Fault intelligent interpretation and stability evaluation method and system of multi-modal geology fusion

Abstract

This invention discloses a method and system for intelligent fault interpretation and stability assessment based on multimodal geological fusion, belonging to the field of risk assessment technology. The method includes the following steps: multimodal geological data preprocessing and fault feature extraction, multimodal fault feature alignment and adaptive fusion, intelligent fault interpretation and classification, and visualization of fault interpretation and risk assessment results, along with platform construction. This invention acquires multimodal geological data from a target area, preprocesses it, extracts features, and spatially aligns it to obtain a multimodal fault feature set; it achieves feature fusion through adaptive weight allocation to form geologically reasonable fused features; it inputs these features into a deep learning model to complete intelligent fault interpretation, conducts risk assessment and classification, improves interpretation accuracy, and avoids misjudgments; finally, it constructs a visualization platform, improving the reliability of intelligent fault interpretation and risk assessment, and solving the problem of low reliability in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of risk assessment technology, and in particular to a method and system for intelligent interpretation and stability assessment of faults based on multimodal geological fusion. Background Technology

[0002] In new work areas, low signal-to-noise ratio data areas, and complex structural areas such as thrust faults and strike-slip faults, prediction accuracy will decrease significantly, and problems such as missed detection of small faults, fault location deviation, and misjudgment of fault strike and dip angle are very likely to occur. These models are essentially purely data-driven, only fitting the distribution and statistical characteristics of training data, without incorporating physical and geological prior constraints such as seismic wave propagation wave equations, sequence stratigraphy, and tectonic deformation mechanics mechanisms. At the same time, mainstream AI (Artificial Intelligence) models are generally based on the basic assumptions of spatiotemporal continuity, information reversibility, and stable distribution in machine learning. However, the actual underground geological structure has strong heterogeneity, abrupt changes, and irreversible evolution, which is fundamentally contradictory to real geological processes such as instantaneous fault activity, stress concentration abrupt changes, sudden fluid decompression, and tectonic uplift and erosion. As a result, the model cannot learn universal features with physical meaning and geological consistency.

[0003] Because the models lack explicit encoding and hard constraints of physical and geological rules, their outputs often contain errors that clearly violate geological common sense, such as unreasonable fault crossings, contradictions between fault displacement and stratigraphic thickness, and fault geometry that deviates from structural patterns. This further exacerbates the problem of failure across work areas. The high-dimensional feature extraction and decision-making processes of deep learning models are highly complex, untraceable, and uninterpretable, exhibiting typical black-box characteristics. They cannot clearly provide the basis for fault identification, the source of key evidence, and the decision-making logic, making it difficult for geologists to verify and correct. The fundamental reason is that the models rely on implicit feature learning and do not explicitly and structurally encode geological rules such as fault geometry, seismic reflection termination relationships, and tectonic coherence. They rely solely on statistical correlation of data to complete the fitting rather than reasoning based on physical mechanisms. In addition, most existing models provide deterministic outputs and lack a sound uncertainty quantification mechanism, making it impossible to quantitatively evaluate the reliability, confidence interval, and error sources of the prediction results. In terms of estimation and visualization, especially in areas with low signal-to-noise ratio and missing data, the model still outputs high-confidence erroneous predictions, leading to serious engineering risks. Existing architectures are mostly single-point optimal outputs, failing to systematically consider noise in the original seismic data, labeling errors, model structural biases, and uncertainties in parameter initialization and optimization. They also fail to integrate probabilistic learning mechanisms such as Bayesian inference and interval prediction, making it impossible to achieve the propagation and quantification of uncertainties across the entire chain from data and model to results. Ultimately, this results in insufficient generalization ability, lack of physical consistency, uninterpretable results, and unassessable reliability in intelligent fault interpretation, making it difficult to apply on a large scale in complex tectonic areas and new work areas. Summary of the Invention

[0004] To address the low reliability of existing fault intelligent interpretation and risk assessment technologies, this invention provides a method and system for fault intelligent interpretation and stability assessment based on multimodal geological fusion. The technical solution is as follows: On the one hand, a method for intelligent fault interpretation and stability assessment based on multimodal geological fusion is provided. This method includes: acquiring multimodal geological data of the target area; extracting fault-related features of corresponding modes based on the preprocessed multimodal geological data; aligning the fault-related features to the same feature space to obtain an aligned multimodal fault feature set; adaptively allocating the weights of each modality feature based on the aligned multimodal fault feature set; fusing the multimodal fault features to remove redundant and false features to obtain a geologically reasonable fused fault feature set; inputting the geologically reasonable fused fault feature set into a pre-trained deep learning model to output intelligent fault interpretation results; performing fault safety level assessment and classification based on the intelligent fault interpretation results to improve the accuracy of fault interpretation and avoid risk misjudgment caused by interpretation bias; and visualizing the intelligent fault interpretation results and fault safety level assessment results by constructing a visualization platform to display the spatial distribution, geometric shape, and fault location of the faults.

[0005] On the other hand, a multimodal geological fusion-based fault intelligent interpretation and stability assessment method system is provided. This system includes: a multimodal geological data preprocessing and fault feature extraction module, a multimodal fault feature alignment and adaptive fusion module, a fault intelligent interpretation and classification module, and a fault interpretation and risk assessment result visualization and platform construction module. Specifically, the multimodal geological data preprocessing and fault feature extraction module acquires multiple modal geological data of the target area, extracts fault-related features of corresponding modalities based on the preprocessed multimodal geological data, and aligns these features to the same feature space to obtain an aligned multimodal fault feature set. The multimodal fault feature alignment and adaptive fusion module is used to perform intelligent interpretation and risk assessment based on the aligned multimodal fault feature set. The system adaptively assigns weights to each modal feature and fuses multimodal fault features to remove redundant and spurious features, resulting in a geologically plausible fused fault feature set. The intelligent fault interpretation and classification module inputs this geologically plausible fused fault feature set into a pre-trained deep learning model, outputting intelligent fault interpretation results. Based on these results, it performs fault safety level assessment and classification to improve the accuracy of fault interpretation and avoid misjudgments caused by interpretation bias. The fault interpretation and risk assessment result visualization and platform construction module visualizes the intelligent fault interpretation results and fault safety level assessment results, building a visualization platform to display the spatial distribution, geometric shape, and fault location of faults.

[0006] Beneficial effects The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. The multimodal geological fusion fault intelligent interpretation and stability assessment method provided by this invention performs unified preprocessing of multi-source heterogeneous geological data such as seismic, well logging, drilling, and geomechanical data, and dynamically adjusts the bandpass filter cutoff frequency based on the fault trace distance. This enhances weak reflection signals in the fault core area, balances fidelity and noise reduction in the transition area, and avoids over-filtering in areas far from the fault, significantly improving the spatial consistency and signal-to-noise ratio of fault-related features of different modes. At the same time, it adaptively and dynamically adjusts the minimum weight value of each modal feature through a time window, ensuring that long-time-window modes are not suppressed and short-time-window modes maintain a reasonable lower limit, achieving a high degree of matching between weight allocation and geological evolution laws. This effectively eliminates redundant features between modes and false features caused by noise, greatly improving the geological rationality, completeness, and reliability of the fused fault features, providing high-quality, strongly constrained feature input for subsequent deep learning models, and reducing fault interpretation bias from the source.

[0007] 2. This invention, by inputting geologically plausible fused fault features into a pre-trained deep learning model, can fully utilize multimodal complementary information to significantly improve the interpretation accuracy and robustness of key parameters such as fault spatial location, geometry, discontinuities, and connectivity. Simultaneously, based on the dynamic mapping and calibration of the risk assessment deviation tolerance threshold using a fluid pressure safety threshold, deviation control is appropriately relaxed in the low fluid pressure range to improve efficiency, while deviation tolerance is strictly tightened in the supercritical pressure range to ensure safety, achieving dynamic adaptation of risk assessment accuracy to engineering scenarios. Through deviation verification, iterative calibration, and linkage with level thresholds, the invention can effectively correct assessment biases caused by interpretation errors, data noise, and geological complexity, significantly reducing the probability of misjudgment and omission of fault risks, making risk level classification more consistent with real geological conditions and engineering safety requirements.

[0008] 3. By integrating multimodal data processing, feature alignment and fusion, intelligent interpretation, risk assessment and control, risk classification and visualization into a complete closed-loop system, each module has a clear division of labor and works in concert to achieve full automation and intelligence from raw data to final decision-making, significantly reducing the subjectivity, multiple interpretations and workload of manual interpretation; by building a visualization platform to intuitively display the spatial distribution, geometric shape, fault location and risk level zoning of faults, supporting 3D interaction, parameter query and result export, the interpretation results and risk conclusions are easier to understand, reuse and implement, effectively improving the work efficiency, decision reliability and field applicability of fault interpretation and risk management, especially suitable for engineering safety assurance and efficient development in high-risk areas such as complex structural areas and high fluid pressure areas. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 Flowchart of the multimodal geological fusion fault intelligent interpretation and stability assessment method provided in the embodiments of this application; Figure 2 A flowchart illustrating the adaptive allocation of weights for each modal feature in the multimodal geological fusion fault intelligent interpretation and stability assessment method provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of the multimodal geological fusion fault intelligent interpretation and stability assessment method system provided in the embodiments of this application. Detailed Implementation

[0011] The following provides explanations for some of the terms used in this application. It should be noted that these explanations are for the convenience of those skilled in the art and do not constitute a limitation on the scope of protection claimed in this application.

[0012] The embodiments of this application involve at least one, including one or more; where "multiple" means two or more. Furthermore, it should be understood that in the description of this specification, terms such as "first," "second," and "third" are used only for descriptive purposes and should not be construed as indicating relative importance or order. For example, "first device" and "second device" do not represent the degree of importance of the two or their order, but are merely for descriptive distinction. In the embodiments of this application, "and / or" merely describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0013] The directional terms mentioned in the embodiments of this application, such as "up", "down", "left", "right", "inner", and "outer", are only for reference to the directions in the accompanying drawings. Therefore, the directional terms used are for better and clearer explanation and understanding of the embodiments of this application, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0014] References to "one embodiment," "in some examples," or "some embodiments" as described in the embodiments of this application mean that one or more embodiments of this specification include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in some examples," "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0016] like Figure 1 The diagram shown is a flowchart of a multimodal geological fusion-based intelligent fault interpretation and stability assessment method provided in this application embodiment. The method includes the following steps: As the first step in the multimodal geological fusion method for intelligent fault interpretation and stability assessment, multiple modal geological data of the target area are acquired. Based on the preprocessed multimodal geological data, fault-related features of the corresponding modes are extracted respectively. The fault-related features are aligned to the same feature space to obtain the aligned multimodal fault feature set.

[0017] It should be understood that the first step is to collect multimodal geological data from the target area, including seismic, logging, drilling, and geomechanical data. Preprocessing operations such as denoising, outlier removal, depth correction, spatial coordinate normalization, and dimensional unification are then performed on these heterogeneous data to eliminate noise interference, missing data, and feature distortion caused by differences between modes, ensuring the stability and reliability of subsequent feature extraction. Based on the preprocessed multimodal geological data, fault-related features such as seismic coherence anomalies, reflection termination, abrupt changes in logging curves, measured values ​​of drilling faults, and stress field anomalies are extracted. This allows each mode to highlight key response information about fault development from its own data dimension, fully leveraging the complementary advantages of multimodal data. Finally, using unified spatial coordinates and stratigraphic time series... Based on the benchmark, the fault-related features extracted from each mode are mapped and aligned to the same feature space. By adaptively adjusting the filtering bandwidth through the distance of the fault trace, weak reflection signals are enhanced in the fault core area, fidelity and noise reduction are balanced in the transition area, and over-filtering is avoided in areas far from the fault. This achieves accurate matching and consistency calibration of multimodal fault features in terms of spatial location, temporal scale, and signal quality, resulting in an aligned multimodal fault feature set with high signal-to-noise ratio, good spatial alignment, and geologically consistent logic. This provides high-quality and reliable basic feature input for subsequent adaptive weight allocation and multimodal feature fusion, reducing feature redundancy, false responses, and interpretation bias caused by modal misalignment from the source, and significantly improving the accuracy and geological rationality of subsequent fault feature fusion and intelligent interpretation.

[0018] It should be further explained that the specific steps for aligning fault-related features to the same feature space are as follows: Obtain the fault trace distribution map of the target area. Based on the fault trace distribution map, obtain the nearest fault trace distance for each sampling point in the seismic data. The nearest fault trace distance is the shortest vertical distance from the sampling point to the nearest fault trace. Based on the comparison between the nearest fault trace distance and the fault trace distance reference interval for each sampling point, the bandpass filter cutoff frequency corresponding to that sampling point is dynamically adjusted as follows: If the nearest fault trace distance is within the fault trace distance reference interval, the sampling point is determined to be located in the fault transition influence area. The nearest fault trace distance is input into the established distance-cutoff frequency mapping relationship, and the low cutoff frequency down-adjustment coefficient and high cutoff frequency up-adjustment coefficient are output. The low cutoff frequency reference value and the low cutoff frequency down-adjustment coefficient are combined to obtain the target low cutoff frequency. The high cutoff frequency reference value and the high cutoff frequency up-adjustment coefficient are combined to obtain the target high cutoff frequency, so as to appropriately improve the filtering bandwidth and balance signal fidelity and noise suppression. The fault trace distance reference interval represents the closed interval formed by the lower limit of the fault trace distance reference and the upper limit of the fault trace distance reference.

[0019] Aligning fault-related features to the same feature space also includes: If the distance to the nearest fault trace is less than the lower limit of the fault trace distance reference, the sampling point is determined to be located in the fault core influence area. The distance to the nearest fault trace is input into the established distance-cutoff frequency mapping relationship, and the low cutoff frequency up adjustment coefficient and the high cutoff frequency down adjustment coefficient are output. The low cutoff frequency reference value and the low cutoff frequency up adjustment coefficient are combined to obtain the target low cutoff frequency. The high cutoff frequency reference value and the high cutoff frequency down adjustment coefficient are combined to obtain the target high cutoff frequency, so as to improve the filtering bandwidth and high frequency response and enhance the recognition accuracy of weak fault reflection signals. If the distance to the nearest fault trace is greater than the upper limit of the fault trace distance reference, the sampling point is determined to be located in the fault transition influence area. The low cutoff frequency reference value and the high cutoff frequency reference value are maintained to avoid over-filtering that causes distortion of the effective layer signal.

[0020] In this embodiment, firstly, a fault trace distribution map of the target area is acquired. Based on this map, the shortest vertical distance from each sampling point in the seismic data to the nearest fault trace is calculated, i.e., the nearest fault trace distance. This achieves accurate quantification of the spatial relationship between the sampling point and the fault, providing a reliable spatial constraint for subsequent adaptive filtering. Next, the nearest fault trace distance of each sampling point is compared with a preset fault trace distance reference interval. Based on the comparison results, the bandpass filter cutoff frequency corresponding to the sampling point is dynamically adjusted. When the nearest fault trace distance falls within the fault trace distance reference interval... When the sampling point is within the closed interval formed by the lower and upper limits, it is determined that the sampling point is located in the fault transition influence area. The distance of the nearest fault trace is input into the pre-constructed distance-cutoff frequency mapping relationship, and the low cutoff frequency down-adjustment coefficient and high cutoff frequency up-adjustment coefficient are output. These are combined with the low cutoff frequency reference value and the high cutoff frequency reference value respectively to obtain the target cutoff frequency. By appropriately widening the filtering bandwidth, the effective reflection signal of the fault transition area is preserved while effectively suppressing environmental noise and random noise, achieving a balance between noise suppression and signal fidelity. When the distance of the nearest fault trace is less than the distance of the fault trace, the sampling point is considered to be located in the fault transition influence area. When the distance to the lower reference limit is reached, the sampling point is determined to be within the fault core influence area. The low cutoff frequency up-adjustment coefficient and high cutoff frequency down-adjustment coefficient are output through the distance-cutoff frequency mapping relationship, and combined to obtain the target cutoff frequency. This improves the filtering bandwidth and high-frequency response capability, enhances the prominence and recognition accuracy of weak fault reflection signals, and solves the problem that traditional fixed filtering struggles to capture weak and concealed reflection features in the fault core area. When the distance to the nearest fault trace is greater than the upper reference limit for fault trace distance, the sampling point is determined to be far from the fault influence area, and the low and high cutoff frequency reference values ​​are directly maintained. The filtering parameters remain unchanged to avoid over-filtering that could weaken or distort the effective layer reflection signal, ensuring the integrity and authenticity of the stratigraphic signal in areas far from the fault. Through the above-mentioned adaptive bandpass filtering adjustment based on spatial location, the filtering parameters of seismic data in different regions are accurately matched with the spatial distribution of the fault, significantly improving the signal-to-noise ratio and spatial consistency of seismic features. This lays a high-quality signal foundation for aligning multimodal fault-related features to the same feature space, effectively improving feature alignment accuracy, spatial matching degree and geological rationality, and reducing feature alignment deviations caused by modal misalignment, signal distortion and noise interference.

[0021] As the second step in the multimodal geological fusion fault intelligent interpretation and stability assessment method, the weights of each modal feature are adaptively allocated based on the aligned multimodal fault feature set, and the multimodal fault features are fused to remove redundant and false features, so as to obtain a geologically reasonable fused fault feature set.

[0022] It should be understood that, based on the multimodal fault feature set that has achieved unified alignment of spatial, temporal, and signal quality, a preset time window matching its geological evolution law is first set for various modal fault features such as seismic, logging, drilling, and geomechanical fault features. Through a pre-constructed time window-minimum weight value mapping relationship, combined with the time window reference interval, reference upper and lower limits, minimum weight reference value, and adjustment coefficient, the minimum weight value that can be assigned to each modal feature is dynamically and adaptively adjusted. When the time window of a modal feature is within the reference interval, the minimum weight reference value is directly adopted to ensure the stability and consistency of weight allocation within the conventional geological time series range, avoiding distortion of fused features due to frequent weight fluctuations. When the time window is greater than the upper limit of the reference, the modal feature is determined to have a longer time coverage and better geological continuity. An upward adjustment coefficient is output through the mapping relationship to improve the minimum weight constraint, avoiding excessive suppression of effective modes in long time windows and fully preserving their geological constraint value over long time scales. When the time window is less than the lower limit of the reference, the modal feature is determined to have stronger temporal focus and local... The features are more prominent. By using mapping relationships to output down-adjustment coefficients to reasonably narrow the minimum weight, it ensures that short-time-window modes participate in the fusion but do not dominate the overall features, achieving temporal adaptability and geological rationality of weight allocation under different time window characteristics. On this basis, global adaptive weight allocation is completed according to the dynamically determined minimum weight and the credibility, correlation, and stability of each modal feature. Multimodal fault features are deeply fused according to weighting rules. Redundant features with high overlap between modes and false features caused by noise and interpretation bias are removed by using correlation thresholds and effective weight thresholds. At the same time, geological prior knowledge is combined to constrain and verify the fusion results, ensuring that the final output fused fault feature set conforms to regional geological laws and tectonic evolution logic in terms of spatial distribution, geometric morphology, temporal evolution, and mechanical response. This significantly improves the reliability, purity, and geological consistency of the fused features, providing a high-quality, strongly constrained, and low-bias input foundation for subsequent deep learning models' intelligent fault interpretation, fundamentally improving fault interpretation accuracy and reducing the probability of misjudgment in subsequent risk assessment.

[0023] It needs to be explained that, such as Figure 2The diagram shows the adaptive weighting flowchart of the fault intelligent interpretation and stability assessment method for multimodal geological fusion provided in this application embodiment. The specific process is as follows: First, obtain the aligned multimodal fault feature set; set a preset time window for each modal fault feature; then determine the time window range of the current modal feature: if the time window is within the baseline interval, directly use the minimum weight baseline value as the minimum weight value of the current mode; if the time window is greater than the baseline upper limit, it is determined to be a long-time-window feature, and the current time window input mapping relationship output adjustment coefficient is used, passing the minimum weight baseline value. The interaction with the adjustment coefficient yields the target minimum weight value to enhance weight constraints and prevent excessive suppression of effective features in long time windows. If the time window is less than the lower limit of the benchmark, it is determined to be a short time window feature. The current time window input mapping relationship is output as the adjustment coefficient. The target minimum weight value is obtained through the interaction between the minimum weight benchmark value and the adjustment coefficient to adapt to the characteristics of short time windows and improve the time adaptability of weight allocation. Finally, the minimum weight value of the current modality is set, realizing the adaptive dynamic matching of the minimum weight values ​​of each modality feature under different time window characteristics, and improving the geological rationality of multimodal feature weight allocation.

[0024] It should be further explained that the specific steps for adaptively assigning weights to each modality feature are as follows: Obtain the aligned multimodal tomographic feature set, set a preset time window for each modal tomographic feature, and dynamically adjust each modal feature based on the time window of each modal tomographic feature to obtain the minimum weight value; A time window-minimum weight value mapping relationship is pre-constructed. This mapping relationship is established using the time window baseline interval, the time window baseline upper limit, the time window baseline lower limit, the minimum weight value baseline value, and the minimum weight value adjustment coefficient. If the time window of each modal tomographic feature is within the time window reference interval, the minimum weight value reference value is used as the current modal feature to obtain the minimum weight value. The minimum weight value is kept without additional adjustment to ensure the stability and consistency of weight allocation within the normal time window range. The time window reference interval represents the closed interval formed by the lower limit of the time window reference and the upper limit of the time window reference.

[0025] Adaptive allocation of weights for each modality feature also includes: If the time window of each modal tomographic feature is greater than the upper limit of the time window baseline, it is determined that the time coverage of the current modal feature is longer. The time window of each modal tomographic feature is input into the pre-constructed time window-minimum weight value mapping relationship, and the minimum weight value adjustment coefficient is output. The minimum weight value baseline value and the minimum weight value adjustment coefficient are interactively processed to obtain the minimum weight value that the target modal feature can obtain, so as to improve the minimum weight constraint of the current modal feature and avoid the long time window effective modal features being over-suppressed. If the time window of each modal fault feature is less than the lower limit of the time window benchmark, it is determined that the current modal feature has stronger temporal focus. The time window of each current modal fault feature is input into the pre-constructed time window-minimum weight value mapping relationship, and the minimum weight value downgrade coefficient is output. The minimum weight value benchmark value and the minimum weight value downgrade coefficient are interactively processed to obtain the minimum weight value that the target modal feature can obtain. This achieves adaptive dynamic matching of the minimum weight value of each modal feature under different time window characteristics, and improves the temporal adaptability and geological rationality of multimodal feature weight allocation.

[0026] In this embodiment, a multimodal fault feature set that has been spatially and temporally aligned is first acquired. This feature set integrates fault-related features from various heterogeneous modes such as seismic, well logging, and drilling, providing a unified benchmark for subsequent adaptive weight allocation. Then, a preset time window is set for each mode's fault features, adapting to its own data characteristics and geological response patterns. The time window setting accurately matches the temporal range of fault feature capture for each mode, providing a core basis for the dynamic adjustment of the minimum weight value, thereby achieving a deep adaptation between weight allocation and geological temporal evolution. Simultaneously, a time window-minimum weight value mapping relationship is pre-constructed. This mapping relationship uses the time window benchmark interval, time window benchmark upper limit, and time window benchmark lower limit as temporal constraint parameters, and the minimum weight value benchmark as the weight benchmark, combined with the minimum weight value adjustment coefficient, to ensure the scientific and geological rationality of the mapping relationship. This provides reliable logical support and quantitative standards for the dynamic calculation of the minimum weight value for each mode, effectively avoiding the subjectivity and blindness of weight adjustment.After completing the above preparations, based on the preset time windows of each modal fault feature, the minimum weight value obtainable by the corresponding modality is dynamically adjusted. The specific adjustment process is as follows: When the time window of each modal fault feature is within the time window benchmark interval (this interval is the closed interval formed by the lower limit and upper limit of the time window benchmark), it is determined that the temporal coverage of the modal feature conforms to the conventional geological scene. At this time, the minimum weight benchmark value is directly used as the minimum weight value obtainable by the current modal feature, and no additional adjustment is made. This operation can effectively ensure the stability and consistency of the weight allocation of each modality within the conventional time window range, avoid the distortion of fused features due to unnecessary weight fluctuations, ensure the stability of multimodal feature fusion, and reduce the weight allocation deviation from the temporal level. Adaptive allocation of the weights of each modal feature also includes adjustments for two special scenarios: If the time window of each modal fault feature is greater than the upper limit of the time window benchmark, it is determined that the temporal coverage of the current modal feature is longer and the temporal continuity of the fault feature is stronger, which can capture fault evolution information at a longer scale. At this time, the time window of the modality is input into the pre-constructed time window - In the minimum weight value mapping relationship, the corresponding minimum weight value adjustment coefficient is accurately output from the mapping relationship. The minimum weight value benchmark value is then interactively processed with this adjustment coefficient (e.g., multiplication) to obtain the target minimum weight value obtainable by the current modal feature. This adjustment can significantly improve the minimum weight constraint of the current modal feature, effectively avoid the excessive suppression of long-term effective modal features by other modalities, and give full play to the core role of long-term modalities in capturing long-term fault evolution features and improving the integrity of fused features. If the time window of each modal fault feature is less than the lower limit of the time window benchmark, it is determined that the current modal feature has stronger temporal focus and can accurately capture the local and short-term subtle features of the fault. At this time, the time window of this modality is input into the pre-constructed time window minus the minimum weight value. In the mapping relationship, the minimum weight value is output as a down-adjustment coefficient. The minimum weight value baseline is then interactively processed with this down-adjustment coefficient to obtain the target minimum weight value obtainable by the current modal feature. This adjustment enables adaptive dynamic matching of the minimum weight values ​​of each modal feature under different time window characteristics. This ensures that the short time window modality can leverage its advantage in capturing local subtle features, while avoiding its excessive weight dominating the fusion process and affecting the global rationality of the fused features. Ultimately, this significantly improves the temporal adaptability and geological rationality of multimodal feature weight allocation, laying a high-quality weight foundation for subsequent multimodal feature fusion and the removal of redundant and false features. It enhances the reliability and accuracy of the fused fault feature set from the source, providing strong support for subsequent intelligent fault interpretation.

[0027] As the third step in the multimodal geological fusion fault intelligent interpretation and stability assessment method, a geologically reasonable fusion fault feature set is input into a pre-trained deep learning model, which outputs fault intelligent interpretation results. Based on the fault intelligent interpretation results, fault safety level assessment, regulation, and level classification are carried out to improve the accuracy of fault interpretation and avoid risk misjudgment caused by interpretation bias.

[0028] It should be understood that, firstly, the fused fault feature set, which possesses complete geological logic and high reliability after adaptive weight fusion and the removal of redundant and false features, is input into a pre-trained deep learning model. This pre-trained deep learning model has been trained with a large number of labeled and standardized multimodal geological samples, fault interpretation samples, and risk case samples, and incorporates explicit encoding of regional geological rules. It can accurately identify the core information contained in the fused feature set, such as fault spatial distribution, geometric morphology, fault location, fault displacement, and connectivity, effectively avoiding the interpretation bias caused by single-modal feature input. At the same time, thanks to the high purity and geological rationality of the fused feature set, the interpretation bias caused by model input noise is significantly reduced, significantly improving the accuracy and robustness of intelligent fault interpretation. Finally, the model stably outputs accurate and comprehensive data. The intelligent fault interpretation results provide high-quality, highly reliable core foundational data for subsequent fault safety level assessment, control, and classification. Subsequently, based on these intelligent fault interpretation results and combined with key parameters such as fluid pressure, stress field, and caprock integrity from multimodal geological data of the target area, the fault safety level assessment and control process is initiated. Through pre-set risk assessment control parameters (risk factor control parameters, assessment model control parameters, and deviation calibration parameters), risk assessment deviations caused by fault interpretation biases, insufficient data accuracy, and complex geological conditions are dynamically corrected. A key focus is on dynamically mapping and calibrating the risk assessment deviation tolerance threshold based on the fluid pressure safety threshold, achieving dynamic adaptation of risk assessment accuracy to engineering scenarios. This effectively solves the problem of "low efficiency in low-risk areas and insufficient accuracy in high-risk areas" caused by traditional fixed deviation control. The pain points were addressed. Finally, based on the completion of risk assessment and control and ensuring the accuracy and reliability of the assessment results, the fault risk level classification work was carried out. Combining the dynamically adjusted risk assessment deviation tolerance threshold with the preset risk level threshold parameters, the calibrated comprehensive fault risk evaluation value was compared and judged, and low, medium, high (and extremely high) risk levels were clearly defined. During the classification process, the adaptability of the deviation tolerance threshold with the current risk level was checked simultaneously. For incompatible thresholds, fine-tuning was carried out and the risk evaluation value was recalibrated to ensure that the level classification results conformed to the actual geological conditions and engineering safety requirements. The whole process realized the coordinated linkage of intelligent fault interpretation, risk assessment, and level classification, which not only greatly improved the accuracy and efficiency of fault interpretation, but also effectively avoided the problems of risk misjudgment and omission caused by interpretation deviation and inaccurate assessment. It provided scientific and accurate decision support for the formulation of subsequent engineering risk prevention and control measures, and effectively ensured the safety and efficiency of engineering operations.

[0029] It should be further explained that the specific steps for assessing and regulating fault safety levels based on the results of intelligent fault interpretation are as follows: Based on the fluid pressure safety threshold corresponding to each mode of fault characteristics, the risk assessment deviation tolerance threshold is dynamically adjusted. The fluid pressure safety threshold is used to determine the safe critical value for whether fluid pressure will induce fault activation. Set a baseline value for the risk assessment deviation tolerance threshold and an adjustment coefficient for the deviation tolerance threshold. The baseline value for the risk assessment deviation tolerance threshold is the initially preset acceptable deviation threshold. Construct a mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold. The fluid pressure threshold ratio represents the ratio of the current fluid pressure safety threshold to the fluid pressure safety baseline threshold.

[0030] Dynamically adjusting the tolerance threshold for risk assessment deviations also includes: If the fluid pressure threshold ratio is within the threshold ratio reference range, the current risk assessment deviation tolerance threshold is set as the risk assessment deviation tolerance benchmark threshold to ensure that the risk assessment deviation is within an acceptable range, while also taking into account assessment efficiency and accuracy, and accurately capturing the assessment deviation caused by potential fluid pressure risks. The threshold ratio reference range represents the closed interval formed by the lower limit of the threshold ratio reference and the upper limit of the threshold ratio reference. If the fluid pressure threshold ratio is less than the lower limit of the threshold ratio reference, the fluid pressure threshold ratio is input into the mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold. The deviation tolerance threshold adjustment coefficient is output, and the deviation tolerance threshold adjustment coefficient is interactively processed with the risk assessment deviation tolerance benchmark threshold to obtain the target risk assessment deviation tolerance threshold. This reduces the frequency of deviation calibration, improves the efficiency of risk assessment, and avoids the waste of computing resources caused by over-calibration.

[0031] Dynamically adjusting the tolerance threshold for risk assessment deviations also includes: If the fluid pressure threshold ratio is less than the upper limit of the threshold ratio reference, the fluid pressure threshold ratio is input into the mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold, and the deviation tolerance threshold reduction coefficient is output. The deviation tolerance threshold reduction coefficient is then interactively processed with the risk assessment deviation tolerance benchmark threshold to obtain the target risk assessment deviation tolerance threshold, thereby avoiding risk misjudgment caused by excessive fluid pressure and loss of control over deviation.

[0032] In this embodiment, firstly, the fluid pressure safety threshold corresponding to each modal fault feature extracted in the previous stage is retrieved. This fluid pressure safety threshold is a core critical value pre-set based on the caprock bearing capacity, fault sealing performance, and engineering safety requirements of the target area. Its core function is to accurately determine whether the current fluid pressure will induce fault activation, providing a scientific and geologically realistic core constraint for the dynamic adjustment of the subsequent risk assessment deviation tolerance threshold. This ensures the correlation between deviation adjustment and fault activation risk from the source, avoiding assessment deviations caused by blind adjustments. Subsequently, a baseline value and adjustment coefficient for the risk assessment deviation tolerance threshold are preset, wherein the baseline value is initially set. The acceptable critical value for deviation, applicable to conventional geological scenarios, provides a unified benchmark for adjusting the deviation tolerance threshold. The deviation tolerance threshold adjustment coefficient is used to quantify the adjustment range of the deviation tolerance threshold under different fluid pressure scenarios. At the same time, a mapping relationship is constructed between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold. Specifically, the fluid pressure threshold ratio is the ratio of the current fluid pressure safety threshold to the fluid pressure safety benchmark threshold. Through this ratio, the deviation degree between the current fluid pressure safety threshold and the conventional benchmark value can be accurately quantified, realizing the accurate mapping between the fluid pressure scenario and the deviation tolerance threshold. This effectively avoids the problem of fluid pressure and deviation adjustment being out of sync, ensuring the rigor and operability of the adjustment logic.After completing the above preparations, based on the established mapping relationship and fluid pressure threshold ratio, the risk assessment deviation tolerance threshold is dynamically adjusted according to different scenarios. The specific adjustment process and technical effects are as follows: If the fluid pressure threshold ratio is within the threshold ratio reference range (this range is a closed interval formed by the lower limit and upper limit of the threshold ratio reference), then the current fluid pressure is determined to be within the normal safe range, and the influence of fluid pressure on fault activation is under control. At this time, the current risk assessment deviation tolerance threshold is directly set as the risk assessment deviation tolerance benchmark value without additional adjustment. This operation can ensure that the risk assessment deviation is always within the preset acceptable range, avoiding misjudgment of risk due to excessive deviation, while also taking into account the risks. The system achieves a balance between efficiency and accuracy in risk assessment without requiring redundant calibration. It precisely captures assessment biases caused by potential fluid pressure risks, resulting in a dynamic adjustment process that adapts to low-fluid-pressure scenarios. If the fluid pressure threshold ratio is less than the lower reference limit, the current fluid pressure is considered to be in a low-risk range, with an extremely low probability of fluid pressure-induced fault activation and minimal impact on risk assessment bias. In this case, the fluid pressure threshold ratio is input into a pre-built mapping relationship between the fluid pressure threshold ratio and the risk assessment bias tolerance threshold. The mapping relationship accurately outputs the corresponding bias tolerance threshold adjustment coefficient, which is then compared with the risk assessment bias tolerance benchmark. The values ​​are processed through interactive operations (such as multiplication) to obtain the target risk assessment deviation tolerance threshold. This adjustment can appropriately relax the deviation control standard, effectively reduce the frequency of deviation calibration, reduce unnecessary calculation operations, significantly improve the efficiency of risk assessment, and avoid the waste of computing resources caused by over-calibration, thus balancing assessment efficiency and resource rationality. The dynamic adjustment process also includes adaptation adjustments for high fluid pressure scenarios: if the fluid pressure threshold ratio is greater than the upper limit of the threshold ratio reference, it is determined that the current fluid pressure is in an abnormally safe range, and the probability of fluid pressure-induced fault activation is significantly increased, which can easily lead to uncontrolled risk assessment deviation. At this time, the fluid pressure threshold ratio is input into a pre-constructed mapping relationship, and the mapping relationship... The system outputs a corresponding deviation tolerance threshold reduction coefficient. This reduction coefficient is then interactively processed with the risk assessment deviation tolerance benchmark value to obtain the target risk assessment deviation tolerance threshold. By appropriately tightening the deviation tolerance standard, the control accuracy of risk assessment deviation can be significantly improved, effectively avoiding the problems of misjudgment and omission of fault risks caused by excessive fluid pressure and uncontrolled deviation. This ensures the accuracy and reliability of risk assessment results under high fluid pressure scenarios, providing a precise basis for deviation control in subsequent fault risk level classification. Ultimately, it achieves adaptive adaptation between the risk assessment deviation tolerance threshold and the fluid pressure scenario, significantly improving the accuracy, pertinence, and reliability of fault safety level assessment, and providing scientific support for engineering safety prevention and control.

[0033] It should be further explained that the specific ratios for the grade divisions are: The dynamically adjusted risk assessment deviation tolerance threshold is compared with the current comprehensive fault risk assessment value, specifically: If the current fault risk comprehensive assessment value is less than the lower limit of the risk assessment deviation tolerance threshold, the current fault is determined to be of low risk level; If the current comprehensive risk assessment value of the fault is within the risk assessment deviation tolerance threshold range, the current fault is determined to be of medium risk level. The risk assessment deviation tolerance threshold range represents the closed interval formed by the lower limit of the risk assessment deviation tolerance threshold and the upper limit of the risk assessment deviation tolerance threshold. If the current comprehensive risk assessment value of the fault exceeds the upper limit of the risk assessment deviation tolerance threshold, the current fault is determined to be of a high-risk level.

[0034] In this embodiment, firstly, the risk assessment deviation tolerance threshold, which was dynamically adjusted based on the fluid pressure safety threshold in the previous stage, is retrieved. This threshold includes a lower limit, an upper limit, and a closed interval. This interval, along with the lower and upper limits, serves as the core criterion for classifying fault risk levels. It has been adaptively adapted to the fluid pressure scenario, effectively avoiding level deviations caused by fixed threshold classifications, and providing accurate and geologically relevant quantitative basis for subsequent level determination. Simultaneously, the comprehensive fault risk assessment value, obtained after risk assessment adjustment and deviation calibration, is retrieved. This assessment value eliminates the impact of interpretation bias and data errors, accurately reflecting the current activation risk and safety status of the fault, ensuring the accuracy and reliability of the basic data for level determination, and reducing the probability of misjudgment in level classification from the source. Subsequently, the dynamically adjusted risk assessment deviation tolerance thresholds (including lower limit, upper limit, and range) are compared one by one with the current comprehensive fault risk assessment value. Accurate fault risk level determination is completed according to preset judgment rules for different scenarios. The specific judgment process and corresponding technical effects are as follows: If the current comprehensive fault risk assessment value is less than the lower limit threshold of the risk assessment deviation tolerance, the current fault is determined to be at a low risk level. This judgment logic can accurately identify scenarios where fault risk is at a low level, clearly indicating that the activation probability of such faults is extremely low and they pose no significant safety threat to engineering operations. This avoids over-control and resource waste caused by misjudging low-risk faults as medium- or high-risk, and accurately defines the range of low-risk faults, providing a basis for subsequently developing low-cost and efficient routine monitoring measures. If the current comprehensive fault risk assessment value is within the closed range of the risk assessment deviation tolerance threshold (i.e., greater than or equal to the lower limit threshold and less than or equal to the upper limit threshold), the risk level is determined accordingly. If the fault's overall risk assessment value exceeds the risk assessment deviation tolerance threshold, the fault is classified as high-risk. This logic accurately identifies scenarios where the fault has a medium risk of activation, clearly indicating that such faults have a moderate activation probability and pose a potential threat to engineering operations. This avoids both misjudging medium-risk faults as low-risk and wasting prevention and control resources due to misjudging them as high-risk, achieving accurate identification and coordinated management of potential risks. If the current fault's overall risk assessment value exceeds the risk assessment deviation tolerance threshold, the fault is classified as high-risk. This logic accurately identifies scenarios where the fault risk is at a high level, clearly indicating that such faults have an extremely high activation probability and pose a serious safety threat to engineering operations. It can promptly identify high-risk hazards, avoiding engineering safety accidents caused by missed or misjudged high-risk faults. This provides a clear basis for subsequently initiating high-frequency monitoring, emergency prevention and control measures, and suspending related operations, effectively ensuring the safety of engineering operations.The entire comparison and judgment process achieved precise linkage between the comprehensive risk assessment value and the deviation tolerance threshold after dynamic adaptation. The judgment rules were clear and the quantitative standards were well-defined. It not only fully leveraged the adaptability advantage of the dynamically adjusted threshold, but also relied on the accurate assessment value to achieve precise differentiation of different risk level faults. This effectively avoided the problems of risk misjudgment, omission, and inappropriate prevention and control caused by the deviation in level classification, ensuring the scientific, targeted, and reliable nature of fault risk level classification. It provided precise level support for the subsequent formulation of differentiated prevention and control measures, realized the hierarchical management and precise prevention and control of fault risks, and further improved the efficiency and safety of fault risk management.

[0035] As the fourth step in the multimodal geological fusion fault intelligent interpretation and stability assessment method, the fault intelligent interpretation results and fault safety level assessment results are visualized and displayed, and a visualization platform is constructed to show the spatial distribution, geometric shape and fault location of the fault.

[0036] It should be understood that, firstly, the system systematically reviews the previously obtained intelligent fault interpretation results and fault safety level assessment results, clarifying the core content of the visualization. The intelligent fault interpretation results cover key parameters such as the spatial distribution range of faults, three-dimensional geometric morphology (strike, dip, displacement, etc.), precise fault location, and fault connectivity. The fault safety level assessment results cover the comprehensive risk evaluation value of each fault, the dynamically adjusted deviation tolerance threshold, and the final risk level (low, medium, high). At the same time, the system associates the spatial coordinates of the target area, stratigraphic depth, and time series information corresponding to each result. Through the classification, organization, and integration of results, the system ensures the completeness, relevance, and standardization of the visualization content, avoiding misunderstandings caused by chaotic display content and missing parameters. This lays a clear data foundation for subsequent visualization and enables the linkage between interpretation results and assessment results, facilitating simultaneous viewing and comparative analysis by staff. Subsequently, 3D visualization rendering technology, zoning coloring technology, and precise labeling technology were used to present the core content in an intuitive and visual way. Specifically, the spatial distribution of faults was shown against the backdrop of the target area's topography, with clear lines of different colors marking the fault's direction and extent. Spatial coordinate calibration ensured the accuracy of the distribution display, allowing staff to quickly grasp the overall distribution pattern of faults in the target area, addressing the pain point of traditional 2D displays' inability to intuitively present the spatial relationships of faults. The geometric shape of the faults was presented in the form of a 3D model, supporting interactive operations such as rotation, scaling, and sectioning, clearly restoring the three-dimensional structural features of the faults. This facilitated detailed observation of the internal morphology and spatial distribution patterns of the faults, overcoming the limitations of traditional planar displays. The fault locations were marked with special highlighted indicators. Clicking on these indicators allowed users to view detailed parameters such as fault depth, lithology, and corresponding risk level in a pop-up window, achieving accurate presentation and rapid query of fault information, effectively improving the efficiency of fault identification and parameter retrieval. Finally, all visualization content is integrated to construct a standardized and intelligent fault interpretation and risk assessment visualization platform. This platform integrates core functions such as result display, parameter query, interactive operation, and result export. It can not only centrally display all core content such as fault spatial distribution, geometric shape, fault location, and risk level, but also realize the linkage between intelligent fault interpretation results and risk assessment results. Staff can quickly retrieve detailed parameters and risk information of faults at any location through the platform. At the same time, it supports exporting visualization results and detailed data into standardized formats for use in engineering report preparation, decision-making, and subsequent data archiving.The construction and visualization of this platform completely solves the problem of traditional fault interpretation and risk assessment results being presented in the form of data and reports, which are abstract, difficult to understand, and hard to interpret quickly. It transforms abstract numerical results and interpretive parameters into intuitive and easy-to-understand 3D graphics and icons, significantly reducing the difficulty for staff to interpret the results and improving work efficiency. At the same time, the platform's interactivity and convenience make the querying and analysis of fault-related information more efficient, providing engineering decision-makers with intuitive and accurate decision-making basis and avoiding decision-making biases caused by untimely or inaccurate interpretation of results. Furthermore, the visualization platform achieves standardized display and centralized management of fault interpretation and risk assessment results, ensuring the traceability and reusability of the results, further improving the convenience, efficiency, and scientific nature of fault risk management, providing strong visual support for the formulation and implementation of subsequent engineering risk prevention and control measures, and effectively ensuring the safety and efficiency of engineering operations.

[0037] like Figure 3 The diagram shown is a structural schematic of the multimodal geological fusion-based fault intelligent interpretation and stability assessment method system provided in this application embodiment. It includes: a multimodal geological data preprocessing and fault feature extraction module, a multimodal fault feature alignment and adaptive fusion module, a fault intelligent interpretation and classification module, and a fault interpretation and risk assessment result visualization and platform construction module. The multimodal geological data preprocessing and fault feature extraction module is used to acquire multiple modal geological data of the target area, extract fault-related features of the corresponding modalities based on the preprocessed multimodal geological data, and align the fault-related features to the same feature space to obtain an aligned multimodal fault feature set. The multimodal fault feature alignment and adaptive fusion module is used to... The feature set adaptively assigns weights to each modal feature and fuses multimodal fault features to remove redundant and spurious features, resulting in a geologically plausible fused fault feature set. The fault intelligent interpretation and classification module inputs the geologically plausible fused fault feature set into a pre-trained deep learning model and outputs intelligent fault interpretation results. Based on these results, fault safety level assessment and classification are performed to improve the accuracy of fault interpretation and avoid misjudgments caused by interpretation bias. The fault interpretation and risk assessment result visualization and platform construction module visualizes the intelligent fault interpretation results and fault safety level assessment results, building a visualization platform to display the spatial distribution, geometric shape, and fault location of faults.

[0038] In this embodiment, the multimodal geological data preprocessing and fault feature extraction module is first activated. The primary function of this module is to acquire multimodal geological data covering the target area, including seismic, logging, drilling, and geomechanical data. By performing preprocessing operations such as denoising, outlier removal, depth correction, spatial coordinate normalization, and dimensional unification on various heterogeneous data, noise interference, data gaps, and feature distortion caused by dimensional differences between modes are effectively eliminated, ensuring the stability and reliability of subsequent feature extraction. Subsequently, based on the preprocessed geological data of each mode, fault-related features for the corresponding modes are extracted. Specifically, features such as reflection termination and coherence anomalies are extracted for the seismic mode, and features such as reflection termination and coherence anomalies are extracted for the logging mode. This module extracts features such as curve abrupt changes, features such as measured fault data from drilling modes, and features such as stress field anomalies from geomechanical modes. It fully leverages the complementary advantages of multimodal data to ensure that the extracted features comprehensively and accurately reflect the different developmental characteristics of faults. Finally, this module aligns the fault-related features extracted from each mode to the same feature space. By adaptively adjusting the filtering bandwidth based on fault trace distance, it achieves precise matching of multimodal features in spatial location and temporal scale, resulting in an aligned multimodal fault feature set with high signal-to-noise ratio, good spatial alignment, and geologically consistent logic. This reduces subsequent interpretation biases caused by feature redundancy and misalignment from the source, providing a high-quality feature input foundation for the next module. Next, the multimodal fault feature alignment and adaptive fusion module is launched. This module takes the aligned multimodal fault feature set as input, first sets a preset time window for each modality of fault features, and dynamically and adaptively adjusts the minimum weight value that can be obtained for each modality based on the mapping relationship between the time window and the minimum weight value. This ensures that the effective modes in the long time window are not overly suppressed and the modes in the short time window are kept reasonably constrained, achieving a high degree of adaptation between the weight allocation and the geological evolution law. Then, the multimodal fault features are deeply fused based on the dynamically allocated weights. At the same time, redundant features with high overlap between modes and false features caused by noise are removed by using correlation thresholds and effective weight thresholds. The rationality of the fusion result is verified by combining geological prior knowledge. Finally, a geologically reasonable fused fault feature set is obtained, which greatly improves the purity and reliability of the features, further optimizes the input quality of subsequent intelligent interpretation, and reduces the ambiguity of interpretation.The next step is to launch the intelligent fault interpretation and classification module. This module inputs a geologically plausible fusion fault feature set into a pre-trained deep learning model. This model incorporates regional geological rule coding and can fully utilize multimodal complementary information to significantly improve the interpretation accuracy and robustness of key parameters such as fault spatial distribution, geometric morphology, and fault location, and stably output accurate and comprehensive intelligent fault interpretation results. Subsequently, based on this interpretation result, the risk assessment deviation tolerance threshold is dynamically adjusted in conjunction with the fluid pressure safety threshold to achieve dynamic adaptation of risk assessment accuracy to engineering scenarios. The assessment deviation is corrected through deviation verification and iterative calibration. Finally, the dynamically adjusted deviation tolerance threshold is compared with the current comprehensive fault risk assessment value to accurately classify low, medium, and high risk levels, effectively avoiding risk misjudgment and omission caused by interpretation deviation and inaccurate assessment, making the risk level classification more consistent with real geological conditions and engineering safety requirements. Finally, the module for visualizing fault interpretation and risk assessment results and building a platform was launched. This module first systematically organizes the intelligent fault interpretation results and risk assessment results, associating them with corresponding spatial coordinates, stratigraphic depth, and other information to ensure the completeness and relevance of the displayed content. Subsequently, it uses 3D visualization rendering, partitioning coloring, and highlighting techniques to intuitively display the spatial distribution, 3D geometric shape, and fault location of the faults. The spatial distribution of the faults is marked with topography as the background, the geometric shape is presented with an interactive 3D model, and the fault location is marked with special markers and supports detailed parameter queries, completely solving the pain points of traditional data reports being abstract, difficult to understand, and hard to interpret. Finally, all visualization content is integrated to build a visualization platform that integrates result display, parameter query, and result export functions, realizing full-process visualization from raw data to decision results. This significantly reduces the difficulty of interpretation for staff, improves work efficiency, and provides intuitive and accurate decision-making basis for engineering decision-makers, ensuring that interpretation results and risk conclusions can be efficiently implemented, further improving the practicality and reliability of fault risk management and decision-making. The four modules work together to form a complete closed loop from raw multimodal data preprocessing, feature processing, intelligent interpretation, risk assessment to visualization implementation, achieving full-process automation and intelligence, significantly reducing human subjectivity and workload, improving the efficiency, accuracy and practicality of fault interpretation and risk management, and providing strong support for the safety assurance of complex engineering projects.

[0039] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)). Where there is no conflict, the solutions in the above embodiments can be combined.

[0040] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0041] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0042] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0043] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0044] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope and intent of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A multimodal geological fusion-based intelligent interpretation and stability assessment method for faults, characterized in that, Includes the following steps: Multiple modal geological data of the target area are acquired. Based on the preprocessed multiple modal geological data, fault-related features of the corresponding modes are extracted respectively. The fault-related features are aligned to the same feature space to obtain the aligned multimodal fault feature set. Based on the aligned multimodal fault feature set, the weights of each modal feature are adaptively assigned, and the multimodal fault features are fused to remove redundant and false features, resulting in a geologically reasonable fused fault feature set. A geologically plausible set of fused fault features is input into a pre-trained deep learning model, which outputs intelligent fault interpretation results. Based on these results, fault safety level assessment, regulation, and classification are performed to improve the accuracy of fault interpretation and avoid misjudgments caused by interpretation bias. The results of intelligent fault interpretation and fault safety level assessment are visualized and displayed, and a visualization platform is built to show the spatial distribution, geometric shape, and fault location of faults.

2. The multimodal geological fusion-based intelligent interpretation and stability assessment method for faults as described in claim 1, characterized in that: The specific steps for aligning fault-related features to the same feature space are as follows: Obtain a fault trace distribution map of the target area. Based on the fault trace distribution map, obtain the nearest fault trace distance for each sampling point in the seismic data. The nearest fault trace distance is the shortest vertical distance from the sampling point to the nearest fault trace. Based on the comparison between the nearest fault trace distance and the fault trace distance reference interval for each sampling point, the bandpass filter cutoff frequency corresponding to that sampling point is dynamically adjusted as follows: If the nearest fault trace distance is within the fault trace distance reference interval, the sampling point is determined to be located in the fault transition influence region. The nearest fault trace distance is input into the established distance-cutoff frequency mapping relationship, and the low cutoff frequency down-adjustment coefficient and the high cutoff frequency up-adjustment coefficient are output. The low cutoff frequency reference value and the low cutoff frequency down-adjustment coefficient are combined to obtain the target low cutoff frequency. The high cutoff frequency reference value and the high cutoff frequency up-adjustment coefficient are combined to obtain the target high cutoff frequency, so as to appropriately improve the filtering bandwidth and balance signal fidelity and noise suppression. The fault trace distance reference interval represents the closed interval formed by the lower limit of the fault trace distance reference and the upper limit of the fault trace distance reference.

3. The fault intelligent interpretation and stability assessment method based on multimodal geological fusion as described in claim 2, characterized in that: The step of aligning fault-related features to the same feature space also includes: If the distance to the nearest fault trace is less than the lower limit of the fault trace distance reference, the sampling point is determined to be located in the fault core influence area. The distance to the nearest fault trace is input into the established distance-cutoff frequency mapping relationship, and the low cutoff frequency up adjustment coefficient and the high cutoff frequency down adjustment coefficient are output. The low cutoff frequency reference value and the low cutoff frequency up adjustment coefficient are combined to obtain the target low cutoff frequency. The high cutoff frequency reference value and the high cutoff frequency down adjustment coefficient are combined to obtain the target high cutoff frequency, so as to improve the filtering bandwidth and high frequency response and enhance the recognition accuracy of weak fault reflection signals. If the distance to the nearest fault trace is greater than the upper limit of the fault trace distance reference, the sampling point is determined to be located in the fault transition influence area. The low cutoff frequency reference value and the high cutoff frequency reference value are maintained to avoid over-filtering that causes distortion of the effective layer signal.

4. The multimodal geological fusion-based intelligent interpretation and stability assessment method for faults as described in claim 1, characterized in that: The specific steps for adaptively allocating the weights of each modality feature are as follows: Obtain the aligned multimodal tomographic feature set, set a preset time window for each modal tomographic feature, and dynamically adjust each modal feature based on the time window of each modal tomographic feature to obtain the minimum weight value; A time window-minimum weight value mapping relationship is pre-constructed, which is established through the time window benchmark interval, the time window benchmark upper limit, the time window benchmark lower limit, the minimum weight value benchmark value, and the minimum weight value adjustment coefficient. If the time window of each modal tomographic feature is within the time window reference interval, the minimum weight value reference value is used as the current modal feature to obtain the minimum weight value. The minimum weight value is kept without additional adjustment to ensure the stability and consistency of weight allocation within the normal time window range. The time window reference interval represents the closed interval formed by the lower limit of the time window reference and the upper limit of the time window reference.

5. The fault intelligent interpretation and stability assessment method based on multimodal geological fusion as described in claim 4, characterized in that: The adaptive allocation of weights for each modality feature further includes: If the time window of each modal tomographic feature is greater than the upper limit of the time window baseline, it is determined that the time coverage of the current modal feature is longer. The time window of each modal tomographic feature is input into the pre-constructed time window-minimum weight value mapping relationship, and the minimum weight value adjustment coefficient is output. The minimum weight value baseline value and the minimum weight value adjustment coefficient are interactively processed to obtain the minimum weight value that the target modal feature can obtain, so as to improve the minimum weight constraint of the current modal feature and avoid the long time window effective modal features being over-suppressed. If the time window of each modal fault feature is less than the lower limit of the time window benchmark, it is determined that the current modal feature has stronger temporal focus. The time window of each current modal fault feature is input into the pre-constructed time window-minimum weight value mapping relationship, and the minimum weight value downgrade coefficient is output. The minimum weight value benchmark value and the minimum weight value downgrade coefficient are interactively processed to obtain the minimum weight value that the target modal feature can obtain. This achieves adaptive dynamic matching of the minimum weight value of each modal feature under different time window characteristics, and improves the temporal adaptability and geological rationality of multimodal feature weight allocation.

6. The multimodal geological fusion-based intelligent interpretation and stability assessment method for faults as described in claim 1, characterized in that: The specific steps for fault safety level assessment and control based on fault intelligent interpretation results are as follows: Based on the fluid pressure safety threshold corresponding to each mode of fault characteristics, the risk assessment deviation tolerance threshold is dynamically adjusted. The fluid pressure safety threshold is used to determine the safety threshold value for whether fluid pressure will induce fault activation. Set a baseline value for the risk assessment deviation tolerance threshold and an adjustment coefficient for the deviation tolerance threshold. The baseline value for the risk assessment deviation tolerance threshold is an initially preset acceptable deviation threshold. Construct a mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold. The fluid pressure threshold ratio represents the ratio of the current fluid pressure safety threshold to the fluid pressure safety baseline threshold.

7. The multimodal geological fusion-based intelligent interpretation and stability assessment method for faults as described in claim 6, characterized in that: The dynamic adjustment of the risk assessment deviation tolerance threshold also includes: If the fluid pressure threshold ratio is within the threshold ratio reference range, the current risk assessment deviation tolerance threshold is set as the risk assessment deviation tolerance benchmark threshold to ensure that the risk assessment deviation is within an acceptable range, while also taking into account assessment efficiency and accuracy, and accurately capturing the assessment deviation caused by potential fluid pressure risks. The threshold ratio reference range represents the closed interval formed by the lower limit of the threshold ratio reference and the upper limit of the threshold ratio reference. If the fluid pressure threshold ratio is less than the lower limit of the threshold ratio reference, the fluid pressure threshold ratio is input into the mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold. The deviation tolerance threshold adjustment coefficient is output, and the deviation tolerance threshold adjustment coefficient is interactively processed with the risk assessment deviation tolerance benchmark threshold to obtain the target risk assessment deviation tolerance threshold. This reduces the frequency of deviation calibration, improves the efficiency of risk assessment, and avoids the waste of computing resources caused by over-calibration.

8. The fault intelligent interpretation and stability assessment method based on multimodal geological fusion as described in claim 7, characterized in that: The dynamic adjustment of the risk assessment deviation tolerance threshold also includes: If the fluid pressure threshold ratio is less than the upper limit of the threshold ratio reference, the fluid pressure threshold ratio is input into the mapping relationship between the fluid pressure threshold ratio and the risk assessment deviation tolerance threshold, and the deviation tolerance threshold reduction coefficient is output. The deviation tolerance threshold reduction coefficient is then interactively processed with the risk assessment deviation tolerance benchmark threshold to obtain the target risk assessment deviation tolerance threshold, thereby avoiding risk misjudgment caused by excessive fluid pressure and loss of control over deviation.

9. The multimodal geological fusion-based intelligent interpretation and stability assessment method for faults as described in claim 1, characterized in that: The specific ratio for the grade division is: The dynamically adjusted risk assessment deviation tolerance threshold is compared with the current comprehensive fault risk assessment value, specifically: If the current fault risk comprehensive assessment value is less than the lower limit of the risk assessment deviation tolerance threshold, the current fault is determined to be of low risk level; If the current fault risk comprehensive assessment value is within the risk assessment deviation tolerance threshold range, then the current fault is determined to be of medium risk level. The risk assessment deviation tolerance threshold range represents the closed interval formed by the lower limit threshold of risk assessment deviation tolerance and the upper limit threshold of risk assessment deviation tolerance. If the current comprehensive risk assessment value of the fault exceeds the upper limit of the risk assessment deviation tolerance threshold, the current fault is determined to be of a high-risk level.

10. A system applying the multimodal geological fusion fault intelligent interpretation and stability assessment method as described in any one of claims 1-9, characterized in that, include: The system includes modules for multimodal geological data preprocessing and fault feature extraction, multimodal fault feature alignment and adaptive fusion, intelligent fault interpretation and classification, and visualization and platform construction of fault interpretation and risk assessment results. The multimodal geological data preprocessing and fault feature extraction module is used to acquire multiple modal geological data of the target area, extract fault-related features of the corresponding modal based on the preprocessed multiple modal geological data, and align the fault-related features to the same feature space to obtain an aligned multimodal fault feature set. The multimodal fault feature alignment and adaptive fusion module is used to adaptively allocate the weights of each modal feature based on the aligned multimodal fault feature set, and fuse the multimodal fault features to remove redundant and false features, thereby obtaining a geologically reasonable fused fault feature set. The fault intelligent interpretation and classification module is used to input a geologically reasonable fused fault feature set into a pre-trained deep learning model, output fault intelligent interpretation results, and perform fault safety level assessment, control and classification based on the fault intelligent interpretation results, so as to improve the accuracy of fault interpretation and avoid risk misjudgment caused by interpretation deviation. The fault interpretation and risk assessment result visualization and platform construction module is used to visualize the intelligent fault interpretation results and fault safety level assessment results, and to build a visualization platform to display the spatial distribution, geometric shape, and fault location of the fault.

Images