A method for advanced geological prediction applied to a tunnel face

Abstract

The application provides an advanced geological prediction method applied to a tunnel face, and relates to the field of data processing. In the method, after the tunnel face is supported and closed, a working condition context is established based on a fixed set of excitation points and a set of receiving points, a reference gather is acquired and shaped to represent the basic wave field response of surrounding rock, and a mutually exclusive state mechanism model is constructed in parallel under a unified time window system and discriminant features are extracted. By introducing azimuth information through small-angle migration of the excitation points, the state mechanism model is eliminated step by step, and if necessary, a unique winning model is determined in combination with a counterexample experiment. The winning model is further mapped into a state risk map containing an axial range, a circumferential offset and an evolution trend, and is associated with a preset treatment strategy library to output a targeted advanced geological prediction treatment strategy. The technical solution provided by the application facilitates improving the accuracy of the advanced geological prediction applied to the tunnel face.

Description

Technical Field

[0001] This application relates to the technical field of data processing, specifically to an advanced geological prediction method applied to tunnel faces. Background Technology

[0002] As tunnel engineering continues to advance towards deeper burial, greater burial depth, complex structures, and high ground stress areas, the engineering behavior of surrounding rock has gradually shifted from being controlled by geometric discontinuities to being controlled by stress evolution and damage accumulation. The main sources of construction safety risks are no longer limited to geological bodies with clear spatial boundaries such as faults, karst caves, or strongly fractured zones, but are more reflected in stress concentration, plastic zone expansion, gradual failure, and the mechanical processes such as large deformation and sudden instability induced by these factors.

[0003] In existing tunnel face geological prediction technologies, prediction systems represented by seismic wave methods such as TSP mainly focus on geological bodies with clear abrupt changes in physical properties, such as faults, fracture zones, karst caves, and water-rich structures. Their processing flow and interpretation logic generally use the imaging quality of reflection events, the continuity of reflection interfaces, and travel-time focusing effects as core criteria. Under this technical system, the surrounding rock is implicitly assumed to be composed of several distinguishable physical property zones, and the seismic response is preferentially interpreted as the result of interface reflection or refraction. However, in actual tunnel construction environments, non-interface risk bodies such as stress concentration zones, the leading edge of plastic zones, and progressive failure zones do not correspond to clear geometric interfaces. Instead, they exist in a form where the mechanical state evolves continuously with space and time. Their seismic responses often exhibit non-interface characteristics such as diffused reflected energy, reduced phase stability, abnormal changes in propagation velocity, and rapid energy attenuation. Such responses are difficult to form stable and focusable reflection evidence in conventional migration imaging and reflection enhancement processing. Moreover, their manifestations are highly similar in appearance to construction disturbances, support coupling, or environmental noise. Therefore, they are usually regarded as uncertain information and are weakened or even ignored directly. As a result, detectable risk signals cannot be transformed into effective early warnings, thus making the accuracy of early geological forecasts low.

[0004] Therefore, there is an urgent need for an advanced geological prediction method applicable to tunnel faces. Summary of the Invention

[0005] This application provides a method for advanced geological prediction applied to tunnel faces, which facilitates the improvement of the accuracy of advanced geological prediction applied to tunnel faces.

[0006] The first aspect of this application provides an advanced geological prediction method for tunnel faces, comprising: after the tunnel face has advanced to a preset mileage and the current shift's support has been closed, determining a set of excitation points and a set of receiver points, and establishing a working condition context bound to the set of excitation points and the set of receiver points; under the constraints of the working condition context, obtaining a reference gather through the set of excitation points and the set of receiver points, and performing source consistency shaping processing on the reference gather to form a reference evidence package, which is used to characterize the basic response characteristics of the surrounding rock to the wavefield under the current working condition; constructing a set of mutually exclusive state mechanism models in parallel based on the reference evidence package, and constructing a discriminative feature combination for the set of state mechanism models under a time window system, the time window system including a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the tunnel face, and a far-field time window for constraining the effect of the medium in front of the tunnel face; and obtaining the azimuth by performing a small-angle circumferential migration on the set of excitation points while keeping the set of receiver points unchanged. A subset of azimuth evidence is obtained, and azimuth evidence packages are acquired through the azimuth subset. Simultaneously, based on the azimuth response patterns corresponding to the discriminant feature combinations extracted from the azimuth evidence packages within the same time window system, a first elimination process is performed on the set of state mechanism models to obtain candidate state mechanism models. If the number of candidate state mechanism models is greater than a preset number, a disproven experiment is performed without changing the set of excitation points and receiver points to perform a second elimination process on the candidate state mechanism models, determining the winning state mechanism model. The winning state mechanism model is mapped to a state risk map, with axial range, circumferential offset, and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combinations within the far-field time window, the circumferential offset is determined by the differences between different azimuth subsets of the discriminant feature combinations, and the evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles. The state risk map is then associated with a preset disposal strategy library to output corresponding advanced geological prediction disposal strategies.

[0007] The second aspect of this application provides an advanced geological prediction device for tunnel faces. The device includes an acquisition module and a processing module. The acquisition module is used to determine the set of excitation points and the set of receiver points after the tunnel face has advanced to a preset mileage and the support closure for the shift has been completed, and to establish a working context bound to the set of excitation points and the set of receiver points. The processing module is used to acquire a reference gather through the set of excitation points and the set of receiver points under the constraints of the working context, and to perform source consistency shaping on the reference gather to form a reference evidence package. The reference evidence package is used to characterize the basic response characteristics of the surrounding rock to the wavefield under the current working condition. The processing module is also used to construct a set of mutually exclusive state mechanism models in parallel based on the reference evidence package, and to construct a discriminant feature combination for the set of state mechanism models under a time window system. The time window system includes a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the tunnel face, and a far-field time window for constraining the effect of the medium in front of the tunnel face. The processing module is also used to perform circumferential small-scale processing on the set of excitation points while keeping the set of receiver points unchanged. After angle migration, an azimuth subset is obtained, and an azimuth evidence package is acquired through the azimuth subset. Simultaneously, based on the azimuth response patterns corresponding to the discriminant feature combinations extracted from the azimuth evidence package within the same time window system, a first elimination process is performed on the set of state mechanism models to obtain candidate state mechanism models. The processing module is further used to perform a rebuttal experiment if the number of candidate state mechanism models is determined to be greater than a preset number, without changing the set of excitation points and receiver points, to perform a second elimination process on the candidate state mechanism models to determine the winning state mechanism model. The processing module is also used to map the winning state mechanism model into a state risk map. The state risk map uses axial range, circumferential offset, and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combination within the far-field time window, the circumferential offset is determined by the difference between the discriminant feature combination and different azimuth subsets, and the evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles. The processing module is also used to associate the state risk map with a preset disposal strategy library to output corresponding advanced geological prediction disposal strategies.

[0008] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0009] In a fourth aspect of this application, a non-transitory computer-readable storage medium is provided, which stores instructions that, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: This system systematically expands tunnel face geological prediction from the traditional single-interface reflection identification model to a comprehensive identification and decision support mechanism oriented towards the evolution of surrounding rock mechanical state. Through the solidification of working condition context, the construction of benchmark evidence packages, and the constraint of time window systems, it ensures strict comparability of wavefield responses under different construction cycles and observation conditions, thereby reducing the interference of construction disturbances and support coupling on prediction results from the source. Furthermore, by constructing a set of mutually exclusive state mechanism models in parallel and introducing a multi-layered elimination mechanism of azimuth subsets and counter-evidence experiments, it enables non-interface-type risk bodies to be stably identified and distinguished through mechanism attribution, avoiding the simplistic attribution of ambiguous anomalies. Introducing noise or uncertain information; further mapping the winning state mechanism model into a state risk map including axial range, circumferential offset, and evolution trend, elevating the forecast results from "whether an anomaly exists" to an operational expression of "where the anomaly is, in which direction it is developing, and whether it is approaching"; finally, through direct association with a pre-set disposal strategy library, achieving a closed-loop connection from wavefield observation and risk identification to construction disposal, enabling detectable risk signals to be effectively transformed into executable early warning and construction decisions, thereby significantly improving the sensitivity, reliability, and engineering applicability of the tunnel face advanced geological prediction system to non-interface risk bodies under complex surrounding rock conditions. Therefore, it facilitates improving the accuracy of advanced geological prediction applied to tunnel faces. Attached Figure Description

[0011] Figure 1 A flowchart illustrating an advanced geological prediction method applied to a tunnel face, provided as an embodiment of this application; Figure 2 A schematic diagram of a module for an advanced geological prediction device applied to a tunnel face, provided as an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0014] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0015] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0016] To address the aforementioned technical problems, this application provides an advanced geological prediction method applied to tunnel faces, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an advanced geological prediction method applied to tunnel faces, provided as an embodiment of this application. The method is applied to a server and includes steps S110 to S170, as follows:

[0017] S110. After advancing to the preset mileage at the tunnel face and completing the on-duty support closure, determine the set of excitation points and the set of receiving points, and establish a working context bound to the set of excitation points and the set of receiving points.

[0018] Specifically, in this application embodiment, a server refers to a computing device or system used to centrally provide computing power, data storage, and business services. The server can centrally receive data from field devices, sensors, or acquisition terminals, parse, store, calculate, or discriminate this data, and return the processing results to the upper-level system or field terminal. This method uses seismic wave reflection to acquire reflected wavefield data at the physical acquisition level. At the data processing level, it models and eliminates mechanisms based on the two-way propagation structure, multi-channel consistency structure, and azimuth-sensitive structure of the reflected wavefield. At the engineering decision-making level, it transforms non-interface anomalies in the reflected wavefield into an operable state-risk map. After the tunnel face advances to the preset mileage and the shift's support is closed, the determination of the excitation point set and the receiver point set are performed first, ensuring that the subsequent seismic wave observation geometry remains stable and reproducible within this blasting cycle. The excitation point set refers to the set of spatial locations near the tunnel face used to generate a controlled wave field. Each excitation point corresponds to a fixed excitation position parameter and a fixed excitation coupling mode parameter. The excitation coupling mode parameter is used to describe the energy transfer contact pattern between the excitation device and the surrounding rock and to avoid contact condition drift at the same excitation point in different measurements. The receiving point set refers to the set of spatial locations near the tunnel face used to collect the wave field response. Each receiving point corresponds to a fixed receiving position parameter, a fixed installation attitude parameter, and a fixed receiving coupling mode parameter. The installation attitude parameter is used to describe the correspondence between the sensor's sensitive axis direction and the tunnel coordinate reference. The receiving coupling mode parameter is used to describe the contact fixation pattern between the sensor and the surrounding rock or support structure to ensure signal comparability. Preset mileage refers to the predefined mileage trigger conditions, which stipulate that the current round of advanced geological forecasting will be carried out after the working face has advanced to a specific position, so that the forecast trigger points between different cycles have a consistent spatial reference; shift support closure refers to the completion of key closure links of support components such as shotcrete, anchor bolts, steel arch frames or linings within the current construction shift, so that the surrounding rock-support system enters a relatively stable state, thereby reducing the disturbance of wave field acquisition during the support construction process.

[0019] The working context refers to the set of parameters describing the construction and structural boundary conditions at the current forecast time. It includes at least the tunnel face mileage marker, blasting cycle marker, support age parameter, surrounding rock exposure time parameter, tunnel face cross-sectional morphology parameter, and construction equipment operating status. Among them, the tunnel face mileage marker is used to provide axial spatial reference, the blasting cycle marker is used to establish a time series correlation between the current forecast and the previous and subsequent cycles, the support age parameter is used to describe the influence of the support material strength on the coupling conditions as time increases, the surrounding rock exposure time parameter is used to describe the influence of surrounding rock unloading and relaxation on the wave field response, the tunnel face cross-sectional morphology parameter is used to describe the influence of under-excavation, over-excavation, and local concavity and convexity on the observation geometry, and the construction equipment operating status is used to describe the structured background vibration conditions generated by tunneling equipment, transportation equipment, or ventilation equipment.

[0020] S120. Under the constraints of the working condition context, a reference gather is obtained by using the set of excitation points and the set of receiver points, and source consistency shaping is performed on the reference gather to form a reference evidence package. The reference evidence package is used to characterize the basic response characteristics of the surrounding rock to the wave field under the current working condition.

[0021] Specifically, when performing observation geometry locking processing on the excitation point set and the receiver point set, an observation geometry data structure is first established on the server to solidify the spatial relationships deployed on-site into reusable digital constraints. Observation geometry locking processing refers to fixing the spatial positions and installation states of the excitation point set and the receiver point set within the current working condition context, ensuring comparability of subsequent acquisitions under the same observation geometry. Unique identifiers are assigned to each excitation point and each receiver point, ensuring that any waveform can be traced back to its corresponding excitation point and receiver point. Spatial position parameters refer to the coordinate values ​​and borehole depth values ​​of the excitation point or receiver point relative to the tunnel coordinate reference. Azimuth parameters refer to the azimuth angle or azimuth number of the excitation point or receiver point relative to the circumferential reference of the tunnel face. Installation attitude parameters refer to the directional relationship between the sensor sensitive axis of the receiver point and the tunnel coordinate reference. Coupling method parameters refer to the contact and fixing method between the excitation device or receiver device and the surrounding rock or support structure, used to constrain the consistency of energy transfer and signal acquisition. When the server associates and stores the unique identifier with the operating condition context, it binds the operating condition identifier of the operating condition context with the observation geometry locking result, so that any subsequent measurement can only write data under the observation geometry constraints corresponding to the same operating condition identifier, thereby implementing the data acquisition comparability as a strong consistency constraint at the data structure level.

[0022] The server first issues the excitation configuration and establishes an excitation scheduling queue, creating a one-to-one correspondence between the excitation point identifier and the excitation order, number of excitations, and excitation interval. The preset excitation configuration refers to the unified setting of the excitation waveform, main frequency band, energy budget, duration, triggering method, and number of repetitions, ensuring consistent excitation conditions between different excitation points and different measurements. Repeated excitation operation refers to exciting the same excitation point multiple times within a short period, allowing for subsequent superposition to suppress random noise and enhance repeatable wavefield components. The excitation timestamp is a high-precision time marker for each excitation trigger, used to strictly bind the excitation event and received data on the timeline. When generating the excitation timestamp bound to the operating condition context, the server writes the excitation timestamp to the event table of the operating condition context, and simultaneously writes the excitation point identifier and excitation repetition number, ensuring that each excitation has a triple index: operating condition traceability, excitation traceability, and repetition number traceability.

[0023] The server issues a unified sampling gating command to all receiving channels, enabling each channel to acquire and transmit waveforms within the same sampling window. The original gather segment refers to the collection of multiple waveforms acquired by multiple receiving channels within the same time window after a single excitation trigger. Synchronous acquisition means that each receiving channel aligns its sampling window under the same time reference to reduce differences in start times between channels. Time alignment processing refers to unifying the waveform start points of each receiving channel to the same reference zero point on the server side, thereby eliminating the influence of sampling start deviations and clock micro-drifts on timing and phase characteristics. Time alignment processing can obtain the time delay correction amount of each receiving channel relative to the reference channel through cross-correlation peak location, and perform waveform shift alignment accordingly. The reference channel can be a receiving channel with stable signal-to-noise ratio and stable coupling, as detailed below:

[0024] in, Indicates the first The delay correction amount for each receiving channel relative to the reference channel, with a value range from... limited, It is determined by both the sampling rate and the maximum possible channel initial deviation; Indicates the reference channel in the discrete sampling index Waveform amplitude at the location; Indicates the first Each receiving point channel in discrete sampling index Waveform amplitude at the location; The number of sample points participating in the cross-correlation calculation is determined by the near-field time window or a stable time window containing the direct wave to improve positioning reliability.

[0025] The server generates quality labels for each raw gather fragment and decides whether to retain or discard it. Simultaneously, it triggers resampling after discarding fragments to maintain integrity in the excitation repetition dimension. Quality constraint screening removes fragments with obvious distortion, saturation, sudden spikes, or noise anomalies before source consistency shaping, preventing these abnormal fragments from contaminating subsequent fingerprint construction and robust overlay. Channel consistency conditions ensure that the waveform energy distribution, phase structure, or critical arrival times across multiple channels under the same excitation do not exceed reasonable thresholds. Saturation conditions indicate that the waveform exhibits a truncation of the upper limit of sampling amplitude or a prolonged flat top, signifying signal distortion. Abnormal spike conditions indicate isolated high-amplitude pulses in the waveform that lack cross-channel consistency. Noise stability conditions ensure that the statistical characteristics of background noise outside the near-field time window remain stable between adjacent repeated excitations. The server can simultaneously calculate the saturation ratio, peak ratio, channel energy dispersion, and background noise drift, and combine them into a segment quality criterion. When the criterion does not meet the preset threshold, the segment is marked as an invalid segment, and supplementary sampling is performed with the same excitation point identifier and the same excitation configuration until the integrity requirement of the excitation repetition number dimension is met, thereby ensuring that each subsequent excitation point is shaped and superimposed on the basis of the same amount of data with the same quality threshold.

[0026] The server first extracts a near-field time window from the valid segments corresponding to each excitation point, forming a set of fingerprint segments to characterize the source output and coupling conditions. The near-field time window refers to the time range closest to the excitation source, least affected by the propagation path, and able to stably reflect the source's initiation and coupling state, used to exclude propagation effects from the fingerprint as much as possible. The source fingerprint refers to the statistical expression of the waveform morphology within the near-field time window, used to describe the amplitude envelope shape, phase initiation structure, and frequency band distribution characteristics of the excitation point under the current operating context. Fingerprint matching and correction refers to selecting the target source fingerprint as a unified reference and mapping other source fingerprints to the expression space of the target source fingerprint through amplitude normalization, phase alignment, and frequency band shaping transformation, enabling superposition between different excitation points or different repeated excitations. The shaping parameter set refers to the set of parameters required for the above mapping, including at least amplitude scaling parameters, phase correction parameters, and frequency band correction parameters. To ensure that the shaping does not introduce excessive distortion, the server selects the target source fingerprint as one with a high signal-to-noise ratio, low saturation ratio, and stable near-field coupling. It also binds and stores the shaping parameter set of each excitation point with the excitation point identifier, so that subsequent shaping transformations of the original gather fragments have consistent parameter basis and traceability.

[0027] The server first applies a set of shaping parameters corresponding to the same excitation point identifier to each original gather segment, causing all repeating segments from the same excitation point to converge in amplitude, phase, and frequency band. Robust stacking is then performed to suppress residual anomalies. Amplitude correction refers to unifying the energy levels of different segments using amplitude scaling and envelope normalization; phase correction refers to aligning the oscillation phase with the key wave group phase using a unified phase reference; frequency band correction refers to shaping the spectral distribution to ensure a consistent main frequency band and suppress abnormally narrow or wide bands; robust stacking processing assigns lower weights to outlier segments during stacking averaging to prevent a small number of outlier segments from dominating the stacking result. Robust stacking can be implemented using a residual-driven weighting function. First, an initial stack is used to obtain a reference waveform. Then, the residual amplitude of each segment relative to the reference waveform is calculated, and the weights are updated accordingly. This process is iterated until the weights converge or the preset number of iterations is reached, thereby suppressing outlier interference without sacrificing repeatable wavefield components. Specifically:

[0028] in, This indicates the normalized waveform after robust superposition at the discrete sampling index. The amplitude at that point; Indicates the first The corrected segments are in the discrete sampling index The amplitude at that point; This indicates the number of valid fragments participating in the stacking under the same excitation point identifier, which is determined by quality constraint screening and supplementary sampling results. Indicates the first The weight of each segment is limited to a non-negative range and can be normalized. The weight is obtained by statistical analysis of the residual between the segment and the reference waveform. The larger the residual, the smaller the weight.

[0029] The server generates a baseline evidence package index according to a fixed data encapsulation format, ensuring that the baseline evidence package contains both the result data and the constraint links that generated the result. The baseline evidence package refers to a unified encapsulation carrier of the basic response characteristics under the current working condition context. The standardized baseline gather is the core data used for subsequent state mechanism model construction and discriminant feature extraction. The working condition context is used to define the interpretation boundary and support cross-blast cycle comparison. The observation geometry locking results are used to ensure spatial reference consistency. The shaping parameter set is used to reproduce and audit the source consistency shaping process. The quality constraint screening records are used to prove that the segments entering the stack meet the unified threshold and explain the reasons for the removed segments. During writing, the server assigns a unique baseline evidence package identifier to the baseline evidence package and writes the identifier back to the event table of the working condition context. This allows any subsequent state risk map or disposal strategy output to be traced back to the original gather segments, shaping parameter set, and removal records through the working condition identifier and the baseline evidence package identifier, thus achieving a complete traceability loop from basic response characteristics to subsequent early warning decisions.

[0030] S130. Construct a set of mutually exclusive state mechanism models in parallel based on the benchmark evidence package, and construct a combination of discriminative features for the set of state mechanism models under the time window system. The time window system includes a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the working face, and a far-field time window for constraining the effect of the medium in front of the working face.

[0031] Specifically, the server first performs evidence structure verification on the benchmark evidence package, ensuring that the standardized benchmark gather, channel index mapping, and time index mapping remain read-only consistent during this round of mechanism construction, thereby avoiding boundary offsets for the same waveform slice in different processing branches. Evidence structure refers to the fixed organization of data fields and index fields in the benchmark evidence package, used to ensure that waveform segments corresponding to the same receiver point identifier are consistently located in any processing branch; the time window system refers to near-field, mid-field, and far-field time windows divided according to unified rules, used to constrain support coupling effects, describe the propagation stability of the medium adjacent to the tunnel face, and describe the state changes of the medium ahead of the tunnel face, respectively, without changing semantic boundaries; parallel construction refers to running the feature extraction and scoring processes of multiple state mechanism models simultaneously on the same benchmark evidence package, enabling each state mechanism model to share the same data source and the same index constraints, thereby achieving comparability. The state mechanism model set includes stress concentration models, low-velocity zone models, and energy dissipation zone models. The stress concentration model characterizes the degradation of phase synchronization and enhanced directional differences caused by the surrounding rock mechanical state field. The low-velocity zone model characterizes the travel-time structural bending and low-frequency migration caused by the reduced equivalent propagation velocity of the preceding medium. The energy dissipation zone model characterizes the shortened energy lifetime and tail coherence failure caused by scattering and enhanced energy dissipation. The mutual exclusion constraint table is a constraint data structure that solidifies the model competition relationship at the output layer. It limits the output to only one winning state mechanism model within the same spatial interval, while requiring rejected state mechanism models to be written into the rejection evidence field for traceability. The same spatial interval refers to the interval unit bounded by the equivalent distance range corresponding to the mapping between the tunnel face axial reference and the time index, used to unify the comparison granularity of different models. The winning state mechanism model is the unique model output selected after passing subsequent elimination and consistency checks under the constraints of the mutual exclusion constraint table.

[0032] The server first performs consistency quantization on the multi-channel phase trajectories within the overlapping region to construct a characterization of phase-locking capability. Phase-locking capability refers to the degree of phase synchronization of different receiving point channels for the same wavegroup within the same time window, reflecting the stability of the propagation path set and whether the medium's microstructure has rearranged. Phase-locking capability degradation refers to a systematic decrease in phase synchronization over time windows or with each burst cycle, related to stress concentration-induced micro-crack opening and closing, contact stiffness fluctuations, and increased wave velocity perturbations. Phase-locking capability can be quantified using phase-locking values. The server first extracts the instantaneous phase of each channel waveform within the target frequency band, then calculates the concentration of cross-channel phase differences, as follows:

[0033] in, Indicates time index The phase lock value at the location ranges from 0 to 1. The closer to 1, the more synchronized the phases are; the closer to 0, the more dispersed the phases are. This indicates the number of channel pairs participating in the statistics. Channel pairs are obtained by combining the receiver point identifiers of the receiver point set. Indicates the first Each channel pair in time index The instantaneous phase difference at a given point is obtained by subtracting the instantaneous phases of the two channels; The imaginary unit is used to map the phase difference onto the unit circle to statistically determine the degree of concentration. Azimuth response bias enhancement refers to the stable directional differences in phase-locked values ​​between different azimuth subsets after the set of excitation points migrates at small angles in the circumference, exhibiting a repeatable bias pattern. The server uses this bias pattern as auxiliary evidence for the stress concentration model and requires that this bias pattern simultaneously occur within the overlapping region along with the degradation of phase-locked capability, thereby avoiding misjudging simple noise fluctuations as stress concentration.

[0034] The server fits the arrival times of multiple channels within the first-stage interval and calculates the deviation from the baseline propagation trend. Travel curvature refers to the nonlinear change in the arrival times of the first wave or main energy at different receiving points as the channel's spatial position changes, reflecting an equivalent velocity reduction or anomaly in the propagation path. The increased proportion of low frequencies indicates a greater concentration of waveform energy towards lower frequencies, caused by the easier scattering and energy attenuation of higher frequencies. Travel curvature can be characterized by the curvature of the arrival time residuals. The server first extracts the arrival times on each channel, then performs a second-order fitting between the arrival times and the channel positions, using the quadratic coefficient as the curvature strength, as calculated below:

[0035] in, Indicates the first The arrival time of each receiving point channel is extracted in the early part of the far-field time window. The extraction method can be based on the arrival time of the envelope peak or cross-correlation aligned arrival time. Indicates the first The channel position scalar of each receiving point channel after projection along the axial or circumferential reference is determined by the observation geometry locking result; This represents the travel-time bending strength parameter. The larger the value, the more pronounced the curvature of the travel curve. and These represent the linear and constant parameters, respectively, used to fit the baseline propagation trend. This fitting captures the nonlinear deviation of the relative position during travel time through a quadratic term, thus transforming the travel time structure changes caused by velocity anomalies into comparable scalar intensities. The increase in the proportion of low-frequency frequencies can be characterized by the proportion of low-frequency energy. The server performs frequency band energy integration on the waveform within the first segment of the same far-field time window, calculating the proportion of low-frequency energy to the total energy, as follows:

[0036] in, This indicates the proportion of low-frequency energy, with a value ranging from 0 to 1; Represents frequency The power spectral density or discrete frequency energy at a given point is calculated from the spectral distribution trajectory within the target time window. This represents the set of low-frequency bands, whose upper and lower limits are determined by the preset excitation main frequency band and the on-site noise spectrum. It represents the total set of frequency bands, covering the effective frequency band range of the spectral distribution trajectory.

[0037] The server performs statistical analysis on the envelope attenuation process and quantifies the consistency across channels at the tail end within the latter half of the interval. Energy lifetime refers to the duration of waveform energy on the time axis from the arrival of the main energy until a certain attenuation threshold, reflecting the energy dissipation and scattering intensity of the medium on the wave field. Shortened energy lifetime refers to a systematic decrease in energy lifetime with blast cycles or azimuth subsets under the same operating context constraints, indicating that energy is dissipated faster or scattered more strongly. Tail coherence refers to the degree to which the far-field tail waveform maintains a similar shape across different receiving point channels, reflecting the stability of the multipath scattering structure. Deterioration of tail coherence refers to a decrease in the cross-channel similarity of the tail waveform, exhibiting a diffuse shape that cannot be stably focused. Energy lifetime can be calculated using the time difference between the envelope energy reaching the threshold. The server first calculates the envelope energy sequence and determines the peak time of the main energy and the cross-threshold time, as follows:

[0038] in, This represents the energy lifetime and takes a non-negative value. This indicates the moment when the envelope energy reaches its peak value, which is obtained by searching within the latter part of the far-field time window. This indicates the moment when the envelope energy first decays to a threshold value. The threshold is either a fixed percentage of the peak energy or a multiple of the background noise energy. The threshold percentage or multiple is determined by a combination of noise stability conditions and engineering experience. Tail coherence can be quantified using the average tail cross-correlation value. The server calculates the normalized cross-correlation peak value for each pair of channels within the latter part of the far-field time window and takes the average to characterize the consistency of the tail shape, as follows:

[0039] in, This indicates tail coherence, with a value ranging from 0 to 1. The larger the value, the more similar the tails are. This indicates the number of channel pairs participating in the statistics; and They represent the first The two tail waveform sequences of each channel pair are taken from the later interval of the far-field time window and truncated according to the same time index. This indicates the permissible micro-scale shift range used to compensate for residual alignment errors. It is determined by the sampling rate and the allowable residual time shift.

[0040] The server first defines a discriminative feature vocabulary and solidifies the feature extraction order, ensuring that the same feature name maintains the same meaning and calculation rules across different time windows and state mechanism models, avoiding distortion caused by homonyms or homonyms. Discriminative feature combinations refer to a set of features with unified naming and semantics, used to support the discrimination and mutually exclusive selection of state mechanism models. Near-field constraint feature vectors refer to feature vectors extracted within the near-field time window, used to determine the stability of support coupling effects and the effectiveness of source consistency shaping, thus serving as pre-thresholds for mid-field and far-field discriminative feature vectors. Mid-field discriminative feature vectors refer to feature vectors extracted within the mid-field time window, used to describe the propagation stability of the medium adjacent to the tunnel face and provide primary evidence input for the stress concentration model. Far-field discriminative feature vectors refer to feature vectors extracted within the far-field time window, used to describe the state changes of the medium ahead of the tunnel face and provide primary evidence input for the low-velocity zone and energy-dissipating zone models. When forming the three types of feature vectors, the server calculates feature terms for each receiver point identifier and each excitation point identifier and performs cross-channel consistency convergence. This ensures that the output feature vectors maintain sensitivity to spatial differences while suppressing single-channel anomalies through consistency convergence. Subsequently, the near-field constraint feature vector, mid-field discriminant feature vector, and far-field discriminant feature vector, along with the mutual exclusion constraint table, are written into the mechanism evidence package. A feature reading path is established for each state mechanism model. The stress concentration model prioritizes reading the phase-locking capability representation in the mid-field discriminant feature vector and uses the passage of the near-field constraint feature vector as a prerequisite. The low-speed band model prioritizes reading the travel time curvature representation and low-frequency energy ratio representation in the far-field discriminant feature vector and uses the passage of the near-field constraint feature vector as a prerequisite. The energy dissipation band model prioritizes reading the energy lifetime representation and tail coherence representation in the far-field discriminant feature vector and uses the passage of the near-field constraint feature vector as a prerequisite. This enables comparable parallel operation and mutually exclusive competitive output of the three types of state mechanism models under the same evidence structure and the same time window system.

[0041] S140. While keeping the set of receiving points unchanged, the set of excitation points is moved circumferentially at a small angle to obtain a azimuth subset. A azimuth evidence package is obtained through the azimuth subset. At the same time, based on the azimuth response pattern corresponding to the combination of discriminative features extracted from the azimuth evidence package under the same time window system, the first elimination process is performed on the set of state mechanism models to obtain candidate state mechanism models.

[0042] Specifically, the server first reads the observation geometry locking results and the zero point of the tunnel face cross-section from the working condition context, establishes a circumferential reference system, and enables small-angle circumferential migration to be repeatedly executed under the same circumferential reference. The tunnel face center axis refers to the axis passing through the geometric center of the tunnel face along the tunnel axial direction, used to define the circumferential angle reference; small-angle circumferential migration refers to making a small-angle controlled change to the position of the excitation point only in the circumferential direction of the tunnel face without changing the receiver point set, thereby changing the incident azimuth of the wave field to trigger an azimuth-sensitive response; the migration azimuth position refers to the new spatial position of each excitation point after migration; the migration angle parameter refers to the angular offset relative to the circumferential reference system; the migration direction parameter refers to the clockwise or counterclockwise migration direction identifier; the hole depth parameter refers to the depth value of the excitation hole or coupling hole, used to constrain coupling and source output stability; the coupling method parameter refers to the contact and fixing method between the excitation device and the surrounding rock or support structure. When generating the migration azimuth positions, the server maps the reference position of each excitation point to the migration position according to a preset angle constraint, and encapsulates the migration angle parameters, migration direction parameters, hole depth parameters, and coupling method parameters into an azimuth geometry locking result. This allows the set of excitation points to form multiple azimuth subsets under different migration azimuths, and each azimuth subset can be reproduced through the azimuth geometry locking result, as detailed below:

[0043] in, Indicates the first The first location subset The migration azimuth coordinates of each excitation point; Indicates the first The reference position coordinates of each excitation point are given by the observation geometry locking results; The coordinates of the geometric center or equivalent rotation center of the tunnel face are determined by the morphological parameters of the tunnel face. Indicated by migration angle parameter The generated two-dimensional rotation matrix, The sign is determined by the preset angle constraint and the migration direction parameter.

[0044] The server first updates the receiver verification record on the receiver point set to confirm that the receiver point identifier, installation attitude parameters, coupling mode parameters, and channel parameters remain unchanged across each azimuth subset, thus ensuring that azimuth differences originate only from the circumferential small-angle migration of the excitation point set. Azimuth acquisition refers to triggering excitation according to the excitation configuration at the migration azimuth position corresponding to a certain azimuth subset, and synchronously acquiring azimuth gathers from the receiver point set. An azimuth gather refers to the set of multiple waveforms acquired by each receiver point channel within that azimuth subset. Source consistency shaping aligns the azimuth gathers to the same reference source fingerprint in terms of amplitude envelope, phase initiation structure, and frequency band distribution, making them comparable across different azimuth subsets. Time alignment eliminates sampling start-up bias between channels using excitation timestamps or cross-correlation alignment, enabling stable comparison of travel time and phase features across azimuth subsets. Standardized azimuth gathers are the azimuth gathers after source consistency shaping and time alignment, used for subsequent extraction of discriminative feature combinations within the same time window system. When the server writes the standardized azimuth gathers, azimuth geometry locking results, reception verification records, and trigger configurations into the azimuth evidence package, it also writes the azimuth subset identifier and trigger point identifier index, so that the azimuth evidence package has a complete link with traceable azimuth subsets, traceable trigger points, traceable reception points, and reproducible processing.

[0045] The server replicates the boundary definitions of the time window system from the baseline evidence package, ensuring that the azimuth evidence package and the baseline evidence package use the same slice range under the same time index mapping. The near-field time window refers to the time range used to constrain the coupling effects of support and the source coupling stability; the mid-field time window refers to the time range used to describe the propagation stability of the medium adjacent to the tunnel face; the far-field time window refers to the time range used to describe the changes in the state of the medium ahead of the tunnel face; slice processing refers to extracting corresponding waveform segments from the standardized azimuth trace set according to the time window boundaries; the discriminant feature combination refers to the feature set extracted according to unified naming and semantic rules, ensuring that features with the same name have the same meaning across different azimuth subsets, different time windows, and different state mechanism models. When extracting discriminant feature combinations within each time window, the server uses the fixed feature extraction rules and feature reading paths in the mechanism evidence package, aggregating the discriminant feature combinations of each azimuth subset into an azimuth feature vector, and binding and storing the azimuth feature vector with the azimuth subset identifier, enabling subsequent azimuth response pattern construction to perform trend comparisons within the same feature space.

[0046] The server uses the migration angle parameter as the independent variable index and the corresponding features in the azimuth feature vector as the dependent variable sequence to generate an angle-feature response curve. Monotonicity and stability constraints are applied to the response curve to eliminate random fluctuations. The azimuth response pattern refers to the directional pattern of the corresponding features as they migrate circumferentially, used to characterize the sensitivity of the surrounding rock or the medium ahead to changes in the incident azimuth. The trend refers to the overall direction and magnitude of change of the response curve across multiple azimuth subsets. The server then compares the azimuth response pattern with the operating status of the construction equipment in the working context. The operating status of the construction equipment refers to the periodic background vibration status code or vibration signature generated by tunneling equipment, transportation equipment, ventilation equipment, etc. The consistency comparison checks whether the azimuth response pattern is highly consistent with the rhythmic vibration of the construction equipment in frequency or phase. If they are consistent, it indicates that the azimuth response pattern may be contaminated by background vibration. Disturbance marking marks the azimuth response patterns determined to be affected by background vibration and reduces their weight or triggers a re-sampling strategy in the first elimination process to avoid misjudging the rhythmic vibration of the construction equipment as true azimuth sensitivity. Specifically:

[0047] in, Representation of features The trend slope in the orientation subset dimension is used to characterize the strength of directionality as the migration angle parameter changes; Indicates the first The migration angle parameters of each azimuth subset are given by the azimuth geometry locking results; This represents the mean value of the migration angle parameter; Indicates the first Among the directional feature vectors of each directional subset, the features with the same name The value of is determined by the result of the discriminant feature combination extraction. This indicates the number of directional subsets involved in trend estimation.

[0048] The server performs an availability check on the near-field constraint feature vectors. This check confirms that the support coupling effect and the source coupling state remain stable during this round of azimuth acquisition. Only models that pass the availability check are allowed to enter the azimuth response mode-driven mechanism competition. A mutual exclusion constraint table limits the output of only a single state mechanism model within the same spatial interval, requiring rejected models to be written into the rejection evidence field. The mechanism discrimination entry table specifies which features with the same name are read by each state mechanism model, within which time windows, and which azimuth response mode indicators are used as consistency criteria. The pre-defined azimuth response expectation refers to the azimuth response pattern fixed for the stress concentration model, low-speed zone model, and energy dissipation zone model, respectively. The pre-defined azimuth response expectation of the stress concentration model is reflected in the phase locking capability related features being more sensitive to the azimuth and having a stable offset direction. The pre-defined azimuth response expectation of the low-speed zone model is reflected in the travel time curvature related features being relatively insensitive to the azimuth and showing more cross-azimuth consistency. The pre-defined azimuth response expectation of the energy dissipation zone model is reflected in the energy lifetime and tail coherence related features showing low azimuth dependence across different azimuth subsets but consistent degradation in the later part of the far-field time window. When the server performs a consistency check between the orientation response pattern of any state mechanism model and its preset orientation response expectation, if the consistency check fails, the state mechanism model is marked as an eliminated state mechanism model and written into the rejection evidence field. The rejection evidence field includes at least the same-name feature identifier of the failed trigger, the corresponding time window identifier, and the corresponding trend slope identifier. If the consistency check passes and the near-field constraint feature vector availability check passes, the state mechanism model is determined as a candidate state mechanism model and a set of candidate state mechanism models is output. This allows subsequent disproven experiments to continue to be performed only on the set of candidate state mechanism models, thereby transforming the directional information brought by the orientation subset into a traceable first-round elimination evidence chain.

[0049] S150. If the number of candidate state mechanism models is greater than the preset number, a proof-of-contrast experiment is performed without changing the set of excitation points and the set of receiver points to perform a second elimination process on the candidate state mechanism models in order to determine the winning state mechanism model.

[0050] Specifically, the candidate state mechanism model refers to a state mechanism model that still satisfies the near-field constraint feature vector availability test and whose azimuth response pattern is consistent with the preset azimuth response expectation after the first elimination process; the preset number refers to the threshold of the number of candidates used to trigger the disproving experiment, which is used to avoid additional acquisition when the number of candidates has converged to a single model; the complementary excitation pair refers to a pair of excitation sequences consisting of the first complementary excitation and the second complementary excitation, which are consistent in energy budget, dominant frequency band and duration, and satisfy the complementary relationship in code pattern phase structure, so that the wave field response of the two excitations under ideal linear propagation conditions has predictable superposition and cancellation behavior; the energy budget refers to the total energy control amount released into the surrounding rock by a single excitation, which is used to constrain the consistency of excitation intensity; the dominant frequency band refers to the frequency range in which the excitation signal energy is concentrated, which is used to ensure that the spectral structure of the two complementary excitations is consistent; the duration refers to the effective duration of the excitation signal, which is used to ensure that the two complementary excitations are comparable in terms of time support; the code pattern phase structure refers to the phase coding sequence of the excitation signal in the time domain, which is used to construct the complementary relationship. When the server binds the complementary excitation pair to each excitation point identifier in the excitation point set, it solidifies the first complementary excitation code pattern, the second complementary excitation code pattern, the complementary excitation interval, and the triggering order for each excitation point identifier, and writes them into the disproving configuration table corresponding to the disproving identifier, so that any subsequent complementary channel set pair can trace back to the complementary excitation pair parameters based on the excitation point identifier and the disproving identifier.

[0051] The server schedules the excitation process according to the excitation point identifier and sends a synchronous acquisition gating to the receiving point set, so that each complementary excitation generates an excitation timestamp bound to it and forms a corresponding original gather fragment. The first complementary gather fragment refers to the set of multi-channel waveform fragments acquired by the receiver set after the first complementary excitation is triggered; the second complementary gather fragment refers to the set of multi-channel waveform fragments acquired by the receiver set after the second complementary excitation is triggered; time alignment processing refers to eliminating the sampling start deviation between channels based on the excitation timestamp or cross-correlation alignment, so that the waveforms corresponding to the two complementary excitations are comparable under the same time index mapping; source consistency shaping processing refers to using the source fingerprint constructed by the near-field time window and the shaping parameter set to perform amplitude correction, phase correction and frequency band correction on the complementary gather fragments, so that the two complementary excitations are comparable after the source output difference and coupling difference are suppressed; complementary gather pair refers to the paired data structure composed of the first complementary gather fragment and the second complementary gather fragment under the same excitation point identifier, which is used to construct the superposition response and differential residual; complementary identifier refers to the unique identifier of the complementary gather pair, which is used to distinguish different excitation point identifiers and different rounds of disproving experiments; disproving identifier refers to the unique identifier of this disproving experiment, which is used to establish the association between the complementary gather pair and the candidate state mechanism model set, the mutual exclusion constraint table and the mechanism discrimination entry table. When the server generates a counter-evidence package, it encapsulates the complementary gather pairs, complementary identifiers, counter-evidence identifiers, complementary excitation pair parameters, observation geometry locking results, and operating context snapshots together, and writes the encapsulation index into the risk memory to ensure that the counter-evidence package is traceable and reproducible.

[0052] The server replicates the boundary definition of the same time window system from the baseline evidence package, ensuring that the slice range of the disproving evidence package and the baseline evidence package is completely consistent in the near-field, mid-field, and far-field time windows. The superposition response refers to the synthetic response obtained by superimposing the same complementary gather pair in the time domain, used to test whether linear superpositionability holds under the same observation geometry and operating context. The differential residual refers to the residual response obtained by differentiating the same complementary gather pair in the time domain, used to quantify the degree of violation of linear superpositionability and reveal whether the violation is spatially concentrated and temporally stable. The superpositionability feature vector refers to a set of uniformly named and semantically unified features extracted from the superposition response and differential residual within the far-field time window, used to characterize whether linear superpositionability holds and the spatial concentration and temporal stability of the violation. Spatial concentration describes whether the differential residual forms a stable concentrated interval within the equivalent axial interval, and temporal stability describes whether the differential residual consistently appears between multiple complementary identifiers and multiple excitation point identifiers. The construction of the superimposed response and differential residual can be performed directly on each receiving point channel and calculated sample-by-sample under the same time index mapping, as follows:

[0053] in, Indicates the first Waveform samples of each receiving point channel after time alignment and source consistency shaping in the first complementary gather segment; Indicates the first Waveform samples of each receiver channel after time alignment and source consistency shaping in the second complementary gather segment; Indicates the first Superimposed response samples from each receiving point channel; Indicates the first Differential residual samples of each receiving point channel; This represents the discrete sampling index, determined by the time index mapping; The receiver identification index is determined by the receiver point set. This construction algebraically synthesizes the waveforms of two complementary excitations on the same sampling index, thereby transforming the validity of linear superposition into the statistical characteristic difference between the superposition response and the differential residual.

[0054] To characterize whether linear superposition holds, the server can calculate the proportion of differential residual energy within the far-field time window as a fundamental component in the superposition eigenvector, and converge it across multiple channels to obtain a unified scalar, as follows: in, This represents the proportion of energy in the differential residual. The larger the value, the stronger the differential residual is, and the more likely the linear superposition property is to be violated. This represents the set of receiving point channels participating in the aggregation, which is determined by the set of receiving points and allows for the removal of channels with abnormal quality labels; The set of sampling indices corresponding to the far-field time window is determined by the time window system and the time index mapping. This represents a stable term to prevent the denominator from being zero. Its value is positive and much smaller than the principal magnitude of the denominator, and it is preset by the server.

[0055] To characterize the spatial concentration of the damage, the server can perform sliding aggregation of the differential residual energy along the time index within a far-field time window and locate the concentrated interval where the energy peak is located. This concentrated interval is then correlated with an equivalent axial reference. Sliding aggregation can be performed using a fixed-length window. The energy trajectory is obtained by moving upwards. The concentrated interval is the continuous index segment of the energy trajectory that exceeds a preset proportional threshold. The threshold proportion is determined jointly by the noise stability condition and the background energy statistics in the benchmark evidence package, thereby avoiding misjudging uniform noise as a concentrated interval. To characterize temporal stability, the server can compare the degree of overlap and peak position drift of the concentrated intervals among multiple complementary identifiers. A high degree of overlap and a small degree of drift indicate strong temporal stability, while a high degree of overlap indicates weak temporal stability. This allows the repeatability of the disruption to be implemented as a stability component of the superimposability feature vector.

[0056] The server reads the anticipatory mapping rules for each candidate state mechanism model from the mechanism discrimination entry table, enabling different models to use the same superimposable feature vector but different consistency criteria during the anticipatory phase. Anticipatory mapping refers to the pre-defined description of the superimposable feature vector patterns that different candidate state mechanism models should exhibit in the anticipatory experiment. Specifically, the anticipatory mapping for the stress concentration model is reflected in the differential residual energy ratio not showing a stable high value and the concentration interval not having cross-complementary identifier consistency; the anticipatory mapping for the low-speed band model is reflected in the differential residual energy ratio maintaining a low value and the concentration interval not forming a stable cluster; and the anticipatory mapping for the energy dissipation band model is reflected in the differential residual energy ratio increasing and the concentration interval having stable clustering and cross-complementary identifier consistency in the later part of the far-field time window. The mechanism discrimination entry table specifies which superimposable feature vector components each state mechanism model reads during the anticipatory phase, and which thresholds or consistency conditions are used to determine pass or fail. The mutual exclusion constraint table restricts the output of only one winning state mechanism model within the same spatial interval, requiring the rejected model to write into the rejection evidence field. During the consistency check, the server matches the counter-evidence expectation of each candidate state mechanism model with the superimposability feature vector. When any candidate state mechanism model does not meet the corresponding counter-evidence expectation, it is marked as an eliminated state mechanism model, and a complementary identifier, the identifier of the superimposability feature vector component that failed to trigger, the corresponding far-field time window index range, and the central interval index range are written into the rejection evidence field. After the second elimination process is completed, the candidate state mechanism models that were not marked as eliminated state mechanism models are determined as winning state mechanism models, and the winning state mechanism model identifier and the counter-evidence package index are written into the mechanism evidence package together, so that the subsequent state risk map mapping and disposal strategy association can directly refer to the traceable evidence chain of the counter-evidence stage.

[0057] S160. Map the winning state mechanism model into a state risk map. The state risk map uses axial range, circumferential bias and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combination within the far field time window. The circumferential bias is determined by the difference between different directional subsets of the discriminant feature combination. The evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles.

[0058] Specifically, the server first reads the evidence-bearing interval rules and feature reading paths corresponding to the winning state mechanism model from the mechanism evidence package. This ensures that the evidence-bearing interval is strictly limited to the time range most relevant to and most interpretable to the winning state mechanism model within the time window system. The evidence-bearing interval refers to the time segment range declared as the main discrimination criterion by the winning state mechanism model within the near-field, mid-field, and far-field time windows. This is used to avoid introducing wavefield segments unrelated to the winning state mechanism model into the mapping. The evidence-bearing channel set refers to the subset of receiving point channels that have been screened and marked as valid by quality constraints in the receiver point set and whose receiving verification records remain stable during the azimuth acquisition phase. This is used to ensure the reliability of cross-channel consistency verification. The channel index mapping refers to the fixed mapping relationship between the receiver point identifier and the acquisition channel sequence number. This is used to align the same receiver point identifier in different evidence packages to the same channel dimension. The time index mapping refers to the mapping relationship between the sampling index and the unified time axis. This is used to align waveform slices from different measurements and different azimuth subsets to the same time reference. Under unified channel index mapping and time index mapping constraints, when the server reassembles the discriminative feature combination into a mapping feature package within the evidence-bearing interval, it merges the features with the same name in the discriminative feature combination according to three semantic categories: primary evidence features, veto evidence features, and constraint evidence features. These three semantic categories are then bound and solidified with the winning state mechanism model. The mapping feature package refers to the structured carrier used to transition from mechanism discrimination to spatial risk expression. Primary evidence features are those that directly drive spatial positioning; veto evidence features are those used to narrow boundaries and eliminate false anomalies; and constraint evidence features are those used to shield the effects of support coupling and construction disturbances and control credibility. To ensure the traceability of the mapping feature package, the server simultaneously binds each feature item in the mapping feature package with a feature name, time window identifier, channel set identifier, and value aggregation rule. This ensures that subsequent axial range fields, circumferential offset fields, and evolution trend fields can all be traced back to the evidence-bearing interval and evidence-bearing channel set of the winning state mechanism model.

[0059] Concentrated scanning refers to forming a time series of the main evidence features on the sampling index set of the far-field time window, and calculating the local clustering intensity on the time series using a sliding window, thereby locating continuous intervals where the main evidence features continuously enhance. Cross-channel consistency testing refers to checking whether the above continuous intervals are simultaneously valid on most channels within the evidence-bearing channel set, used to exclude false concentrations caused by single-channel coupling defects or occasional noise. Consistency comparison of the disproving evidence package refers to conducting a co-occurrence test between the continuous intervals and the differential residual concentration intervals obtained in the disproving stage, so that only continuous intervals that also show stable destruction signs in the disproving stage are confirmed as evidence concentration intervals, thereby connecting superimposed destruction with spatial risk location. The evidence concentration interval refers to the continuous sampling index interval confirmed within the far-field time window, used as the location basis for the axial range field; the axial distance interval refers to the interval expression after mapping the evidence concentration interval from the time domain to the distance domain in front of the tunnel face; the axial range field refers to the field in the state risk map that expresses the influence range of the risk body with axial reference. Concentrated scanning can construct the cluster intensity trajectory of the main evidence features using sliding energy aggregation, and consecutive index segments exceeding a threshold are used as candidate intervals. The threshold is jointly constrained by the background statistics and quality constraints of the corresponding main evidence features in the benchmark evidence package, as follows:

[0060] in, Indicates sampling index Starting point, window length is The higher the clustering strength, the more concentrated the energy or consistency of the main evidence features are within that window; The number of sampling points covered by the sliding window is determined by the minimum interpretable duration expected within the far-field time window and the sampling rate. The set of channel indices representing the evidence-bearing channel set is determined by the evidence-bearing channel set; Indicates the first Each channel in the sampling index The time series values ​​of the main evidence features are extracted by the mapped feature package within the far-field time window according to the channel index mapping and aligned under the time index mapping. This expression aggregates the main evidence features of each channel within the window by square, so that paragraphs that are continuously enhanced and consistent across channels are... A clear peak is formed on the upper part, which can be used to locate the interval of candidate evidence set. The interval of evidence set is determined as the sampling index interval. Subsequently, when the server maps it to an axial distance interval, it needs to convert the time interval corresponding to the sampling index interval into a propagation distance and bind it to the face mileage identifier to output the axial range field. The propagation speed can be estimated by the equivalent speed of the stable segment in the baseline evidence package or determined by the reference speed given by the engineering prior. At the same time, the speed selection method is written into the constraint evidence feature for auditing purposes, as follows:

[0061] in, It represents the axial distance in front of the working face corresponding to a certain arrival time; This represents the reference propagation velocity, the value of which is determined by the propagation stability statistics of the baseline evidence package or by the reference table corresponding to the surrounding rock category. This represents the time difference from the excitation time to a representative time within the evidence set interval. The representative time can be the center time or the peak time of the evidence set interval, obtained by converting the sampling index to time using a time index mapping. The 2 in the denominator reflects the round-trip propagation path assumption from excitation to reception, making the time difference correspond to the one-way distance. The server uses this to... The corresponding time interval mapping is and will Together with the face mileage marker, they form the axial range field, enabling the axial range field to simultaneously express relative distance and provide a traceable mileage reference.

[0062] The server performs difference alignment on the same-named features of each azimuth subset within the same evidence-bearing interval, ensuring that the difference alignment strictly uses channel index mapping and time index mapping to avoid differences arising from slice drift. Difference distribution refers to the distribution of differences between the same-named features in different azimuth subsets as the circumferential azimuth changes; azimuth response difference refers to the directional expression of the difference distribution in the circumferential reference frame, used to capture directional scattering or directional energy dissipation caused by non-interface-type risk bodies; directional consistency test checks whether the azimuth response difference exhibits a stable monotonic or stable peak direction with the migration angle parameter, thereby excluding random fluctuations; construction equipment operating status disturbance verification compares the main frequency bands or main rhythms of the azimuth response difference with the rhythmic vibration characteristics of the construction equipment operating status; if highly consistent, a disturbance label is applied to the azimuth response difference and its confidence level is reduced. The central azimuth refers to the circumferential azimuth angle corresponding to the point where the azimuth response difference reaches its peak or the stable offset is at its maximum, used to characterize the main direction of the risk body's influence; the coverage sector refers to the circumferential angular range around the central azimuth, used to characterize the extent of the risk body's influence direction; the circumferential offset field refers to the field expressing the directionality of the risk body in the state risk map, composed of the central azimuth and the coverage sector, and consistent with the azimuth geometry locking result. The azimuth response difference can be constructed using the difference between the corresponding features of each azimuth subset and the reference azimuth, and the central azimuth is determined by angle-weighted vector synthesis, thereby avoiding the reduction of robustness due to relying only on the extreme points of a single azimuth subset, as detailed below:

[0063] in, Indicates features with the same name The azimuth response synthesis vector; This indicates the number of azimuth subsets, determined by the azimuth geometry locking result; Indicates the first Features with the same name in each location set The difference relative to the reference azimuth subset or the deviation relative to the mean of the azimuth subset is obtained from the azimuth feature vector; Indicates the first The circumferential azimuth angles corresponding to each azimuth subset are determined by the migration angle parameters and the circumferential reference frame. The center azimuth can be taken as... The direction angle and coverage sector can be determined by the angle range of differential distribution exceeding a preset proportional threshold. The threshold is jointly constrained by the noise stability condition and the disturbance mark state, so that the coverage sector automatically shrinks and the credibility of the circumferential offset field is reduced when the disturbance mark is activated. This allows the pollution of the directional response by the rhythmic vibration of the construction equipment to be explicitly incorporated into the constrained evidence features and output traceably.

[0064] The server performs field alignment on the state risk map of the previous blasting cycle and the current state risk map under the same coordinate semantics, enabling direct comparison of the axial range field and the circumferential offset field under the same axial reference, the same circumferential reference, and the same mileage reference. Coordinate semantics refers to the unified definition of axial reference, circumferential reference, and time reference, used to ensure that cross-cycle comparisons do not produce false migrations due to reference changes; field alignment refers to mapping the leading edge and trailing edge positions of the axial range field of the previous blasting cycle and the center orientation and coverage sector of the circumferential offset field of the current blasting cycle to the same expression scale; migration status refers to the amount of change of the above fields with the blasting cycle sequence, used to characterize whether the risk body is approaching the working face, whether directional drift has occurred, or whether expansion or contraction has occurred; continuity constraint judgment means that only when the migration status remains in the same direction or maintains the same pattern for no less than a preset number of consecutive cycles will a clear trend be output; otherwise, an unstable trend will be output and the credibility will be reduced. The evolution trend field refers to the field in the state-risk map that expresses the direction and stability of the risk body's evolution over time. It includes at least the trend type and trend confidence level, enabling the association of treatment strategies to distinguish between continuously approaching high-risk states and low-confidence states with random fluctuations. To implement continuity constraint judgment, the server constructs a cross-cycle change sequence for the center distance of the axial range field and the center orientation of the circumferential offset field, and applies joint constraints of sign consistency and amplitude threshold to the change sequence. The amplitude threshold is estimated from the repeated measurement fluctuation range of the baseline evidence package, thereby distinguishing between interpretable changes and random fluctuations, as detailed below:

[0065] in, This represents the consistency index of axial migration signs, with values ​​ranging from negative to positive. The larger the absolute value, the more consistent the migration in the same direction. The number of consecutive blasting cycles involved in the judgment is determined by the number of historical state risk maps available in the risk memory. Indicates the first The first blasting cycle is relative to the first The change in axial center distance for each blasting cycle, the axial center distance is determined by the center position of the axial range field; The sign function is used to determine the direction of change. This indicator characterizes whether the migration continues in the same direction by taking the sign of continuous changes in different directions and averaging the results. When the signs continuously approach the working face, the signs remain consistent in the same direction, thus... The absolute value increases; the sign alternates when the change is repeated. Approaching zero. When the server outputs the evolution trend field, it will... Constrained together with the amplitude threshold, only when After repeatedly exceeding the amplitude threshold and A clear trend is output only when a preset consistency threshold is reached, and the consistency threshold, magnitude threshold, and corresponding historical cycle interval are written into the constraint evidence features, so that the evolution trend field can be used for engineering decision-making and can be audited and reproduced on the evidence chain.

[0066] S170. Associate the status risk map with the preset disposal strategy library to output the corresponding advanced geological forecast disposal strategy.

[0067] Specifically, the pre-configured disposal strategy library refers to a set of disposal strategies that are pre-configured and sustainably maintained; strategy semantic solidification refers to establishing a unified naming and semantic field structure for disposal strategies and fixing the meaning of each field; the disposal strategy identifier refers to the unique identifier of each disposal strategy; the strategy applicability condition set refers to the set of condition fields that allow the disposal strategy to be triggered, and these condition fields must at least include the winning state mechanism model identifier matching condition, axial range threshold condition, circumferential offset coverage sector condition, evolution trend type condition, and credibility threshold condition; the strategy action set refers to the set of engineering action fields that the disposal strategy requires to be executed, and these action fields must at least include the action identifier, action parameters, and action execution window for the main disposal action and the verification disposal action; the strategy constraint set refers to the set of construction condition fields that must be met when the disposal strategy is executed, and these construction condition fields must at least include construction method constraints, equipment resource constraints, traffic organization constraints, and safety boundary constraints; the strategy verification condition set refers to the set of condition fields used for closed-loop verification, and these condition fields must at least include retest trigger conditions and monitoring trigger conditions. After solidification, the disposal strategy identifier is indexed and associated with the above field sets, enabling subsequent retrieval processes to match each field and record the reasons for elimination.

[0068] When constructing a strategy query key based on the state risk map and binding it to the working condition context, the server extracts the winning state mechanism model identifier, axial range field, circumferential offset field, and evolution trend field from the state risk map, and combines the above extraction results into the strategy query key, making the strategy query key the sole input for retrieving and filtering disposal strategies. The strategy query key refers to the set of retrieval key values ​​formed by the fieldization of risk identification results; the axial range field expresses the axial distance range of the risk body's influence in front of the tunnel face; the circumferential offset field expresses the center orientation and coverage sector of the risk body's influence in the circumferential direction; and the evolution trend field expresses the migration direction and stability of the risk body in adjacent blasting cycles. When binding the strategy query key to the working condition context, the tunnel face mileage identifier, blasting cycle identifier, and construction equipment operating status are also bound simultaneously. The tunnel face mileage identifier provides absolute mileage reference, the blasting cycle identifier provides cross-cycle comparison reference, and the construction equipment operating status constrains the executability and reliability of the disposal strategy under the current equipment rhythm and resource status.

[0069] Multi-stage screening refers to the process of sequentially performing field matching and consistency verification through multiple screening stages; the candidate disposal strategy set refers to the set of disposal strategies that still meet the policy applicability conditions after passing the screening; model matching screening refers to retaining similar disposal strategies based on the matching condition of the winning state mechanism model identifier; distance interval consistency screening refers to retaining disposal strategies that can cover the current axial range field based on the axial range threshold condition; azimuth coverage consistency screening refers to retaining disposal strategies that can cover the current circumferential offset field based on the circumferential offset coverage sector condition; trend type consistency screening refers to retaining disposal strategies that are consistent with the current evolution trend field based on the evolution trend type condition. Each level of screening is recorded in the screening record field, which is used to record the disposal strategy identifier and the condition field identifier that triggers failure, so that the specific reason for the elimination of each disposal strategy can be traced later.

[0070] The executability constraint check refers to performing a consistency check on each item of the set of strategy constraints corresponding to the candidate disposal strategies based on the working context. Construction method refers to the combined state of the current excavation method, support method, and work organization method. Equipment resources refer to the availability of drilling rigs, grouting equipment, drainage equipment, surveying equipment, and work team resources. Traffic organization refers to the organizational restrictions on transportation and ventilation, as well as work face conflict constraints. Safety boundary refers to the boundary constraints of the support closure state, monitoring alarm thresholds, and risk control red lines. When a conflict is detected, the candidate disposal strategy is marked as an unexecutable strategy and written into the conflict reason field, which records the conflict type and conflict object. Strategy downgrading refers to replacing the unexecutable strategy with an alternative disposal strategy that has lower resource requirements, less intrusiveness, or less construction disturbance, while maintaining the matching conditions of the winning state mechanism model identifier. Simultaneously, the strategy verification condition set is adjusted to ensure that verification is still feasible, thereby obtaining an alternative disposal strategy that meets the current construction conditions.

[0071] After completing the executability constraint verification, when performing strategy optimization and strategy assembly to form an advanced geological prediction and treatment strategy for the remaining candidate treatment strategies, the server first determines the risk priority and then outputs the directly executable treatment strategy results. Strategy optimization refers to selecting the treatment strategy with the highest priority that best matches the current state risk map from the remaining candidate treatment strategies; risk priority refers to the result of sorting the urgency of risk and the necessity of treatment based on the axial range field, circumferential offset field, and evolution trend field; strategy assembly refers to combining the set of strategy actions corresponding to the optimized treatment strategy and the set of strategy verification conditions into a unified output, so that the output includes both the main treatment action and the verification treatment action. The main treatment action refers to the set of engineering actions that directly reduce risk, such as adjusting the excavation rhythm, adjusting support parameters, implementing advanced reinforcement, or implementing drainage and pressure reduction; the verification treatment action refers to the set of actions used to verify the effect of the main treatment action, such as retesting and monitoring densification. The purpose of the verification treatment action is to ensure that the next blasting cycle can form a traceable chain of evidence for the treatment effect.

[0072] Risk memory refers to persistent storage space used across blasting cycles to store state risk maps, response strategy outputs, filter record fields, conflict cause fields, and verification results. Write-back refers to writing the retest trigger conditions and monitoring trigger conditions from the strategy verification condition set into the operational context, enabling the next blasting cycle to perform targeted data collection and effect verification based on the retest trigger conditions during the baseline gather acquisition, azimuth acquisition, and discriminant feature combination extraction stages. Retest trigger conditions refer to the set of condition fields that trigger retesting after a response action is completed, including retest time, retest configuration, and retest target fields. Monitoring trigger conditions refer to the set of condition fields that trigger encrypted monitoring during and after the execution of a response action, including monitoring indicators, monitoring thresholds, and monitoring frequency fields. This allows the response strategy to not only output actions but also solidify verification requirements into the operational context of the next cycle.

[0073] This application also provides an advanced geological prediction device for use at tunnel faces, referring to... Figure 2 , Figure 2This application provides a schematic diagram of a module for an advanced geological prediction device applied to a tunnel face. The device is a server, which includes an acquisition module 21 and a processing module 22. The acquisition module 21 is used to determine the set of excitation points and the set of receiving points after the tunnel face has advanced to a preset mileage and completed the on-duty support closure, and to establish a working context bound to the set of excitation points and the set of receiving points. The processing module 22 is used to obtain a reference gather through the set of excitation points and the set of receiving points under the constraints of the working context, and to perform source consistency adjustment on the reference gather. The processing module 22 is used to form a baseline evidence package, which is used to characterize the basic response characteristics of the surrounding rock to the wavefield under the current working conditions. The processing module 22 is also used to construct a set of mutually exclusive state mechanism models in parallel based on the baseline evidence package, and to construct a set of discriminative features for the set of state mechanism models under a time window system. The time window system includes a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the working face, and a far-field time window for constraining the effect of the medium in front of the working face. The processing module 22 is also used to process the excitation... After the circumferential small-angle migration of the starting point set, an azimuth subset is obtained, and an azimuth evidence package is acquired through the azimuth subset. Simultaneously, based on the azimuth response pattern corresponding to the discriminant feature combination extracted from the azimuth evidence package under the same time window system, a first elimination process is performed on the state mechanism model set to obtain candidate state mechanism models. Processing module 22 is also used to perform a disproven experiment without changing the starting point set and the receiving point set if the number of candidate state mechanism models is determined to be greater than a preset number, so as to perform a second elimination process on the candidate state mechanism models to determine the winning state mechanism model. Processing module 22 is also used to map the winning state mechanism model into a state risk map. The state risk map uses axial range, circumferential offset, and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combination within the far-field time window, the circumferential offset is determined by the difference between the discriminant feature combination and different azimuth subsets, and the evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles. Processing module 22 is also used to associate the state risk map with a preset disposal strategy library to output the corresponding advanced geological prediction disposal strategy.

[0074] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0075] The communication bus 32 is used to enable communication between these components.

[0076] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0077] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0078] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0079] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for an advanced geological prediction method applied to the tunnel face.

[0080] exist Figure 3 In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call an application program stored in the memory 35 for an advanced geological prediction method applied to the tunnel face. When executed by one or more processors, the electronic device executes one or more methods as described in the above embodiments.

[0081] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0082] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for advanced geological prediction applied to tunnel faces, characterized in that, The method includes: After the tunnel face has advanced to the preset mileage and the shift support has been closed, the set of excitation points and the set of receiving points are determined, and a working context bound to the set of excitation points and the set of receiving points is established. Under the constraints of the operating condition context, a reference gather is obtained through the set of excitation points and the set of receiver points, and source consistency shaping is performed on the reference gather to form a reference evidence package. The reference evidence package is used to characterize the basic response characteristics of the surrounding rock to the wave field under the current operating condition. Based on the benchmark evidence package, a set of mutually exclusive state mechanism models is constructed in parallel, and a set of discriminative features is constructed for the set of state mechanism models under a time window system. The time window system includes a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the working face, and a far-field time window for constraining the effect of the medium in front of the working face. While keeping the set of receiving points unchanged, the set of excitation points is migrated circumferentially at a small angle to obtain a azimuth subset, and a azimuth evidence package is obtained through the azimuth subset. At the same time, based on the azimuth response pattern corresponding to the combination of discriminative features extracted from the azimuth evidence package under the same time window system, the set of state mechanism models is subjected to a first elimination process to obtain candidate state mechanism models. If it is determined that the number of candidate state mechanism models is greater than the preset number, then a proof-of-contrast experiment is performed without changing the set of excitation points and the set of receiving points, so as to perform a second elimination process on the candidate state mechanism models and determine the winning state mechanism model. The winning state mechanism model is mapped into a state risk map, which uses axial range, circumferential offset and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combination within the far field time window. The circumferential offset is determined by the difference between different directional subsets of the discriminant feature combination. The evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles. The state risk map is associated with a preset disposal strategy library to output corresponding advanced geological forecast disposal strategies.

2. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, Under the constraints of the operating condition context, a reference gather is obtained through the set of excitation points and the set of receiver points, and source consistency shaping is performed on the reference gather to form a reference evidence package, specifically including: An observation geometry locking process is performed on the set of excitation points and the set of receiving points to bind a unique identifier to each excitation point and each receiving point and solidify the corresponding spatial position parameters, orientation parameters, installation attitude parameters and coupling mode parameters. The unique identifier is then associated with the working context to ensure the consistency of the observation geometry between each measurement. After the observation geometry locking process is completed, the repeated excitation operation is performed on each excitation point in the set of excitation points according to the preset excitation configuration, and an excitation timestamp bound to the working condition context is generated each time the excitation is triggered. After each excitation trigger, the wave field response is synchronously acquired using the set of receiving points to form an original gather segment corresponding to the excitation timestamp, and time alignment processing is performed on each original gather segment based on the excitation timestamp to eliminate sampling start deviation between channels. The original gather fragments that have completed time alignment processing are subjected to quality constraint screening to remove invalid fragments that do not meet the channel consistency condition, saturation condition, abnormal spike condition or noise stability condition, and the integrity of the excitation point set in the excitation repetition number dimension is ensured by supplementary sampling. After the quality constraint screening process is completed, a source fingerprint is constructed for each excitation point based on the near-field time window, and fingerprint matching correction is performed on the source fingerprint to generate a set of shaping parameters. The original gather fragments are subjected to amplitude correction, phase correction and frequency band correction using the set of shaping parameters, and the corrected gather fragments are subjected to robust superposition processing according to the excitation point dimension to obtain a standardized reference gather. The standardized benchmark gather, along with the operating condition context, observation geometry locking results, the shaping parameter set, and the quality constraint screening records, are written into the benchmark evidence package.

3. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, The step of constructing a set of mutually exclusive state mechanism models in parallel based on the benchmark evidence package, and constructing a discriminative feature combination for the set of state mechanism models under a time window system, specifically includes: Based on the benchmark evidence package, a set of state mechanism models is constructed in parallel under the same evidence structure and the same time window system. The set of state mechanism models includes stress concentration model, low speed zone model and energy dissipation zone model, and is limited to the same spatial interval by mutual exclusion constraint table, and allows a single state mechanism model to be output as the winning state mechanism model. For the stress concentration model, phase-locking capability degradation and azimuth response bias enhancement are used as observable indicators, and the evidence-bearing interval is limited to the overlapping region of the mid-field and far-field time windows; for the low-speed band model, travel time curvature and increased low-frequency proportion are used as observable indicators, and the evidence-bearing interval is limited to the first part of the far-field time window; for the energy-dissipating band model, shortened energy lifetime and tail coherence degradation are used as observable indicators, and the evidence-bearing interval is limited to the second part of the far-field time window. A unified naming and semantic discriminative feature combination is constructed around the time window system, forming a near-field constraint feature vector, a mid-field discriminative feature vector, and a far-field discriminative feature vector, respectively. The near-field constraint feature vector is used to constrain the coupling effect of the support, the mid-field discriminative feature vector is used to describe the propagation stability of the medium adjacent to the tunnel face, and the far-field discriminative feature vector is used to describe the state changes of the medium in front of the tunnel face.

4. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, While keeping the set of receiving points unchanged, a directional subset is obtained by performing a small-angle circumferential migration on the set of excitation points, and a directional evidence packet is obtained through the directional subset. Simultaneously, based on the directional response pattern corresponding to the discriminative feature combination extracted from the directional evidence packet within the same time window system, a first elimination process is performed on the set of state mechanism models to obtain candidate state mechanism models, specifically including: Using the central axis of the tunnel face as a circumferential reference, each excitation point in the set of excitation points is migrated at a small circumferential angle under a preset angle constraint to generate a migration azimuth position. The corresponding migration angle parameters, migration direction parameters, hole depth parameters and coupling method parameters are then solidified into an azimuth geometry locking result so that the set of excitation points can form multiple azimuth subsets under different migration azimuths. After the azimuth geometry locking result takes effect, each of the azimuth subsets is controlled to perform azimuth acquisition to obtain azimuth gathers. The azimuth gathers are then subjected to source consistency shaping and time alignment processing consistent with the reference gathers to form standardized azimuth gathers. The standardized azimuth gathers, the azimuth geometry locking result, the receiver verification record, and the excitation configuration are then written into the azimuth evidence package. Based on the time window system consistent with the benchmark evidence package, the standardized azimuth gathers in the azimuth evidence package are sliced ​​using near-field time windows, mid-field time windows and far-field time windows. Within each time window, discriminative feature combinations are extracted according to unified naming and unified semantic rules to form azimuth feature vectors for each corresponding azimuth subset. By comparing the trend of the same azimuth feature vector changing with circumferential migration in different azimuth subsets, an azimuth response pattern is constructed, and the consistency of the azimuth response pattern with the operating state of the construction equipment in the working condition context is compared, so as to mark the perturbation of the azimuth response pattern affected by the rhythmic vibration of the construction equipment. Based on the azimuth response pattern, the mutual exclusion constraint table, and the mechanism discrimination entry table, the first elimination process is performed on the set of state mechanism models. If the azimuth response pattern of any state mechanism model is inconsistent with the preset azimuth response expectation, it is marked as an eliminated state mechanism model. If any state mechanism model passes the near-field constraint feature vector availability test and the azimuth response pattern is consistent with the azimuth response expectation, it is determined as a candidate state mechanism model.

5. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, If it is determined that the number of candidate state mechanism models is greater than a preset number, then a proof-of-contrast experiment is performed without changing the set of excitation points and the set of receiver points to perform a second elimination process on the candidate state mechanism models, in order to determine the winning state mechanism model, specifically including: If it is determined that the number of candidate state mechanism models is greater than the preset number, a counter-evidence excitation configuration in the form of complementary excitation pairs is constructed. The complementary excitation pair consists of a first complementary excitation and a second complementary excitation. The first complementary excitation and the second complementary excitation are consistent in energy budget, main frequency band and duration and satisfy a complementary relationship in code pattern phase structure. The complementary excitation pair is then bound to each excitation point identifier in the excitation point set. After the complementary excitation pair is solidified, the first complementary excitation and the second complementary excitation are triggered sequentially, and the first complementary gather fragment and the second complementary gather fragment are obtained through the set of receiving points. The first complementary gather fragment and the second complementary gather fragment are subjected to time alignment processing and source consistency shaping processing respectively to form a complementary gather pair. The complementary gather pair is bound with a complementary identifier and a disproving identifier to form a disproving evidence package. Under the unified time window system, the superposition response and differential residual are constructed based on the complementary gather pairs, and the superpositionability feature vector is extracted within the far-field time window. The superpositionability feature vector is used to characterize whether linear superpositionability holds and the spatial concentration and temporal stability of the violation. Based on the superposition feature vector and combined with the mechanism discrimination entry table and mutual exclusion constraint table, the candidate state mechanism model is subjected to a second elimination process. When the counter-evidence expectation of any candidate state mechanism model is inconsistent with the superposition feature vector, it is marked as an eliminated state mechanism model and written into the rejection evidence field. After the second elimination process is completed, the candidate state mechanism models that are not marked as eliminated state mechanism models are determined as winning state mechanism models.

6. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, The process of mapping the winning state mechanism model into a state risk map specifically includes: Based on the winning state mechanism model, the evidence carrying interval and evidence carrying channel set are determined. Under the unified channel index mapping and time index mapping constraints, the discrimination features are combined and reorganized into a mapping feature package within the evidence carrying interval. The mapping feature package includes main evidence features, veto evidence features and constraint evidence features. Under the constraints of the mapping feature package, by focusing on the main evidence features within the far-field time window and combining cross-channel consistency test with the consistency comparison of the counter-evidence package, the evidence concentration interval is determined, and the evidence concentration interval is mapped to the axial distance interval in front of the face of the tunnel to form the axial range field of the state risk map. Based on the difference distribution of the discriminant feature combination corresponding to different azimuth subsets within the same evidence carrying interval, the azimuth response difference is generated and after passing the direction consistency test and the construction equipment operation status disturbance verification, the center azimuth and coverage sector of the effective difference are determined to form the circumferential offset field of the state risk map. Based on the field alignment results of the state risk maps corresponding to adjacent blasting cycles under the same coordinate semantics, a continuity constraint judgment is performed on the migration of the axial range field and the circumferential offset field to generate the evolution trend field of the state risk map.

7. The method for advanced geological prediction applied to tunnel faces according to claim 1, characterized in that, The step of associating the state risk map with a preset response strategy library to output corresponding advanced geological forecasting and response strategies specifically includes: The preset handling strategy library is subjected to strategy semantic solidification processing, each handling strategy is bound to a unique handling strategy identifier, and the set of applicable conditions, set of action, set of constraint, and set of verification conditions associated with the handling strategy identifier are solidified, so that the handling strategies in the preset handling strategy library can establish a searchable relationship with the status risk map based on a unified field. Based on the state risk map, a strategy query key is constructed. The strategy query key includes the winning state mechanism model identifier, axial range field, circumferential offset field, and evolution trend field. The strategy query key is bound to the face mileage identifier, blasting cycle identifier, and construction equipment operating status in the working condition context as a unified input for handling strategy retrieval and filtering. Based on the strategy query key, a multi-stage screening process is performed in the preset disposal strategy library to form a candidate disposal strategy set. The multi-stage screening process includes model matching screening based on the winning state mechanism model identifier, distance interval consistency screening based on the axial range field, azimuth coverage consistency screening based on the circumferential offset field, and trend type consistency screening based on the evolution trend field. An executability constraint check is performed on the candidate disposal strategy set. The executability constraint check performs a consistency check on the strategy constraint set corresponding to the candidate disposal strategy based on the working context. When any candidate disposal strategy is found to conflict with the current construction method, equipment resources, traffic organization or safety boundary, the corresponding candidate disposal strategy is marked as an unexecutable strategy and strategy downgrading is triggered to obtain an alternative disposal strategy that meets the current construction conditions. After completing the executability constraint verification, the remaining candidate disposal strategies are optimized and assembled to form an advanced geological forecast disposal strategy that includes the main disposal action and the verification disposal action. The strategy optimization determines the risk priority based on the axial range field, circumferential offset field and evolution trend field of the state risk map. The strategy assembly combines the set of strategy actions corresponding to the optimized disposal strategy with the set of strategy verification conditions. The advanced geological forecasting and handling strategy, along with the corresponding status risk map, strategy query key, and working condition context, are written into the risk memory. The retest triggering conditions and monitoring triggering conditions in the strategy verification condition set are written back to the working condition context to support the verification of the handling effect and the continuous updating of the forecast closed loop in subsequent blasting cycles.

8. An advanced geological prediction device applied to the tunnel face, characterized in that, The apparatus is used to execute the advanced geological prediction method applied to tunnel faces as described in any one of claims 1 to 7, the apparatus comprising an acquisition module and a processing module, wherein... The acquisition module is used to determine the set of excitation points and the set of receiving points after the tunnel face has advanced to a preset mileage and the support closure for the shift has been completed, and to establish a working context bound to the set of excitation points and the set of receiving points. The processing module is used to obtain a reference gather through the set of excitation points and the set of receiving points under the constraints of the working condition context, and to perform source consistency shaping processing on the reference gather to form a reference evidence package. The reference evidence package is used to characterize the basic response characteristics of the surrounding rock to the wave field under the current working condition. The processing module is also used to construct a set of mutually exclusive state mechanism models in parallel based on the benchmark evidence package, and to construct a combination of discriminative features for the set of state mechanism models under a time window system. The time window system includes a near-field time window for constraining the coupling effect of the support, a mid-field time window for constraining the effect of the medium adjacent to the working face, and a far-field time window for constraining the effect of the medium in front of the working face. The processing module is further configured to, while keeping the set of receiving points unchanged, perform a small-angle circumferential migration on the set of excitation points to obtain a azimuth subset, and obtain a azimuth evidence package through the azimuth subset. At the same time, based on the azimuth response pattern corresponding to the combination of discriminative features extracted from the azimuth evidence package under the same time window system, perform a first elimination process on the set of state mechanism models to obtain candidate state mechanism models. The processing module is further configured to perform a proof-of-contrast experiment without changing the set of excitation points and the set of receiving points if it is determined that the number of candidate state mechanism models is greater than a preset number, so as to perform a second elimination process on the candidate state mechanism models to determine the winning state mechanism model. The processing module is further configured to map the winning state mechanism model into a state risk map. The state risk map uses axial range, circumferential offset, and evolution trend as unified output fields. The axial range is determined by the concentrated interval of the discriminant feature combination within the far-field time window. The circumferential offset is determined by the difference between different directional subsets of the discriminant feature combination. The evolution trend is obtained by comparing the state risk maps corresponding to adjacent blasting cycles. The processing module is also used to associate the state risk map with a preset disposal strategy library to output the corresponding advanced geological forecast disposal strategy.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images