A natural park geological disaster hidden danger early identification and early warning system

Abstract

The application discloses a natural park geological disaster hidden danger early identification and early warning system and relates to the technical field of geological disaster monitoring and early warning, and comprises a medium self-adaptive inversion module; based on real-time seismic wave signals of a target natural park, a medium impedance matching algorithm and an adaptive neural network are combined to dynamically invert surface medium parameters to construct a single-point early warning model and to adaptively correct parameter thresholds of the single-point early warning model. The application can dynamically analyze the influence of complex factors such as vegetation, soil humidity and lithology on signal propagation by means of the medium impedance matching algorithm and the adaptive neural network to adaptively correct the parameter thresholds of the single-point early warning model, thereby fundamentally overcoming the technical bottleneck of inaccurate early warning parameters and insufficient environmental adaptability caused by the variability of medium structures in the traditional method and improving the accuracy and reliability of signal identification in complex geological environments such as natural parks.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically to an early identification and early warning system for potential geological disaster hazards in natural parks. Background Technology

[0002] With the intensification of global climate change and human activities, nature parks are facing increasing risks of geological disasters, such as landslides, mudslides, and collapses. These geological disasters not only threaten natural landscapes and biodiversity, but may also pose serious threats to the lives of tourists. Therefore, it is particularly important to establish an effective monitoring system and early warning mechanism.

[0003] For example, the geological hazard identification method based on seismic wave dynamics and kinematic characteristics proposed in Chinese Patent Publication No. CN120949299A can be applied to the monitoring and early warning of geological hazards caused by human or natural activities, such as flash floods, dam breaks, landslides, and collapses. It has broad prospects for engineering technology applications and scientific research.

[0004] In existing technologies, the surface medium structure of natural parks is complex and variable. Seismic wave signals are significantly affected by vegetation, soil moisture, lithology, etc. during propagation. Due to the lack of adaptive interpretation and real-time correction mechanisms for medium differences, problems such as inaccurate threshold parameters and insufficient environmental adaptability of single-point early warning models occur. Secondly, after confirming the triggering of geological disasters, only real-time alarms can be achieved, making it difficult to dynamically predict the disaster scale, expansion path, and energy evolution trend based on multi-point time-series seismic data. It is also difficult to support advanced judgment and graded early warning of disaster evolution. Therefore, an early identification and early warning system for geological disaster hazards in natural parks is proposed to solve the above-mentioned problems. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an early identification and warning system for geological disaster hazards in a natural park, comprising:

[0006] The medium adaptive inversion module, based on the real-time seismic wave signal of the target natural park, combines the medium impedance matching algorithm and adaptive neural network to dynamically invert the surface medium parameters to construct a single-point early warning model, and adaptively corrects the parameter thresholds of the single-point early warning model to obtain the corrected time-series seismic wave data, thereby realizing real-time identification of complex surface media and dynamic correction of early warning parameters, and improving the environmental adaptability of the model.

[0007] The temporal fusion and inversion module is used to fuse time-series seismic wave data after correction from multiple measurement points. It uses a time-series fusion spatiotemporal propagation inversion algorithm to accurately locate geological hazard sources and dynamically reconstruct their initial motion trajectories, thereby achieving collaborative analysis of multi-source data and improving the accuracy of hazard source location and motion trajectory reconstruction.

[0008] The scale and energy evolution simulation module, based on the location results and time-series seismic wave data, combines the evolution trend prediction model to dynamically simulate the disaster scale and energy evolution, analyze the trend of disaster source expansion and energy distribution changes over time, realize the dynamic prediction of disaster scale and energy evolution, and support the advanced judgment of the evolution process.

[0009] The path evolution prediction module, based on simulation results and topographic data of the target natural park, predicts the expansion path and impact range of geological disasters. It dynamically updates the evolution path in combination with time-series changes, realizing dynamic prediction of disaster paths and impact ranges, and supporting risk zoning and evacuation planning.

[0010] The intelligent early warning output module is used to integrate the evolution prediction results, generate hierarchical early warning information and visualize the evolution status, and realize the visualization management and decision support of the entire process of real-time early warning release and disaster dynamic monitoring of downstream risk areas.

[0011] Preferably, the medium adaptive inversion module includes a medium parameter real-time inversion unit and an adaptive threshold correction unit;

[0012] The real-time inversion unit of the medium parameters is used to perform real-time inversion of the surface medium parameters of the seismic wave signal of the target natural park using the medium impedance matching algorithm, to obtain medium structure parameters including surface vegetation, soil moisture and lithology, and to construct a single-point early warning model of the target natural park, thereby establishing a single-point early warning model that reflects the actual medium structure and providing a basis for subsequent adaptive correction.

[0013] The adaptive threshold correction unit dynamically analyzes the inverted medium structure parameters based on an adaptive neural network, adaptively corrects the parameter thresholds of energy ratio and frequency range covered in the single-point early warning model, adapts to medium changes, and outputs corrected time-series seismic wave data from multiple measuring points, thereby achieving real-time matching between early warning parameters and environmental changes and improving monitoring accuracy and stability.

[0014] Preferably, the execution steps of the real-time inversion unit for medium parameters include:

[0015] The system receives real-time three-component seismic wave signals from multiple measuring points in the target natural park, performs bandpass filtering and instrument response correction, acquires preprocessed time-series seismic data streams, removes environmental and instrument noise, and restores the true amplitude and phase information of the seismic signals.

[0016] Based on the medium impedance matching algorithm, the wave impedance and dispersion characteristics of the time-series seismic data stream are calculated point by point. The three-dimensional medium structure profile covering the surface vegetation cover index, soil moisture content and lithological stiffness parameters is obtained by inversion, which quantitatively reveals the medium inhomogeneity and its key physical properties in the depth range below the surface.

[0017] Based on the inverted three-dimensional medium structure profile, a single-point early warning model is dynamically constructed for each measuring point. The model embeds the medium-dependent energy ratio baseline value, frequency response function, and attenuation coefficient to form a medium-adaptive initial early warning parameter set. This ensures that the basic parameters of the early warning model are accurately matched with the actual geological conditions below the measuring point, thereby improving the accuracy of the initial threshold setting.

[0018] Preferably, the execution steps of the adaptive threshold correction unit include:

[0019] An adaptive neural network model is constructed by inputting medium structure parameters and historical disaster event data. The network is trained to learn the dynamic impact of medium changes on early warning parameters, thereby realizing the intelligent mapping of early warning parameters to medium changes and significantly improving the model's environmental adaptability.

[0020] Based on the trained neural network, the changes in the retrieved medium parameters are analyzed in real time, and the energy ratio threshold and frequency response range of the long and short time windows in the single-point early warning model are dynamically adjusted to achieve environmental adaptive correction of the parameter thresholds. This enables the single-point early warning threshold to follow the changes in the surface environment in real time, significantly reducing the false alarm and missed alarm rates caused by differences in the medium.

[0021] The output corrected time-series seismic wave data includes medium-corrected energy ratio sequences, dominant frequency trajectories, and power spectral density curves, providing a consistent data foundation for multi-point fusion. The output data eliminates parameter deviations caused by medium differences between measurement points, providing reliable input for subsequent high-precision fusion inversion.

[0022] Preferably, the time series fusion and inversion module includes a multi-source time series data fusion unit and a spatiotemporal propagation inversion unit;

[0023] The multi-source time-series data fusion unit is used to perform time alignment and spatial fusion on the time-series seismic wave data after adaptive correction of multiple measurement points, construct a unified fused time-series dataset, support subsequent disaster source location and path inversion, form a high-quality dataset with spatiotemporal consistency, and provide reliable input for accurate location and inversion.

[0024] The spatiotemporal propagation inversion unit employs a spatiotemporal propagation inversion algorithm to calculate the propagation path and time difference of seismic waves based on the fused time-series seismic wave data. This is used to locate geological hazard sources and reconstruct their initial trajectory, enabling three-dimensional dynamic positioning of hazard sources and millisecond-level reconstruction of their trajectories, thus supporting evolution analysis.

[0025] Preferably, the execution steps of the multi-source time-series data fusion unit include:

[0026] The time-series seismic wave data after correction of multiple measuring points are time-stamp aligned and the sampling rate is unified to eliminate the time asynchronous error caused by transmission delay and differences in acquisition equipment, and to achieve strict alignment of multi-source data on the time axis, laying a precise time-series foundation for subsequent fusion;

[0027] Based on the spatial location relationship of the measuring points, a spatiotemporal grid model is established, and the data of each measuring point is interpolated to a unified grid node to form a spatially continuous temporal seismic wavefield data volume. This constructs a spatially continuous wavefield data volume under physical constraints, thus solving the problem of spatial discontinuity of discrete measuring point data.

[0028] A weighted fusion algorithm is adopted, which combines the signal-to-noise ratio and media consistency index of each measurement point to generate a high-confidence fused time series dataset to support disaster source location and propagation path reconstruction. Through dynamic weighted fusion, noise and media anomaly interference are effectively suppressed, significantly improving data quality and inversion reliability.

[0029] Preferably, the execution steps of the spatiotemporal propagation inversion unit include:

[0030] Based on the fused time series dataset, the arrival time difference of seismic waves at each measuring point is extracted. Combined with the medium wave velocity model, the propagation path and relative time difference sequence of seismic waves in the three-dimensional medium are calculated to obtain a high-precision time difference sequence, which provides a reliable spatiotemporal constraint for wave propagation for subsequent source localization.

[0031] The temporal back projection algorithm is used to retrieve the location of seismic wave energy radiation source time by time, and to perform spatiotemporal dynamic positioning of geological hazard source and identification of initial rupture point. This achieves millisecond-level dynamic positioning of hazard source with meter-level accuracy and accurately locks the hazard initiation rupture point.

[0032] By combining the propagation path and time difference sequence, the initial motion trajectory of the disaster source is reconstructed, and the trajectory point set and motion direction vector are output to provide input for large-scale evolution simulation. A smooth three-dimensional motion trajectory and velocity sequence are generated to provide an accurate initial kinematic state for evolution simulation.

[0033] Preferably, the execution steps of the scaled energy evolution simulation module include:

[0034] Based on the location of the disaster source and the time-series seismic wave data, energy release rate curves and spectral evolution characteristics are extracted to construct a dynamic energy release model for disasters. This model can quantitatively describe the dynamics of energy release during a disaster and provide core physical drivers for scale prediction.

[0035] An evolution trend prediction algorithm is adopted, which combines the media attenuation characteristics and terrain coupling effect to simulate the volume change and energy distribution evolution of the disaster source scale over time. This enables an advanced quantitative prediction of the future volume growth trend of the disaster source, and outputs disaster scale evolution curves, energy spatial distribution maps and time-series evolution profiles to support path expansion prediction and graded early warning decision-making. The algorithm also provides a graphical representation of key parameters of disaster evolution, providing direct data support for the delineation of downstream risk zones and the determination of early warning levels.

[0036] Preferably, the execution steps of the path evolution prediction module include:

[0037] Based on the results of large-scale energy simulation and the topographic data of the target natural park, a fluid dynamics and particle flow coupled model is used to analyze the movement path and accumulation trend of the disaster source in complex terrain. The model outputs the spatial distribution of flow depth, velocity vector field and accumulation thickness, so as to realize the refined movement simulation of the disaster source in real terrain and improve the physical reliability of path prediction.

[0038] By combining the changes in the energy distribution of time-series seismic waves, the path prediction results are dynamically corrected, generating a disaster expansion probability map and a time-series impact range boundary sequence. Real-time monitoring data is used to continuously calibrate the prediction, improving the accuracy and timeliness of impact range judgment. The system also outputs disaster evolution path animation, impact zoning map and risk level distribution, providing spatial decision-making basis for early warning issuance and evacuation planning, providing intuitive and visual decision support products, and assisting in the formulation of accurate emergency evacuation and control plans.

[0039] Preferably, the execution steps of the intelligent early warning output module include:

[0040] Based on the combined path prediction, scale simulation and energy distribution results, a four-level early warning information system (red-orange-yellow-blue) is generated, which includes the expected arrival time, impact range and recommended response measures. This achieves the structuring of early warning information and the quantification of risk levels, thereby improving the accuracy of emergency response.

[0041] Construct a visualization interface for the evolution of geological disasters, integrating real-time monitoring data, predicted paths, risk zones and early warning information, supporting dynamic overlay of multiple layers, providing a panoramic dynamic view, and enhancing situational awareness and collaborative decision-making capabilities;

[0042] Through wireless networks and public early warning platforms, real-time warnings are issued to downstream risk areas, and evolution status maps and decision support reports are pushed to achieve full-process visualized management and emergency response coordination, ensuring timely information delivery and synchronized action.

[0043] This invention provides an early identification and warning system for potential geological hazards in natural parks. It has the following beneficial effects:

[0044] (I) This early identification and warning system for geological hazards in natural parks uses a medium impedance matching algorithm and an adaptive neural network to perform real-time surface medium parameter inversion on seismic wave signals. It can dynamically analyze the influence of complex factors such as vegetation, soil moisture, and lithology on signal propagation, and adaptively correct the parameter thresholds of the single-point warning model accordingly. This fundamentally overcomes the technical bottleneck of traditional methods, which suffer from inaccurate warning parameters and insufficient environmental adaptability due to the variable medium structure. It improves the accuracy and reliability of signal identification in complex geological environments such as natural parks.

[0045] (II) This early identification and warning system for geological hazards in natural parks constructs a three-dimensional seismic wavefield data volume that covers the target area and is spatiotemporally continuous by performing spatiotemporal alignment and spatial interpolation on the data after correction of multiple discrete measuring points. It introduces topographic and wave velocity constraints to ensure that the fusion results conform to physical laws, transforming scattered monitoring point information into coherent surface and even volume information, forming a unified dataset with high confidence, and providing consistent data support for high-precision disaster source location, trajectory reconstruction and evolution simulation.

[0046] (III) This early identification and warning system for geological hazards in natural parks adopts a time-series fusion spatiotemporal propagation inversion algorithm. By utilizing fused data and a high-precision three-dimensional wave velocity model, it can dynamically locate energy radiation sources and reconstruct the initial movement trajectory of disaster sources, thus achieving real-time and accurate characterization of disasters. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating the workflow of an early identification and warning system for potential geological hazards in a natural park, as described in this invention.

[0048] Figure 2 This is a data flow diagram of an early identification and warning system for potential geological hazards in a natural park, as described in this invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: an early identification and warning system for geological disaster hazards in natural parks, comprising:

[0051] The medium adaptive inversion module, based on the real-time seismic wave signal of the target natural park, combines the medium impedance matching algorithm and adaptive neural network to dynamically invert the surface medium parameters to construct a single-point early warning model, and adaptively corrects the parameter thresholds of the single-point early warning model to obtain the corrected time-series seismic wave data. This enables real-time identification of complex surface media and dynamic correction of early warning parameters, improving the model's environmental adaptability. The medium adaptive inversion module includes a real-time inversion unit for medium parameters and an adaptive threshold correction unit.

[0052] The real-time medium parameter inversion unit is used to perform real-time surface medium parameter inversion on the seismic wave signals of the target natural park using a medium impedance matching algorithm. This acquires medium structure parameters including surface vegetation, soil moisture, and lithology, and constructs a single-point early warning model for the target natural park. This model reflects the actual medium structure and provides a foundation for subsequent adaptive correction. It receives real-time three-component seismic wave signals from multiple measuring points in the target natural park, performs bandpass filtering and instrument response correction, acquires the preprocessed time-series seismic data stream, removes environmental and instrument noise, and restores the true amplitude and phase information of the seismic signal based on the medium... Impedance matching algorithm calculates wave impedance and dispersion characteristics point by point in time-series seismic data stream, and inverts to obtain a three-dimensional medium structure profile covering surface vegetation cover index, soil moisture content and lithological stiffness parameters. It quantitatively reveals the medium heterogeneity and its key physical properties in the depth range below the surface. Based on the inverted three-dimensional medium structure profile, a single-point early warning model for each measuring point is dynamically constructed. The model embeds the medium-dependent energy ratio baseline value, frequency response function and attenuation coefficient to form a medium-adaptive initial early warning parameter set, so that the basic parameters of the early warning model are accurately matched with the actual geological conditions below the measuring point, and the accuracy of the initial threshold setting is improved.

[0053] The specific work involves deploying multiple real-time transmission three-component broadband seismometers within the natural park's geological disaster monitoring network. These seismometers acquire raw seismic wave signals in the vertical, east-west, and north-south directions. The signals are first noise-suppressed using a 0.5–30Hz bandpass filter, and then the actual ground motion is recovered through instrument response correction. The pre-processed signals are used to generate a continuous time-series seismic data stream at a 100Hz sampling rate, stored as independent minute-level data files for each channel, ensuring the temporal integrity and frequency domain authenticity of the input data. Based on the pre-processed time-series seismic data stream, a medium impedance matching algorithm is used to calculate the wave impedance spectrum and phase velocity dispersion curve point-by-point. Through iterative inversion, a three-dimensional medium structure profile with a depth of 0–30m centered on the measuring point is obtained. Key parameters include: surface vegetation cover index, soil volumetric water content, lithological shear wave velocity, and Young's modulus. Based on the inversion results, a single-point early warning model for each measuring point is dynamically constructed, embedding a medium-adaptive mechanism within the model. The parameter set includes the energy ratio baseline value, frequency response function, and attenuation coefficient. The energy ratio baseline value for both long and short time windows is adjusted based on the surface wave velocity; the dominant frequency response function is set according to the soil resonance characteristics; and the signal attenuation coefficient is related to the water content. Using the inverted three-dimensional medium structure profile, medium-dependent parameters are initialized in the single-point early warning model for each measuring point. The energy ratio threshold is dynamically calibrated based on the surface wave velocity and vegetation index: the threshold is increased by 10%–20% in low-vegetation areas and decreased by 5%–15% in high-exposed rock areas. The frequency response range is adaptively constrained by the soil resonance frequency band and lithological stiffness; for example, the dominant frequency range for soft soil is set to 2–5Hz, and extended to 5–12Hz in bedrock areas. The attenuation coefficient is used to correct the signal attenuation model in real time based on the water content and lithological combination, ensuring that the energy calculation matches the environment. Finally, an adaptive initial early warning parameter set for each measuring point is generated, including the dynamic energy ratio threshold, frequency response boundary, and attenuation correction coefficient.

[0054] The adaptive threshold correction unit dynamically analyzes the inverted medium structure parameters based on an adaptive neural network. It adaptively corrects the energy ratio and frequency range parameter thresholds covered in the single-point early warning model to adapt to medium changes and output corrected time-series seismic wave data from multiple measuring points. This achieves real-time matching between early warning parameters and environmental changes, improving monitoring accuracy and stability. An adaptive neural network model is constructed, inputting medium structure parameters and historical disaster event data. The network is trained to learn the dynamic influence of medium changes on early warning parameters, achieving intelligent mapping of early warning parameters to medium changes and significantly improving the model's environmental adaptability. Based on the trained neural network, it analyzes the changes in inverted medium parameters in real time, dynamically adjusting the long and short time window energy ratio thresholds and frequency response ranges in the single-point early warning model. This achieves environmental adaptive correction of parameter thresholds, enabling single-point early warning thresholds to follow surface environment changes in real time, significantly reducing false alarms and missed alarms caused by medium differences. The corrected time-series seismic wave data is output, including medium-corrected energy ratio sequences, dominant frequency trajectories, and power spectral density curves, providing a consistent data foundation for multi-measuring point fusion. The output data eliminates parameter deviations caused by medium differences between measuring points, providing reliable input for subsequent high-precision fusion inversion.

[0055] The specific work involves constructing and training an adaptive neural network model to achieve environmental adaptive correction of early warning parameter thresholds. The input layer of this model consists of media structure parameters obtained through real-time inversion, specifically including: surface vegetation cover index, soil volumetric water content, surface shear wave velocity, and Young's modulus of lithology. Simultaneously, the input layer also incorporates historical geological disaster event data, corresponding to the measured values ​​of the aforementioned media parameters and the associated true values ​​of the early warning parameters at the time of the disaster. The network structure employs a fully connected feedforward network with three hidden layers, containing 64, 32, and 16 neurons respectively. The activation function used is ReLU. The training objective is to learn the nonlinear mapping relationship between the combination of media parameters and the ideal early warning parameters. The model uses the mean squared error loss function and employs the Adam optimizer for backpropagation. The learning rate is set to 0.001, the batch size to 32, and the training cycle is no less than 500 rounds until the model's prediction error on the validation set stabilizes. The trained neural network model is deployed in a real-time processing pipeline to perform dynamic correction of warning parameters. It receives the latest 3D media structure profile data in real time, extracts the vegetation cover index, water content, wave velocity, and modulus parameters of each measuring point at the current moment, and combines them into a feature vector, which is then input into the loaded neural network model. After forward propagation, the model outputs the optimal adjustment amount of the warning parameters for the current media environment. For the energy ratio threshold of the long and short time windows, the model outputs a proportional adjustment coefficient, which is multiplied by... The model dynamically calibrates the threshold by adjusting it upwards or downwards based on the previous moment or baseline threshold. For the frequency response range, the model outputs adjusted values ​​for the lower and upper limits of the dominant frequency. These values ​​are adaptively constrained within the original set range according to the soil type. For example, when the water content increases significantly, the effective frequency band is automatically narrowed to suppress high-frequency noise interference. The entire calibration process is completed within seconds, ensuring that the single-point early warning model responds instantly to changes in medium properties caused by rainfall, snowmelt, etc. After completing the dynamic parameter calibration, the new threshold and range are immediately applied to reprocess the continuous time-series seismic data stream of the current measuring point. First, the updated energy ratio threshold is used to recalculate the energy ratio sequence of long and short time windows, where the time window length is fixed at 5 seconds for the first window and 20 seconds for the second window, ensuring that the energy ratio calculation is consistent with the medium. The signal is matched with the attenuation characteristics. Secondly, based on the corrected frequency response range, time-frequency analysis is performed on the signal (using short-time Fourier transform, window length of 10 seconds, overlap rate of 50%) to extract the main frequency trajectory of the signal and retain only the power spectrum components within the effective frequency band. Finally, based on the attenuation coefficient associated with water content, the power spectral density curve of the signal is attenuated to compensate for the high-frequency energy loss caused by medium absorption, and a power spectral density curve reflecting the true source energy is obtained. All data after medium correction, including energy ratio sequence, main frequency trajectory and power spectral density curve, are reorganized into a continuous time-series data stream at a sampling rate of 100Hz, and the corresponding medium parameter version number and correction timestamp are marked and output to the data buffer.

[0056] The time series fusion and inversion module is used to fuse time series seismic wave data after correction from multiple measurement points. It uses a time series fusion spatiotemporal propagation inversion algorithm to accurately locate geological hazard sources and dynamically reconstruct their initial motion trajectories, thereby achieving collaborative analysis of multi-source data and improving the accuracy of hazard source location and motion trajectory reconstruction. The time series fusion and inversion module includes a multi-source time series data fusion unit and a spatiotemporal propagation inversion unit.

[0057] The multi-source time-series data fusion unit is used to perform time alignment and spatial fusion on the adaptively corrected time-series seismic wave data from multiple measuring points, constructing a unified fused time-series dataset to support subsequent disaster source location and path inversion. This results in a high-quality dataset with spatiotemporal consistency, providing reliable input for accurate location and inversion. It performs timestamp alignment and sampling rate unification on the corrected time-series seismic wave data from multiple measuring points, eliminating time asynchrony errors caused by transmission delays and differences in acquisition equipment, achieving strict alignment of multi-source data on the time axis, and laying a precise time-series foundation for subsequent fusion. Based on the spatial location relationship of the measuring points, a spatiotemporal grid model is established, interpolating the data from each measuring point to a unified grid node to form a spatially continuous time-series seismic wave field data volume. This constructs a spatially continuous, physically constrained wave field data volume, solving the problem of spatial discontinuity in discrete measuring point data. A weighted fusion algorithm is used, combining the signal-to-noise ratio and medium consistency index of each measuring point, to generate a high-confidence fused time-series dataset to support disaster source location and propagation path reconstruction. Dynamic weighted fusion effectively suppresses noise and medium anomaly interference, significantly improving data quality and inversion reliability.

[0058] The specific work involves: to achieve effective fusion of multi-source seismic monitoring data, spatiotemporal alignment of time-series data acquired by three-component seismometers distributed within the natural park and after medium adaptive correction. The data is stored in minute-level files with a uniform sampling rate of 100Hz. Due to differences in clock synchronization accuracy and data transmission network latency among the various measuring points, precise timestamp correction is performed. An absolute time reference based on GPS timing signals is used to compensate for millisecond-level time offsets in the data streams from each measuring point, ensuring strict time synchronization across all channels. Furthermore, digital resampling is performed to address any existing sampling clock drift. An anti-aliasing FIR filter was used to unify all data to a standard sampling rate of 100Hz, retaining the effective frequency band components of 0.5-30Hz. Based on the three-dimensional coordinates (latitude, longitude, and elevation) of the measuring points and the monitoring network topology, a regularized spatiotemporal grid model covering the target area was established. The grid plane resolution was set to 50-100 meters according to the average instrument spacing, and the vertical direction was fixed at the surface layer (0-30 meter medium profile). The time axis was consistent with the data acquisition sequence. The Kriging spatial interpolation algorithm was used to interpolate the corrected time-series seismic wave parameters (including energy ratio sequence, dominant frequency trajectory, and power spectral density curve) of each measuring point to each grid node. The process incorporates topographic slope factor and medium wave velocity field as constraints for spatial variability functions, ensuring that the spatial distribution of the wavefield conforms to the physical laws of seismic wave propagation under complex terrain conditions. This ultimately generates a time-continuous three-dimensional seismic wavefield data volume with grid nodes as units. Each node contains a complete time-series record of three-component waveform parameters, forming a standardized fusion data foundation that is spatially continuous and temporally synchronized. Based on the spatiotemporal grid data volume, a dynamic weighted fusion algorithm based on signal-to-noise ratio (SNR) and medium consistency is employed. For each grid node, the SNR index of its adjacent source measurement points is calculated (defined as the effective signal power of 1-30Hz and the low-frequency signal power of 0.5-1Hz). The ratio of signal-to-noise ratio to power and the medium consistency coefficient (obtained by comparing the local variation of shear wave velocity and water content parameters obtained by inversion at each measurement point) are used to dynamically allocate the fusion weights based on the product of the square of the signal-to-noise ratio and the medium consistency coefficient. For each time sampling point, the parameter values ​​interpolated from each measurement point to that node are weighted and averaged to generate the fusion parameter value of that grid node. At the same time, the fusion confidence index is calculated, which is determined by the number of measurement points participating in the fusion, the uniformity of the weight distribution, and the parameter variance. The final output is a three-dimensional fusion time series dataset containing a complete spatiotemporal grid node parameter sequence, fusion weight distribution, and confidence evaluation.

[0059] The spatiotemporal propagation inversion unit employs a spatiotemporal propagation inversion algorithm to calculate the propagation path and time difference of seismic waves based on fused time-series seismic wave data. This is used to locate geological hazard sources and reconstruct their initial trajectories, achieving three-dimensional dynamic positioning and millisecond-level trajectory reconstruction of the hazard source. This supports evolution analysis. Based on the fused time-series dataset, the arrival time difference of seismic waves at each measuring point is extracted. Combined with the medium wave velocity model, the propagation path and relative time difference sequence of seismic waves in the three-dimensional medium are calculated to obtain a high-precision time difference sequence, providing reliable spatiotemporal constraints for subsequent source positioning. A time-series back projection algorithm is used to invert the location of seismic wave energy radiation sources time-by-time, performing spatiotemporal dynamic positioning and initial rupture point identification of geological hazard sources. This achieves millisecond-level dynamic positioning of hazard sources with meter-level accuracy, accurately locking the hazard's initial rupture point. Combining the propagation path and time difference sequence, the initial trajectory of the hazard source is reconstructed, outputting a trajectory point set and motion direction vector, providing input for large-scale evolution simulation, and generating a smooth three-dimensional motion trajectory and velocity sequence, providing accurate initial kinematic states for evolution simulation.

[0060] The specific work involves: In practical operation, accurately extracting the initial arrival times of seismic waves recorded at each measuring point from the fused time-series dataset, locking the 1-15Hz frequency band as the effective frequency band for initial detection. Specifically, for the three-component fused waveform data of each grid node, an improved AIC (Akaike Information Criterion) picking algorithm is used to automatically detect the initial arrival times of P-wave and S-wave phases. Polarization analysis is then used to further confirm the dominant phase. To calculate the accurate propagation path and relative time difference, a pre-stored three-dimensional wave velocity model based on previous medium inversion is invoked. This model stores the spatial distribution of shear wave velocity within a depth range of 0-30 meters in the target area using a 50m × 50m × 5m (horizontal × horizontal × depth) grid resolution. For any two measuring points (i, j), based on their three-dimensional coordinates, the rapid travel method is used to calculate the theoretical minimum travel time path and travel time difference of the seismic wave from the hypothetical source point to these two measuring points. In practical processing, an iterative approach is adopted: assuming an initial plane wavefront, the three-dimensional wave velocity model... The theoretical time difference between all measuring points is calculated. By comparing the theoretical time difference with the actual time difference observed from the fused data, the estimation of wavefront shape and propagation path is continuously adjusted until the residuals of the two meet the preset convergence criteria. Finally, the geometric description of the path evolution and relative time difference sequence of seismic waves in the actual three-dimensional non-homogeneous medium between all measuring point pairs are output. After obtaining high-precision observation time difference and propagation path constraints, the temporal and spatial dynamic positioning of geological hazard sources is carried out using a time-series back projection algorithm. Operationally, the study area is discretized on the horizontal plane with a 20m × 20m grid, and the depth range is fixed from the surface to within 30m. For each possible potential source point (grid node), the following steps are performed at each time sampling point: the theoretical travel time from the potential source point to all effective measuring points is calculated according to the three-dimensional wave velocity model. Secondly, from the actual fused waveform data of each measuring point (usually using the vertical component or the waveform envelope with energy correction), a section with a length of 0 ohms centered on the theoretical arrival time is extracted.Five-second time-window data are used. All time-window data are aligned and coherently superimposed. The amplitude of the superimposed energy represents the probability that a potential source point will radiate seismic wave energy at that moment. Searching and superimposing calculations are performed on the entire spatial grid step by step (sliding time window). When a spatial location exhibits coherent superimposed energy significantly higher than the background noise level across multiple consecutive time steps, it is identified as an active energy radiation source. By tracking the initial position and evolution sequence of active energy radiation sources in the spatiotemporal domain, the initial rupture point of the disaster is identified, i.e., the location where energy is first concentratedly released. Millisecond-level spatiotemporal dynamic positioning of multiple radiation sources throughout the entire disaster event is achieved. The positioning results are output as a source point sequence containing timestamps, three-dimensional coordinates (longitude, latitude, and depth), and instantaneous radiation energy levels. Based on the spatiotemporal sequence of source points generated by dynamic positioning, the initial trajectory of the disaster source is further reconstructed. Specifically, the source points located in consecutive time periods are connected in chronological order to form the initial trajectory. The trajectory point set is processed using a velocity-constrained Kalman filter to smooth the trajectory, eliminating jitter caused by positioning errors and extracting the true motion trend. The filter's state variables include position and velocity. The covariance matrix of its process noise and observation noise is adaptively adjusted based on the confidence level of the positioning results (derived from the confidence level of fused data and the energy signal-to-noise ratio superimposed by back projection). The smoothed trajectory point set clearly depicts the movement path of the disaster source from the initial rupture point. Simultaneously, by calculating the spatial displacement vector between trajectory points at adjacent time points, the motion direction and instantaneous velocity sequence of the disaster source are obtained. The spatial displacement vector includes horizontal (eastward and northward) components as well as vertical (elevation) change components, characterizing the three-dimensional kinematic features of the disaster source in complex terrain. Finally, a structured trajectory file is output, recording the three-dimensional coordinates, corresponding absolute time, motion direction vector (azimuth and inclination), and instantaneous velocity scalar of each trajectory point in chronological order.

[0061] The scale and energy evolution simulation module, based on the location results and time-series seismic wave data, combines the evolution trend prediction model to dynamically simulate the disaster scale and energy evolution, analyze the trend of disaster source expansion and energy distribution changes over time, realize the dynamic prediction of disaster scale and energy evolution, and support the advanced judgment of the evolution process.

[0062] The path evolution prediction module, based on simulation results and topographic data of the target natural park, predicts the expansion path and impact range of geological disasters. It dynamically updates the evolution path in combination with time-series changes, realizing dynamic prediction of disaster paths and impact ranges, and supporting risk zoning and evacuation planning.

[0063] The intelligent early warning output module is used to integrate evolution prediction results, generate hierarchical early warning information and visualize the evolution status output, realize the visualization management and decision support of the entire process of real-time early warning release and disaster dynamic monitoring of downstream risk areas, realize the visualization and emergency linkage of the entire process from monitoring to early warning, and improve the efficiency of decision response.

[0064] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the execution steps of the scale energy evolution simulation module include: based on the location of the disaster source and the time-series seismic wave data, extracting the energy release rate curve and spectral evolution characteristics, constructing a disaster energy dynamic release model, which can quantitatively describe the energy release dynamics during the disaster process, providing core physical drivers for scale prediction, adopting an evolution trend prediction algorithm, combining the medium attenuation characteristics and terrain coupling effect, simulating the volume change and energy distribution evolution process of the disaster source scale as it expands over time, thereby achieving an advanced quantitative prediction of the future volume growth trend of the disaster source, and outputting the disaster scale evolution curve, energy spatial distribution map and time-series evolution profile, supporting path expansion prediction and graded early warning decision-making, and intuitively displaying the key parameters of disaster evolution in a graphical way, providing direct data support for the delineation of downstream risk areas and the determination of early warning levels;

[0065] The specific work involves: based on the spatiotemporal sequence of disaster sources and the corresponding time-series seismic wave data output by the spatiotemporal fusion inversion module, extracting dynamic features characterizing the disaster process; for each located active source point, calculating its instantaneous radiated energy at 0.1-second intervals, forming an energy release rate curve with time as the horizontal axis and energy release rate as the vertical axis; simultaneously, extracting a 0.5-second time window centered on the theoretical arrival time from the three-component fused waveform data corresponding to the source point, performing a Fast Fourier Transform to extract its centroid frequency, spectral width, and energy ratio of low-frequency (1-5Hz) to high-frequency (5-15Hz) components, forming a spectral evolution feature sequence reflecting the evolution of the source mechanism; using the extracted energy release rate curve and spectral features, constructing a dynamic disaster energy release model. This model recursively uses a time step of 0.1 seconds, and its state equation takes the energy release rate and spectral features of the current moment as input, predicting the energy state of the next moment through a transfer function containing a medium attenuation term, thus initially simulating the accumulation and dissipation process of energy in space and time; based on the dynamic disaster energy release model... Based on this, an evolution trend prediction algorithm combining physical constraints is adopted to simulate the spatiotemporal expansion of the disaster source scale. The algorithm input includes: the output of the energy dynamic release model, the three-dimensional medium parameter profile obtained from the previous inversion, and the DEM topographic data of the target area at a scale of 1:1000. The simulation process is advanced with a time step of 0.5 seconds. In each step, the effective radiation radius of energy in three-dimensional space is calculated based on the current energy release rate and medium attenuation characteristics (using the frequency-related quality factor Q(f) model, whose value is dynamically assigned in the range of 20-80 according to the inverted water content and lithology). The topographic coupling effect is introduced. Based on the topographic slope and curvature calculated by DEM, the potential transport direction and accumulation trend of material driven by energy are corrected by empirical formula. Through iterative calculation, the change of the equivalent volume of the disaster source over time is simulated, and the disaster scale evolution curve (time-equivalent volume) is generated. At the same time, the energy spatial distribution calculated at each time step is accumulated and gridded (the grid is consistent with the fused data volume), and a series of energy spatial distribution maps and time-series evolution profile maps along the main movement direction are output.

[0066] The expression for calculating the effective radiation radius of energy in three-dimensional space is as follows:

[0067] ;

[0068] In the formula: To simulate time step At any given moment, the effective radiation radius of energy in three-dimensional space is defined as the radius of the spherical space that the energy released by the disaster can affect within the current time step, after considering the absorption by the medium. In order to time step Energy release rate at any given moment; To simulate the time step; The density of the medium; The energy conversion efficiency coefficient represents the proportion of seismic wave energy that is converted into effective mechanical energy to drive the movement (displacement, fragmentation) of the disaster source. It is determined by calibration using historical data based on the type of disaster. The quality factor is frequency-dependent and dynamically assigned within the range of 20-80 based on the inverted water content and lithology. The higher the value, the weaker the medium absorbs seismic waves, the slower the energy attenuation, and the larger the effective radiation radius.

[0069] The expression for calculating the potential transport direction of matter is as follows:

[0070] ;

[0071] ;

[0072] In the formula: In order to time step At any given moment, the unit vector of the potential transport direction of matter after terrain coupling correction; The terrain control weighting coefficient determines the relative influence of terrain and energy on the direction of transport, especially for highly fluid disasters (such as debris flows). Approaching 1, terrain-dominated, for collapses or rigid slides, Smaller values ​​have a greater impact from energy direction. This is the terrain gradient vector, calculated from the input high-precision DEM (Digital Elevation Model); For terrain height, The coordinates are in the horizontal direction. The coordinates are in the horizontal direction; Indicates the terrain in Rate of change of direction (slope in the east-west direction); Indicates the terrain in Rate of change of direction (slope in the north-south direction); The energy-driven direction vector is defined as the horizontal direction with the maximum energy release rate at the current moment, or the direction from the initial rupture point to the simulated center of mass at the previous moment.

[0073] The expression for calculating the accumulation trend is as follows:

[0074] ;

[0075] In the formula: For in position and time step At a given location, the change in the thickness of material deposition (positive value) or erosion (negative value) after correction for topographic curvature; This is the theoretical accumulation thickness before considering topographic curvature, which is determined by the effective radiation radius. The volume distribution of the internal substances was calculated. The curvature influence coefficient controls the intensity of the influence of topographic curvature on deposition / erosion thickness. It is obtained by inversion calibration using historical disaster depositional fan morphology and topographic curvature data. For in position The terrain curvature at a given location is calculated from the input DEM; This indicates a convex terrain (such as a mountain ridge), where materials do not easily accumulate, leading to... Increase, manifested as erosion or thinning, This indicates that the terrain is concave (such as a valley), where materials tend to accumulate, leading to... Increase, manifested as accumulation and thickening. This indicates a flat or uniform slope, where the terrain does not affect the original deposition thickness.

[0076] After completing the evolution simulation, the results are standardized, packaged, and output to directly support path expansion prediction and graded early warning decisions. The core output results include: a disaster scale evolution curve, stored in CSV format, containing absolute timestamps, corresponding equivalent volumes, and volume change rates; a sequence of energy spatial distribution maps, stored in GeoTIFF format, with each file corresponding to a simulation time point, and pixel values ​​representing the cumulative relative energy intensity received at that location, with spatial reference consistent with the input terrain data; and a temporal evolution profile, which is a two-dimensional raster sequence extracted along the main motion direction of the simulation, stored as an image sequence, visually displaying the changes in the thickness and energy intensity of the disaster source on the profile over time. All output data are accompanied by complete metadata, recording the version of the medium parameters used, the version of the terrain data, the model parameters, and the start and end times of the simulation.

[0077] The execution steps of the path evolution prediction module include: based on the results of large-scale energy simulation and the topographic data of the target natural park, using a fluid dynamics and particle flow coupled model, analyzing the movement path and accumulation trend of the disaster source under complex terrain, outputting the spatial distribution of flow depth, velocity vector field and accumulation thickness, realizing a refined movement simulation of the disaster source in real terrain, improving the physical credibility of path prediction, combining the changes in energy distribution of time-series seismic waves, dynamically correcting the path prediction results, generating a disaster expansion probability map and a time-series impact range boundary sequence, continuously calibrating the prediction using real-time monitoring data, improving the accuracy and timeliness of impact range judgment, and outputting disaster evolution path animation, impact zoning map and risk level distribution, providing spatial decision-making basis for early warning issuance and evacuation planning, providing intuitive and visual decision support products, and assisting in the formulation of accurate emergency evacuation and control plans;

[0078] The specific work involves: obtaining the initial spatial location, equivalent volume, and energy distribution of the disaster source; initializing the fluid dynamics and particle flow coupled model for calculation; setting the model's physical parameters based on the previously obtained three-dimensional medium parameter profile, including the internal friction angle, viscosity coefficient, and pore water pressure ratio dynamically assigned according to lithology and water content; providing surface boundary conditions from 1:1000 scale DEM data; setting the horizontal grid resolution of the model's computational domain to 5 meters and the vertical layering to 1 meter; and automatically adjusting the grid according to CFL stability conditions. The integral time step is usually less than 0.01 seconds. After the calculation starts, the model takes the equivalent volume of the disaster source as the initial material source quantity. Under the drive of gravity, it simultaneously solves the governing equations describing the fluid motion in saturated / unsaturated porous media and the discrete particle dynamics equations describing the collision, friction and shear dilatation behavior of solid particles. The two-way coupling is achieved through the momentum exchange term between the fluid phase and the particle phase. The model simulates the entire process of the disaster source's movement, sorting, siltation and cessation in complex terrains such as ditches and slopes. It outputs the flow depth, velocity vector field and spatial distribution of the accumulation thickness for each calculation time step.

[0079] Flow depth The governing equations are expressed as follows:

[0080] ;

[0081] In the formula: For in position and time The source of the disaster is deep; For calculating time; For divergence operators; For in position and time The depth-average velocity vector at that location; For source / sink terms, representing the local rate of change of flow depth caused by various physical processes;

[0082] The governing equations for the velocity vector field are expressed as follows:

[0083] ;

[0084] In the formula: This is the convective acceleration term, which describes the contribution of the spatial variation of the velocity field itself to the rate of change over time, and is a manifestation of the nonlinearity of fluid motion. This is the gravitational acceleration vector, with a magnitude of 9.81 m / s², and its direction is vertically downward. The slope angle of the ground surface The sine value; The linear damping coefficient represents linear friction from viscous dissipation from the substrate or interior. It is the second friction coefficient, representing turbulent friction or particle collision friction that is proportional to the square of the flow velocity; This is the fluid-particle momentum exchange term;

[0085] The formula for calculating the packing thickness is as follows:

[0086] ;

[0087] In the formula: At the end of the simulation, at position The final thickness of the accumulated debris at the disaster source; To simulate the total duration; For in position and time The deposition rate is dynamically determined and calculated by the model during the calculation process based on the particle dynamics equation and local flow conditions, reflecting the process of particles changing from a moving state to a stationary state. To calculate the time step;

[0088] A dynamic assimilation and correction mechanism based on real-time seismic wave energy distribution data is established to overcome the uncertainties in the initial conditions and physical parameters of the model. While the coupled model performs continuous simulation, it receives time-series energy spatial distribution maps with 0.5-second time intervals from the large-scale energy evolution simulation module in parallel. At each new observation moment, the simulated energy intensity field (characterized by the product of flow depth and the square of flow velocity) and the observed energy field at that moment are extracted. An ensemble Kalman filter algorithm is used for data assimilation: using the model state set from the previous moment as the background field and the current observed energy field as a constraint, the increment is calculated and analyzed, and all state variables and key physical parameters of the model are corrected accordingly. This allows the model-simulated spatiotemporal pattern of energy release to rapidly approximate the observed facts. At each observation moment... The process is repeated continuously to achieve constant calibration between the model's predicted trajectory and real-time seismic monitoring evidence. The dynamically corrected coupled model outputs a high-confidence spatiotemporal sequence of disaster evolution. Based on this sequence, the probability of each grid cell being covered by the disaster source during the entire simulation period is calculated to generate the final disaster spread probability map. Next, the frontal boundary of the disaster source with a thickness greater than 0.1 meters at each key prediction moment is extracted to form a temporal influence range boundary sequence. Finally, the above results are automatically integrated, and a disaster evolution path animation is generated through a 3D rendering engine. A risk level distribution map (divided into four levels of risk: extremely high, high, medium, and low) is produced by combining the spread probability, maximum flow depth, and arrival time. All maps and data are integrated and displayed through the geographic information platform of the early warning output module.

[0089] The execution steps of the intelligent early warning output module include: integrating path prediction, scale simulation, and energy distribution results to generate four-level early warning information (red-orange-yellow-blue), including estimated arrival time, impact range, and suggested response measures. This achieves structured early warning information and quantified risk levels, improving the accuracy of emergency response. It also constructs a visual interface for the evolution of geological disasters, integrating real-time monitoring data, predicted paths, risk zones, and early warning information. It supports dynamic overlay of multiple layers, provides a panoramic dynamic view, enhances situational awareness and collaborative decision-making capabilities, and issues real-time early warnings to downstream risk areas through wireless networks and public early warning platforms. It also pushes evolutionary situation maps and decision support reports, achieving full-process visualized management and emergency linkage, ensuring timely information delivery and synchronized action.

[0090] The specific work involves: based on the time-series impact range boundary sequence and risk level distribution map output by the path evolution prediction module, and combined with the scale evolution curve provided by the scale energy evolution simulation module, generating standardized four-level early warning information (red-orange-yellow-blue). The early warning level criteria are based on quantitative thresholds: a red warning is defined as the expected arrival time of the disaster front less than 5 minutes; an orange warning is defined as the arrival time of 5-15 minutes; a yellow warning is defined as the arrival time of 15-30 minutes; and a blue warning is defined as the arrival time of more than 30 minutes. The impact range is encapsulated in vector area data, accurate to 10 meters, with time attribute labels. For each warning level, a pre-set response measure library is automatically matched to generate structured data. Recommended response measures for a red alert include: immediate evacuation of personnel within 500 meters downstream along pre-set evacuation routes; closure of key road nodes; activation of the emergency broadcast system; and encapsulation of all warning information, spatiotemporal range, response measures, and generation timestamps into XML format data packets conforming to the CAP standard. A WebGIS-based visualization interface for the evolution of geological hazards should be constructed as a unified monitoring, early warning, and decision-making dashboard. The interface adopts a layered architecture, with the base layer being 1:1000 scale DOM and DEM data, dynamically overlaid with multiple thematic layers: including a real-time seismic wave energy distribution heatmap refreshed every 0.5 seconds, and a data map updated every 10 seconds. The system displays predicted path vector lines and frontal boundaries, a four-level risk zoning areal layer rendered based on a risk level distribution map, and pop-up windows for currently active warning information at each level. The interface supports independent control and transparent overlay of multiple layers. A timeline control bar allows users to rewind or fast-forward the evolution process. Key parameters such as flow depth and velocity raster data are accessed in real time via WMS service and displayed using color gradient rendering. All visualization elements comply with the relevant symbols and color scheme specifications of the National Geographic Information Public Service Platform. Through 4G / 5G wireless networks, the generated standardized warning information packages and situation snapshots are automatically pushed to the public warning information release platform, the emergency management department's command system, and authorized social networks. The media portal adheres to a five-element standard for content release: disaster type, warning level, affected area, estimated time, and response recommendations. For red and orange warnings, the system simultaneously triggers mandatory releases via associated emergency broadcast terminals, LED displays, and mobile cellular broadcasts within specific areas. Simultaneously, it automatically generates a decision support report containing key frames of the evolving situation, risk analysis conclusions, and action recommendations, which is then pushed to relevant commanders in PDF format via the intranet. The entire release process follows the principles of automatic generation, manual verification, and one-click release, and records complete operation logs to ensure the timeliness and traceability of warning information, achieving closed-loop management of the entire process from monitoring and analysis to emergency response.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An early identification and warning system for potential geological hazards in a natural park, characterized in that, include: The medium adaptive inversion module, based on the real-time seismic wave signal of the target natural park, combines the medium impedance matching algorithm and the adaptive neural network to dynamically invert the surface medium parameters to construct a single-point early warning model, and adaptively corrects the parameter thresholds of the single-point early warning model to obtain the corrected time-series seismic wave data; The time-series fusion and inversion module is used to fuse time-series seismic wave data after correction from multiple measurement points. It uses a time-series fusion spatiotemporal propagation inversion algorithm to accurately locate geological hazard sources and dynamically reconstruct their initial trajectories. The scale and energy evolution simulation module, based on the location results and time-series seismic wave data, combined with the evolution trend prediction model, dynamically simulates the disaster scale and energy evolution, and analyzes the trend of disaster source expansion and energy distribution changes over time. The path evolution prediction module predicts the expansion path and impact range of geological disasters based on simulation results and topographic data of the target natural park, and dynamically updates the evolution path in combination with time-series changes. The intelligent early warning output module is used to integrate evolution prediction results and generate hierarchical early warning information and visualize the evolution status.

2. The early identification and warning system for geological disaster hazards in a natural park according to claim 1, characterized in that: The medium adaptive inversion module includes a real-time inversion unit for medium parameters and an adaptive threshold correction unit; The real-time inversion unit of the medium parameters is used to perform real-time inversion of the surface medium parameters of the seismic wave signal of the target natural park using the medium impedance matching algorithm, to obtain medium structure parameters including surface vegetation, soil moisture and lithology, and to construct a single-point early warning model of the target natural park. The adaptive threshold correction unit performs dynamic analysis of the inverted medium structure parameters based on an adaptive neural network, adaptively corrects the parameter thresholds in the single-point early warning model, adapts to medium changes, and outputs time-series seismic wave data corrected from multiple measuring points.

3. The early identification and warning system for geological disaster hazards in a natural park according to claim 2, characterized in that: The execution steps of the real-time inversion unit for medium parameters include: The system receives real-time three-component seismic wave signals from multiple measuring points in the target natural park, performs bandpass filtering and instrument response correction, and obtains the preprocessed time-series seismic data stream. Based on the medium impedance matching algorithm, the wave impedance and dispersion characteristics of the time-series seismic data stream are calculated point by point, and a three-dimensional medium structure profile covering the surface vegetation cover index, soil moisture content and lithological stiffness parameters is obtained by inversion. Based on the inverted three-dimensional medium structure profile, a single-point early warning model for each measuring point is dynamically constructed. The model embeds the medium-dependent energy ratio baseline value, frequency response function, and attenuation coefficient to form an initial early warning parameter set that is adaptive to the medium.

4. The early identification and warning system for geological disaster hazards in a natural park according to claim 2, characterized in that: The execution steps of the adaptive threshold correction unit include: An adaptive neural network model is constructed, and the input media structure parameters and historical disaster event data are used to train the network to learn the dynamic impact of media changes on early warning parameters; Based on the trained neural network, the changes in the inverted medium parameters are analyzed in real time, and the long and short time window energy ratio thresholds and frequency response ranges in the single-point early warning model are dynamically adjusted to achieve environmental adaptive correction of the parameter thresholds. Output corrected time-series seismic wave data, including medium-corrected energy ratio sequence, dominant frequency trajectory, and power spectral density curve.

5. The early identification and warning system for geological hazards in a natural park according to claim 2, characterized in that: The time series fusion and inversion module includes a multi-source time series data fusion unit and a spatiotemporal propagation inversion unit; The multi-source time-series data fusion unit is used to perform time alignment and spatial fusion on the time-series seismic wave data after adaptive correction from multiple measurement points to construct a fused time-series dataset. The spatiotemporal propagation inversion unit uses a spatiotemporal propagation inversion algorithm to calculate the propagation path and time difference of seismic waves based on the fused time-series seismic wave data, in order to locate the geological hazard source and reconstruct the initial movement trajectory of the hazard source.

6. The early identification and warning system for potential geological hazards in a natural park according to claim 5, characterized in that: The execution steps of the multi-source time-series data fusion unit include: Timestamp alignment and sampling rate unification are performed on time-series seismic wave data after correction at multiple measuring points to eliminate time asynchronous errors caused by transmission delay and differences in acquisition equipment. Based on the spatial relationship of the measuring points, a spatiotemporal grid model is established, and the data of each measuring point are interpolated to a unified grid node to form a spatially continuous temporal seismic wavefield data volume. A weighted fusion algorithm is used to generate a high-confidence fused time series dataset by combining the signal-to-noise ratio and medium consistency index of each measurement point.

7. The early identification and warning system for geological disaster hazards in a natural park according to claim 5, characterized in that: The execution steps of the spatiotemporal propagation inversion unit include: Based on the fused time series dataset, the arrival time difference of seismic waves at each measuring point is extracted, and combined with the medium wave velocity model, the propagation path and relative time difference sequence of seismic waves in the three-dimensional medium are calculated. The temporal back projection algorithm is used to retrieve the location of seismic wave energy radiation source time by time, and to perform spatiotemporal dynamic positioning of geological hazard source and identification of initial rupture point; By combining the propagation path and time difference sequence, the initial trajectory of the disaster source is reconstructed, and the trajectory point set and motion direction vector are output.

8. The early identification and warning system for geological disaster hazards in a natural park according to claim 5, characterized in that: The execution steps of the scaled energy evolution simulation module include: Based on the location of the disaster source and the time-series seismic wave data, energy release rate curves and spectral evolution characteristics are extracted to construct a dynamic disaster energy release model; An evolution trend prediction algorithm is used, which combines the media attenuation characteristics and the terrain coupling effect to simulate the volume change and energy distribution evolution of the disaster source scale over time, and then outputs the disaster scale evolution curve, energy spatial distribution map and time series evolution profile.

9. The early identification and warning system for geological disaster hazards in a natural park according to claim 8, characterized in that: The execution steps of the path evolution prediction module include: Based on the results of large-scale energy simulation and the topographic data of the target natural park, a fluid dynamics and particle flow coupled model is used to analyze the movement path and accumulation trend of the disaster source under complex terrain, and output the spatial distribution of flow depth, velocity vector field and accumulation thickness. By combining the changes in the energy distribution of time-series seismic waves, the path prediction results are dynamically corrected, a disaster expansion probability map and a time-series impact range boundary sequence are generated, and an animation of the disaster evolution path, an impact zoning map and a risk level distribution are output.

10. The early identification and warning system for geological disaster hazards in a natural park according to claim 9, characterized in that: The execution steps of the intelligent early warning output module include: Based on the combined path prediction, scale simulation and energy distribution results, a four-level early warning information system (red-orange-yellow-blue) is generated, which includes the expected arrival time, the scope of impact and recommended response measures. Construct a visual interface for the evolution of geological disasters, integrating real-time monitoring data, predicted paths, risk zoning, and early warning information; Real-time warnings are issued to downstream risk areas through wireless networks and public early warning platforms, and evolution status maps and decision support reports are pushed out.

Images