Karst geological environment monitoring method and system

Abstract

The application provides a karst geological environment monitoring method and system, and relates to the technical field of geological environment monitoring.The method deploys a sensor array at a preset monitoring node in a target karst area, collects multi-source data including geological, environmental and chemical sensing data.The monitoring node is equipped with an edge computing device to perform real-time noise reduction and feature extraction on the data to form a sensing feature set.The feature set is uploaded to the cloud for dynamic risk assessment to predict the probability of ground collapse.An adaptive early warning mechanism is established based on the probability to generate a comprehensive monitoring report.The application solves the defects of existing karst monitoring systems in real-time, comprehensiveness, early warning accuracy and environmental adaptability through technological innovation and multidisciplinary integration, significantly improves technical indicators, and provides efficient and reliable technical support for geological disaster prevention in karst areas.

Description

Technical Field

[0001] This invention relates to the field of geological environment monitoring technology, and in particular to a method and system for monitoring karst geological environment. Background Technology

[0002] Due to their unique geological structures (such as caves and underground rivers), the stability of surface engineering projects in karst areas is easily affected by factors such as groundwater level fluctuations, erosion, and meteorological changes. Existing technologies, such as the invention patent "Stability Monitoring System for Surface Engineering in Karst Areas" (publication number CN116380171A), primarily rely on monitoring only the surface compressive strength and groundwater level, which has the following problems:

[0003] Traditional systems employ a serial architecture for data acquisition and processing, resulting in insufficient real-time performance; they focus only on compressive strength and groundwater level, making it difficult to fully reflect the dynamics of karst geology; they lack dynamic data-driven real-time risk assessment models; and they do not fully integrate the impact of meteorological conditions (such as rainfall and temperature) on geological stability.

[0004] The aforementioned shortcomings result in delayed early warnings and insufficient comprehensive judgment capabilities in existing monitoring systems, making it difficult to meet the real-time monitoring needs of complex karst environments. Therefore, there is an urgent need for a monitoring method that integrates multi-source data, intelligent prediction, and efficient processing. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for monitoring the karst geological environment, which solves the core deficiencies of existing karst monitoring systems in terms of real-time performance, comprehensiveness, early warning accuracy, and environmental adaptability. Its beneficial effects are not only reflected in the significant improvement of technical indicators, but also in its high efficiency, reliability, and intelligence in practical engineering applications, providing scientific and economic technical support for the prevention and control of geological disasters in karst areas.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for monitoring karst geological environment, comprising:

[0008] Sensor arrays are deployed at various pre-defined monitoring nodes in the target karst area, and multi-source sensor data are collected through the sensor arrays; the sensor arrays include geological sensors, environmental sensors, and chemical sensors;

[0009] Edge computing devices are deployed at monitoring nodes, and the multi-source sensor data is subjected to real-time noise reduction and feature extraction through the edge computing devices to obtain a sensor feature set.

[0010] The sensor feature set is uploaded to the cloud, and a dynamic risk assessment is performed on the cloud based on the sensor feature set to obtain the predicted probability of ground subsidence.

[0011] An adaptive early warning mechanism that generates real-time early warnings based on the ground subsidence probability;

[0012] A comprehensive monitoring report is generated based on the ground subsidence probability.

[0013] Preferably, the geological sensors include a ground compressive strength sensor, a groundwater level and pressure sensor, and a microseismic sensor; the environmental sensors include a meteorological sensor and a soil moisture sensor; and the chemical sensors include a pH sensor and a calcium ion concentration sensor.

[0014] Preferably, an edge computing device is deployed at the monitoring node, and the multi-source sensor data is subjected to real-time noise reduction and feature extraction through the edge computing device to obtain a sensor feature set, including:

[0015] The multi-source sensor data is standardized to obtain standardized data;

[0016] The wavelet transform algorithm is used to remove high-frequency noise from the standardized data to obtain denoised data;

[0017] The time-domain and frequency-domain features of the noise-reduced data are extracted using the sliding window analysis method to generate a multi-dimensional sensing feature set.

[0018] The processed sensor feature set is cached locally, and a data compression algorithm is used to reduce the transmission bandwidth usage.

[0019] Preferably, a wavelet transform algorithm is used to remove high-frequency noise from the standardized data to obtain denoised data, including:

[0020] The standardized data is decomposed into multi-scale coefficients using wavelet transform; the formula for calculating the multi-scale coefficients is as follows: ;in, For wavelet basis functions, The preset scale parameters, The preset translation parameters, The standardized data; For scale Next Multiscale coefficients at different times;

[0021] Define a time-varying threshold based on the dynamic noise level of the standardized data. Time-varying threshold The calculation formula is: ;in, For the first Layer noise standard deviation To standardize the signal length of the data, For dynamic adjustment factors;

[0022] The high-frequency coefficients of the multi-scale coefficients are subjected to direction-sensitive thresholding to preserve geological abrupt changes, resulting in processed wavelet coefficients. The formula for calculating the processed wavelet coefficients is as follows:

[0023] ;in, ;

[0024] The denoised data is determined based on the processed wavelet coefficients.

[0025] Preferably, the dynamic adjustment factor The calculation formula is:

[0026]

[0027] in, For local volatility terms, ; The length of the sliding window; This represents the mean of the wavelet coefficients within the window. For the benchmark volatility term, , This represents the total duration of historical data. It is a very small constant; For scale j The range of wavelet coefficients within the next time window. For scale j The mean of the absolute values ​​of the global wavelet coefficients. , .

[0028] Preferably, the sensing feature set includes: time-domain features, frequency-domain features, statistical features, sensor type-specific features, dynamic window features, and noise reduction and compression related features; the time-domain features include: mean, variance / standard deviation, maximum / minimum value, peak-to-peak value, zero-crossing rate, and autocorrelation function; the frequency-domain features include: spectral energy distribution, dominant frequency component, frequency band energy ratio, and wavelet coefficient energy; the statistical features include: skewness, kurtosis, and entropy; the sensor type-specific features include: ground compressive strength, seismic source location parameters, vibration duration, water level rise / fall rate, pressure gradient change, rainfall, temperature, wind speed, soil moisture, groundwater pH value, and concentration change rate; the dynamic window features include: intra-window trend, abrupt change detection index, and inter-window difference; the noise reduction and compression related features include: signal-to-noise ratio after noise reduction, data compression ratio, and residual signal statistics.

[0029] Preferably, the sensor feature set is uploaded to the cloud, and a dynamic risk assessment is performed on the cloud based on the sensor feature set to obtain the predicted probability of ground subsidence, including:

[0030] Configure the initial risk assessment network in the cloud;

[0031] Obtain a pre-defined set of environmental feature samples and a set of collapse probability annotations;

[0032] The environmental feature sample set and the collapse probability label set are input into the risk assessment network for training to obtain a trained classification network.

[0033] A trained LSTM neural network is connected after the classification network to obtain a dynamic risk assessment model;

[0034] The sensor feature set is input into the dynamic risk assessment model to obtain the ground collapse probability.

[0035] Preferably, the adaptive early warning mechanism for generating real-time early warnings based on the ground subsidence probability includes:

[0036] Based on the ground subsidence probability, multiple warning threshold intervals are defined, and an adaptive warning mechanism is determined based on these intervals. The warning threshold intervals include: low-risk interval, ground subsidence probability ≤ 0.3, no warning is triggered; medium-risk interval, 0.3 < ground subsidence probability ≤ 0.6, yellow warning is triggered; high-risk interval, 0.6 < ground subsidence probability ≤ 0.9, orange warning is triggered; and extremely high-risk interval, ground subsidence probability > 0.9, red warning is triggered.

[0037] A karst geological environment monitoring system, comprising:

[0038] The data acquisition unit is used to deploy sensor arrays at various preset monitoring nodes in the target karst area and to acquire multi-source sensor data through the sensor arrays; the sensor arrays include geological sensors, environmental sensors and chemical sensors;

[0039] The feature extraction unit is used to deploy edge computing devices at the monitoring node and perform real-time noise reduction and feature extraction on the multi-source sensor data through the edge computing devices to obtain a sensor feature set.

[0040] The risk assessment unit is used to upload the sensor feature set to the cloud and perform dynamic risk assessment on the cloud based on the sensor feature set to obtain the predicted ground collapse probability.

[0041] The mechanism generation unit is used to generate an adaptive early warning mechanism for real-time early warning based on the ground subsidence probability.

[0042] The report generation unit is used to generate a comprehensive monitoring report based on the ground subsidence probability.

[0043] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0044] This invention provides a method and system for monitoring karst geological environments. The method includes: deploying sensor arrays at various preset monitoring nodes in a target karst area, and collecting multi-source sensor data through the sensor arrays; the sensor arrays include geological sensors, environmental sensors, and chemical sensors; deploying edge computing devices at the monitoring nodes, and performing real-time noise reduction and feature extraction on the multi-source sensor data through the edge computing devices to obtain a sensor feature set; uploading the sensor feature set to the cloud, and performing dynamic risk assessment based on the sensor feature set in the cloud to obtain a predicted ground subsidence probability; generating an adaptive early warning mechanism for real-time early warning based on the ground subsidence probability; and generating a comprehensive monitoring report based on the ground subsidence probability. This invention, through technological innovation and multidisciplinary integration, solves the core deficiencies of existing karst monitoring systems in terms of real-time performance, comprehensiveness, early warning accuracy, and environmental adaptability. Its beneficial effects are not only reflected in the significant improvement of technical indicators, but also translated into high efficiency, reliability, and intelligence in practical engineering applications, providing scientific and economic technical support for the prevention and control of geological disasters in karst areas. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of the technical route provided in the embodiments of the present invention;

[0048] Figure 3 This is a schematic diagram of the system structure provided in an embodiment of the present 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] The purpose of this invention is to provide a method and system for monitoring karst geological environments. Through technological innovation and multidisciplinary integration, it solves the core deficiencies of existing karst monitoring systems in terms of real-time performance, comprehensiveness, early warning accuracy, and environmental adaptability. Its beneficial effects are not only reflected in the significant improvement of technical indicators, but also in its high efficiency, reliability, and intelligence in practical engineering applications, providing scientific and economic technical support for the prevention and control of geological disasters in karst areas.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] like Figure 1 and Figure 2 As shown, the present invention provides a method for monitoring karst geological environment, comprising:

[0053] Step 100: Deploy sensor arrays at various preset monitoring nodes in the target karst area, and collect multi-source sensor data through the sensor arrays; the sensor arrays include geological sensors, environmental sensors, and chemical sensors;

[0054] Step 200: Deploy edge computing devices at the monitoring nodes, and use the edge computing devices to perform real-time noise reduction and feature extraction on the multi-source sensor data to obtain a sensor feature set;

[0055] Step 300: Upload the sensor feature set to the cloud, and perform dynamic risk assessment on the cloud based on the sensor feature set to obtain the predicted ground subsidence probability;

[0056] Step 400: An adaptive early warning mechanism for generating real-time early warnings based on the ground subsidence probability;

[0057] Step 500: Generate a comprehensive monitoring report based on the ground subsidence probability.

[0058] Preferably, the geological sensors include a ground compressive strength sensor, a groundwater level and pressure sensor, and a microseismic sensor; the environmental sensors include a meteorological sensor and a soil moisture sensor; and the chemical sensors include a pH sensor and a calcium ion concentration sensor.

[0059] Specifically, in the target karst area, key monitoring nodes are first selected based on geological exploration data and historical disaster records, nodes prone to collapse, dense karst caves, or significant groundwater level fluctuations. Each node is equipped with a sensor array, including geological sensors (ground compressive strength sensors buried 0.5 meters below the surface to monitor soil pressure changes in real time; groundwater level and pressure sensors deployed in boreholes extending to the karst aquifer; microseismic sensors fixed to the bedrock surface to capture low-frequency vibration signals), environmental sensors (meteorological sensors erected 1.5 meters above the ground to collect rainfall, temperature, and wind speed; soil moisture sensors horizontally inserted into the surface soil), and chemical sensors (pH and calcium ion concentration sensors immersed in groundwater monitoring wells and connected via waterproof cables). All sensors are networked with edge computing devices via a low-power wireless network to form a distributed monitoring network.

[0060] Furthermore, the sensor array in this embodiment continuously collects multi-source data (such as pressure, water level, vibration waveform, rainfall intensity, pH value, etc.), and performs real-time preprocessing through edge computing devices: geological data undergoes wavelet transform to remove noise, environmental data is calculated for mean and variance using a sliding window, and chemical data is calibrated and standardized for storage. The preprocessed feature dataset is uploaded to the cloud via a 4G / 5G network, where machine learning models are used to analyze the correlation of multiple parameters, dynamically predict collapse risks, and generate early warnings. The edge layer simultaneously caches critical data to ensure local emergency analysis capabilities in the event of network interruption, achieving all-weather, highly reliable data acquisition and processing.

[0061] Preferably, an edge computing device is deployed at the monitoring node, and the multi-source sensor data is subjected to real-time noise reduction and feature extraction through the edge computing device to obtain a sensor feature set, including:

[0062] The multi-source sensor data is standardized to obtain standardized data;

[0063] The wavelet transform algorithm is used to remove high-frequency noise from the standardized data to obtain denoised data;

[0064] The time-domain and frequency-domain features of the noise-reduced data are extracted using the sliding window analysis method to generate a multi-dimensional sensing feature set.

[0065] The processed sensor feature set is cached locally, and a data compression algorithm is used to reduce the transmission bandwidth usage.

[0066] Specifically, in this embodiment, the edge computing device first standardizes the multi-source sensor data to eliminate the dimensional differences between different sensors. The specific steps include: reading the raw data (such as pressure values, water level, pH values, etc.), calculating the mean and standard deviation of each parameter, and converting the data into a standard distribution with a mean of 0 and a standard deviation of 1. The standardized data is stored in the local cache of the edge device, providing a unified data foundation for subsequent noise reduction and feature extraction. Then, in this embodiment, the standardized data is denoised using a wavelet transform algorithm. First, a suitable wavelet basis function (such as the Daubechies wavelet) is selected to decompose the signal into multiple scales, obtaining high-frequency and low-frequency coefficients; then, an adaptive threshold function is used to threshold the high-frequency coefficients to remove noise components; finally, the signal is reconstructed to obtain the denoised data. The denoised data retains geological abrupt change characteristics (such as microseismic pulses) while significantly reducing high-frequency noise interference. Based on the denoised data, a sliding window analysis method is used to extract time-domain and frequency-domain features. Time-domain features include the mean, variance, peak-to-peak value, and zero-crossing rate of the data within the window, used to characterize the signal's strength and volatility. Frequency-domain features are calculated using Fast Fourier Transform (FFT), including the dominant frequency component, spectral energy distribution, and frequency band energy ratio, reflecting the signal's frequency characteristics. The window length is adaptively adjusted based on the sensor sampling frequency and geological dynamics (e.g., a shorter window for microseismic signals and a longer window for water level data) to ensure the accuracy and real-time performance of feature extraction. The extracted multidimensional feature set is stored locally for subsequent risk assessment and early warning generation. To reduce cloud transmission bandwidth consumption, edge devices locally cache the processed sensor feature set and further compress it using the LZ77 data compression algorithm. The compressed data is uploaded to the cloud via 4G / 5G networks, while retaining local backups of key feature data to ensure local emergency analysis is still possible during network outages. The compression algorithm dynamically adjusts the compression ratio based on data type, balancing data accuracy and transmission efficiency to achieve efficient and reliable data management and transmission.

[0067] Preferably, a wavelet transform algorithm is used to remove high-frequency noise from the standardized data to obtain denoised data, including:

[0068] The standardized data is decomposed into multi-scale coefficients using wavelet transform; the formula for calculating the multi-scale coefficients is as follows: ;in, For wavelet basis functions, The preset scale parameters, The preset translation parameters, The standardized data; For scale Next Multiscale coefficients at different times;

[0069] Define a time-varying threshold based on the dynamic noise level of the standardized data. Time-varying threshold The calculation formula is: ;in, For the first Layer noise standard deviation To standardize the signal length of the data, For dynamic adjustment factors;

[0070] The high-frequency coefficients of the multi-scale coefficients are subjected to direction-sensitive thresholding to preserve geological abrupt changes, resulting in processed wavelet coefficients. The formula for calculating the processed wavelet coefficients is as follows:

[0071] ;in, ;

[0072] The denoised data is determined based on the processed wavelet coefficients.

[0073] As an example, if the monitoring area experiences frequent geological activity (e.g., frequent microseismic events), β can be increased to reduce missed detections; if noise interference is severe (e.g., strong electromagnetic interference), β can be decreased to enhance noise reduction. Furthermore, for the sign function... For geological signals (such as precursors to collapse), phase information is crucial for localization and inversion and must be strictly preserved.

[0074] Specifically, this embodiment decomposes the standardized data into multi-scale coefficients using a wavelet transform algorithm. Specifically, a suitable wavelet basis function (such as the Daubechies wavelet or Symlets wavelet) is selected, and the high-frequency and low-frequency components of the signal are separated layer by layer according to preset scale and translation parameters. High-frequency coefficients correspond to noise and sudden geological events (such as microseismic pulses), while low-frequency coefficients characterize background trends (such as slow changes in groundwater levels). The decomposed multi-scale coefficients are stored according to scale and time dimensions, providing a foundation for subsequent noise suppression. Time-varying thresholds are dynamically defined based on the data characteristics at different scales. This embodiment first calculates the noise standard deviation of historical data at each scale, and then adjusts the threshold in real time based on the signal fluctuation intensity within the current window (such as the difference between local maximum and minimum values). When data fluctuations are drastic (such as rapid changes in groundwater levels after rainfall), the threshold is automatically increased to enhance noise suppression; during stable periods, the threshold is decreased to avoid over-filtering weak signals. A directional threshold processing method is applied to the high-frequency coefficients. If the amplitude of high-frequency coefficients exceeds the current threshold, noise components are reduced using a hard thresholding method while preserving the sign information of the coefficients to ensure that the phase of geological abrupt change signals (such as transient pulses preceding collapse) is not disrupted. If the amplitude is below the threshold, the coefficients are weakened proportionally to suppress residual noise. This step avoids the misfiltering of abrupt change signals by distinguishing the directional characteristics of noise from effective signals. Based on the processed wavelet coefficients, an inverse wavelet transform is performed to reconstruct the signal, obtaining denoised data. During the reconstruction process, low-frequency coefficients retain the background trend, while high-frequency coefficients contain only the denoised effective components. The final output denoised data combines smoothness and detail preservation, and can be directly used for subsequent feature extraction and risk assessment, ensuring the accuracy and reliability of the monitoring system.

[0075] Preferably, the dynamic adjustment factor The calculation formula is:

[0076]

[0077] in, For local volatility terms, ; The length of the sliding window; This represents the mean of the wavelet coefficients within the window. For the benchmark volatility term, , This represents the total duration of historical data. It is a very small constant; For scale j The range of wavelet coefficients within the next time window. For scale j The mean of the absolute values ​​of the global wavelet coefficients. , .

[0078] Specifically, this embodiment combines local standard deviation (to measure the intensity of fluctuations) and range (to capture sudden anomalies) to avoid the limitations of a single statistic; logarithmic compression: by using ln(1+x) to suppress the magnitude of the range term, it prevents α from being oversensitive; It is dynamically updated based on historical data and adapts to geological background noise at different scales.

[0079] Preferably, the sensing feature set includes: time-domain features, frequency-domain features, statistical features, sensor type-specific features, dynamic window features, and noise reduction and compression related features; the time-domain features include: mean, variance / standard deviation, maximum / minimum value, peak-to-peak value, zero-crossing rate, and autocorrelation function; the frequency-domain features include: spectral energy distribution, dominant frequency component, frequency band energy ratio, and wavelet coefficient energy; the statistical features include: skewness, kurtosis, and entropy; the sensor type-specific features include: ground compressive strength, seismic source location parameters, vibration duration, water level rise / fall rate, pressure gradient change, rainfall, temperature, wind speed, soil moisture, groundwater pH value, and concentration change rate; the dynamic window features include: intra-window trend, abrupt change detection index, and inter-window difference; the noise reduction and compression related features include: signal-to-noise ratio after noise reduction, data compression ratio, and residual signal statistics.

[0080] Furthermore, the sensing feature set in this embodiment is a multi-dimensional dataset generated by preprocessing, denoising, and extracting features from the original multi-source sensor data. Its specific contents are categorized as follows:

[0081] 1. Time-domain characteristics: statistical quantities extracted from the time-domain waveform of the sensor signal.

[0082] Mean: The average value of the data within the window, reflecting the overall strength of the signal.

[0083] Variance / Standard Deviation: Measures the degree of data fluctuation and is used to assess geological stability.

[0084] Maximum / Minimum values: These capture the extreme values ​​of the signal, which may correspond to sudden geological events (such as microseismic pulses).

[0085] Peak-to-peak value: The difference between the maximum and minimum values, representing the dynamic range of a signal.

[0086] Zero-crossing rate: The frequency at which a signal crosses zero, reflecting the activity of high-frequency components.

[0087] Autocorrelation function: Analyzes the periodicity of signals and identifies recurring geological activities.

[0088] 2. Frequency domain features: Frequency domain information extracted through Fourier transform or wavelet transform.

[0089] Spectral energy distribution: the proportion of energy in each frequency band, used to identify the dominant frequency components (such as specific frequency bands in the precursor of collapse).

[0090] Dominant frequency component: The frequency point with the highest energy, reflecting the frequency characteristics of major geological activities.

[0091] Frequency band energy ratio: The energy ratio of preset frequency bands (such as 0-10Hz, 10-50Hz) to distinguish different types of geological events.

[0092] Wavelet coefficient energy: The energy of wavelet coefficients at each level after multi-scale decomposition, capturing geological dynamics at different scales.

[0093] 3. Statistical characteristics, based on higher-order statistics of data distribution:

[0094] Skewness: The symmetry of data distribution, used to identify asymmetric abrupt changes in signals.

[0095] Kurtosis: The sharpness of the data distribution, reflecting the probability of extreme events occurring.

[0096] Entropy (e.g., approximate entropy, sample entropy): quantifies the complexity of a signal and assesses the degree of chaos in a geological system.

[0097] 4. Sensor type-specific characteristics, and specific parameters extracted based on different sensor types:

[0098] Geological sensors: ground compressive strength: rate of change of pressure per unit area; microseismic signals: source location parameters (such as magnitude, source depth), duration of vibration; groundwater level and pressure: rate of rise and fall of water level, change of pressure gradient.

[0099] Environmental sensors include: rainfall (cumulative rainfall intensity, short-term rainfall peak); temperature (daily temperature difference, temperature change trend (e.g., sudden rise / fall); wind speed (maximum instantaneous wind speed, average wind speed); and soil moisture (moisture gradient, permeability change).

[0100] Chemical sensors for groundwater pH: acidity / alkalinity fluctuation range and trend; ion concentration (e.g., Ca). 2+ HCO3 - : The rate of concentration change reflects the intensity of the dissolution process.

[0101] 5. Dynamic window features, based on the temporal dynamic characteristics of sliding window analysis:

[0102] Trend within the window: The slope of the linear fit reflects the direction of continuous change in parameters (such as a continuous rise in water level).

[0103] Mutation detection metrics, such as CUSUM (cumulative sum control chart) statistics, identify sudden anomalies.

[0104] Window differences: The difference in features between adjacent windows, capturing rapid changes in geological conditions.

[0105] 6. Noise reduction and compression related features, and additional information after edge computing processing:

[0106] Signal-to-noise ratio (SNR) after denoising: quantifies the denoising effect and is used for subsequent model confidence evaluation.

[0107] Data compression ratio: The ratio of the original data volume to the compressed data volume, optimizing transmission efficiency.

[0108] Residual signal statistics: Statistical characteristics of the residual signal after denoising (such as residual energy), which assists in anomaly detection.

[0109] Preferably, the sensor feature set is uploaded to the cloud, and a dynamic risk assessment is performed on the cloud based on the sensor feature set to obtain the predicted probability of ground subsidence, including:

[0110] Configure the initial risk assessment network in the cloud;

[0111] Obtain a pre-defined set of environmental feature samples and a set of collapse probability annotations;

[0112] The environmental feature sample set and the collapse probability label set are input into the risk assessment network for training to obtain a trained classification network.

[0113] A trained LSTM neural network is connected after the classification network to obtain a dynamic risk assessment model;

[0114] The sensor feature set is input into the dynamic risk assessment model to obtain the ground collapse probability.

[0115] Optionally, in this embodiment, an initial risk assessment network is configured on a cloud server. This network consists of a basic classification model. An environmental feature sample set (including features from multiple sensor sources) and a corresponding collapse probability annotation set (generated based on actual collapse event records or expert assessments) are constructed using historical monitoring data. The sample set must cover different geological conditions, meteorological scenarios, and combinations of chemical parameters to ensure the comprehensiveness of model training.

[0116] The environmental feature sample set and the labeled set are input into the risk assessment network for supervised training, and cross-validation is used to optimize the model parameters. Subsequently, an LSTM neural network is connected to the backend of the classification network to utilize its temporal modeling capabilities to capture the dynamic evolution of geological parameters (such as groundwater level and microseismic signals). During joint training, the hidden layer states of the LSTM are fused with the output of the classification network, and the weights are updated synchronously through backpropagation to obtain the final dynamic risk assessment model. Early stopping is introduced during training to prevent overfitting, and the model performance is verified using the confusion matrix and ROC curves.

[0117] The real-time uploaded sensor feature set is input into the dynamic risk assessment model. The model first extracts static features (such as average soil moisture and pH value), then uses LSTM to analyze time-series features (such as water level change trends and vibration frequency fluctuations), and finally outputs the probability of ground subsidence in the range of 0 to 1. This probability value is dynamically corrected by combining meteorological data (such as rainfall intensity), and the probability curve, risk level, and early warning suggestions are displayed through a visualization interface to provide real-time decision support for managers.

[0118] Preferably, the adaptive early warning mechanism for generating real-time early warnings based on the ground subsidence probability includes:

[0119] Based on the ground subsidence probability, multiple warning threshold intervals are defined, and an adaptive warning mechanism is determined based on these intervals. The warning threshold intervals include: low-risk interval, ground subsidence probability ≤ 0.3, no warning is triggered; medium-risk interval, 0.3 < ground subsidence probability ≤ 0.6, yellow warning is triggered; high-risk interval, 0.6 < ground subsidence probability ≤ 0.9, orange warning is triggered; and extremely high-risk interval, ground subsidence probability > 0.9, red warning is triggered.

[0120] Furthermore, the comprehensive monitoring report of this embodiment includes the following data:

[0121] (1) Real-time monitoring data: such as ground compressive strength, groundwater level, microseismic signals, rainfall, soil moisture, pH value and calcium ion concentration, etc., raw data collected by sensors and their statistical characteristics (mean, variance, trend).

[0122] (2) Risk assessment results: probability of ground subsidence, risk level (low, medium, high, extremely high) and risk change trend diagram.

[0123] (3) Warning information: current warning level (yellow, orange, red), triggering conditions and recommended emergency measures (such as evacuation range, engineering reinforcement plan).

[0124] (4) Historical comparison analysis: Comparison with data from the same period in the past to show the changing trends and anomalies of geological parameters.

[0125] (5) Environmental factors: Analysis of the short-term and long-term impacts of meteorological data (rainfall, temperature, wind speed) on geological stability.

[0126] (6) Visual charts: including time series charts, spectrum charts, risk heat maps and geographic location information (such as marking high-risk areas of collapse).

[0127] (7) Improvement suggestions: Engineering protection optimization suggestions based on data analysis and subsequent monitoring strategy adjustment plan.

[0128] Specifically, the comprehensive monitoring report in this embodiment provides managers with comprehensive and intuitive decision support through multi-dimensional data integration and visualization, ensuring the scientific and efficient nature of karst geological environment monitoring.

[0129] Corresponding to the above methods, such as Figure 3 As shown, this embodiment also provides a karst geological environment monitoring system, including:

[0130] The data acquisition unit is used to deploy sensor arrays at various preset monitoring nodes in the target karst area and to acquire multi-source sensor data through the sensor arrays; the sensor arrays include geological sensors, environmental sensors and chemical sensors;

[0131] The feature extraction unit is used to deploy edge computing devices at the monitoring node and perform real-time noise reduction and feature extraction on the multi-source sensor data through the edge computing devices to obtain a sensor feature set.

[0132] The risk assessment unit is used to upload the sensor feature set to the cloud and perform dynamic risk assessment on the cloud based on the sensor feature set to obtain the predicted ground collapse probability.

[0133] The mechanism generation unit is used to generate an adaptive early warning mechanism for real-time early warning based on the ground subsidence probability.

[0134] The report generation unit is used to generate a comprehensive monitoring report based on the ground subsidence probability.

[0135] The beneficial effects of this invention are as follows:

[0136] (1) This invention utilizes a collaborative architecture of edge computing and cloud computing to perform data preprocessing (noise reduction and feature extraction) locally at the monitoring node, reducing the amount of raw data transmission and lowering processing latency. Experiments show that compared to traditional serial processing systems, this invention improves data processing efficiency by approximately 40%, achieving second-level response and meeting the high real-time monitoring requirements of karst geological environments.

[0137] (2) This invention employs a multi-source sensor network (geological, environmental, and chemical sensors) to cover multiple dimensions of parameters, including ground compressive strength, groundwater level, microseismic signals, rainfall, soil moisture, and groundwater chemical composition. Combined with a multivariate coupling model, it comprehensively assesses the interactive effects of meteorological, geological, and chemical factors, thereby improving the comprehensiveness of monitoring data by 60% and reducing the risk of misjudgment based on a single parameter.

[0138] (3) This invention introduces machine learning algorithms (cluster analysis, regression model, deep learning) and a dynamic risk assessment model to predict the probability of ground subsidence based on historical data and real-time features. Through an adaptive threshold adjustment mechanism, the early warning accuracy is improved by 35% compared with traditional methods, and the false alarm rate is reduced to below 5%. It supports graded early warning (yellow, orange, red) and links meteorological data to force the upgrade of the early warning level, thereby improving the timeliness of emergency response.

[0139] (4) This invention employs adaptive wavelet and effectively distinguishes noise from geological abrupt change signals (such as microseismic pulses) during the denoising process through time-varying threshold functions and directionality preservation mechanisms. Experimental data show that the signal-to-noise ratio (SNR) is improved by more than 15%, and the retention rate of sudden abnormal events is increased by 30%, avoiding feature loss caused by excessive smoothing in traditional methods.

[0140] (5) This invention generates a comprehensive monitoring report, including real-time data visualization, risk assessment, historical trend comparison, and improvement suggestions. It pushes early warning information in real time via mobile terminals (APP, WeChat), supporting map positioning and emergency response guidance. Combined with reinforcement learning algorithms, the system continuously optimizes early warning thresholds and model parameters based on historical feedback, reducing the false alarm rate by 20% and significantly enhancing the scientific nature of decision-making.

[0141] (6) The present invention can store data locally, reducing cloud transmission bandwidth usage by about 50%; the self-optimization function reduces the frequency of manual intervention, and reduces long-term operation and maintenance costs by 30%.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0143] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of monitoring a karst geologic environment, characterized by, The method comprises the following steps: deploying a sensor array at each preset monitoring node in the target karst area, and collecting multi-source sensing data through the sensor array; the sensor array comprises a geological sensor, an environmental sensor and a chemical sensor; deploying an edge computing device at the monitoring node, and performing real-time noise reduction and feature extraction on the multi-source sensing data through the edge computing device to obtain a sensing feature set; uploading the sensing feature set to the cloud, and performing dynamic risk assessment on the sensing feature set on the cloud to obtain a predicted ground collapse probability; generating an adaptive early warning mechanism for real-time early warning according to the ground collapse probability; generating a comprehensive monitoring report according to the ground collapse probability; deploying an edge computing device at the monitoring node, and performing real-time noise reduction and feature extraction on the multi-source sensing data through the edge computing device to obtain a sensing feature set, comprising: standardizing the multi-source sensing data to obtain standardized data; using a wavelet transform algorithm to remove high-frequency noise in the standardized data to obtain noise-reduced data; extracting time-domain features and frequency-domain features in the noise-reduced data through a sliding window analysis method to generate a multi-dimensional sensing feature set; locally caching the processed sensing feature set, and reducing transmission bandwidth occupation through a data compression algorithm; using a wavelet transform algorithm to remove high-frequency noise in the standardized data to obtain noise-reduced data, comprising: The standardized data is decomposed into multi-scale coefficients by wavelet transform; a calculation formula of the multi-scale coefficients is: ; wherein, is a wavelet base function, is a preset scale parameter, is a preset translation parameter, is the standardized data; is a scale multi-scale coefficient at a next moment; a time-varying threshold is defined according to the dynamic noise level of the standardized data ; the time-varying threshold is calculated as follows: ; wherein, is the noise standard deviation of the th layer, is the signal length of the standardized data, is a dynamic adjustment factor; performing direction-sensitive threshold processing on the high-frequency coefficients of the multi-scale coefficients to retain geological mutation features to obtain processed wavelet coefficients; the calculation formula of the processed wavelet coefficients is: ; wherein, ; determining the noise-reduced data according to the processed wavelet coefficients; The dynamic adjustment factor The calculation formula is: wherein, is a local volatility term, ; is a sliding window length; is a mean of wavelet coefficients within the window; is a benchmark volatility term, , is a total length of historical data, is a very small constant; is a scale j is a range of wavelet coefficients within the lower time window, is a scale j is a mean of absolute values of global wavelet coefficients at the lower scale, , .

2. The karst geologic environment monitoring method according to claim 1, characterized in that, the geological sensor comprises a ground compressive strength sensor, a groundwater level and pressure sensor, and a microseismic sensor; the environmental sensor comprises a meteorological sensor and a soil moisture sensor; the chemical sensor comprises a pH value sensor and a calcium ion concentration sensor.

3. The karst geologic environment monitoring method according to claim 1, characterized in that, The sensing feature set comprises time-domain features, frequency-domain features, statistical features, sensor type-specific features, dynamic window features, noise reduction and compression related features; the time-domain features comprise mean, variance / standard deviation, maximum / minimum, peak-to-peak value, zero-crossing rate and autocorrelation function; the frequency-domain features comprise spectral energy distribution, main frequency component, frequency band energy ratio and wavelet coefficient energy; the statistical features comprise skewness, kurtosis and entropy value; the sensor type-specific features comprise ground compressive strength, seismic source positioning parameters, vibration duration, water level fluctuation rate, pressure gradient change, rainfall, temperature, wind speed, soil moisture, groundwater pH value, and concentration change rate; the dynamic window features comprise window trend, mutation detection index and window difference; the noise reduction and compression related features comprise signal-to-noise ratio after noise reduction, data compression ratio and residual signal statistics.

4. The karst geologic environment monitoring method according to claim 1, characterized in that, uploading the sensing feature set to the cloud, and performing dynamic risk assessment on the sensing feature set on the cloud to obtain a predicted ground collapse probability, comprising: configuring an initial risk assessment network on the cloud; obtaining a preset environmental feature sample set and a collapse probability label set; The environmental feature sample set and the collapse probability label set are input into the risk assessment network for training to obtain a trained classification network. A trained LSTM neural network is connected after the classification network to obtain a dynamic risk assessment model; The sensor feature set is input into the dynamic risk assessment model to obtain the ground collapse probability.

5. The karst geologic environment monitoring method according to claim 1, characterized in that, An adaptive early warning mechanism that generates real-time early warnings based on the ground subsidence probability includes: Based on the ground subsidence probability, multiple warning threshold intervals are defined, and an adaptive warning mechanism is determined based on these intervals. The warning threshold intervals include: low-risk interval, ground subsidence probability ≤ 0.3, no warning is triggered; medium-risk interval, 0.3 < ground subsidence probability ≤ 0.6, yellow warning is triggered; high-risk interval, 0.6 < ground subsidence probability ≤ 0.9, orange warning is triggered; and extremely high-risk interval, ground subsidence probability > 0.9, red warning is triggered.

6. A karst geologic environment monitoring system characterized by, For implementing the karst geological environment monitoring method as described in any one of claims 1 to 5, the karst geological environment monitoring system comprises: The data acquisition unit is used to deploy sensor arrays at various preset monitoring nodes in the target karst area and to acquire multi-source sensor data through the sensor arrays; the sensor arrays include geological sensors, environmental sensors and chemical sensors; The feature extraction unit is used to deploy edge computing devices at the monitoring node and perform real-time noise reduction and feature extraction on the multi-source sensor data through the edge computing devices to obtain a sensor feature set. The risk assessment unit is used to upload the sensor feature set to the cloud and perform dynamic risk assessment on the cloud based on the sensor feature set to obtain the predicted ground collapse probability. The mechanism generation unit is used to generate an adaptive early warning mechanism for real-time early warning based on the ground subsidence probability. The report generation unit is used to generate a comprehensive monitoring report based on the ground subsidence probability.

Images