Intelligent early warning system for shallow landslide

Abstract

The application discloses a kind of shallow landslide intelligent early warning system, belong to disaster early warning monitoring technical field.It includes parameter management module, for storing and updating three-dimensional parameter matrix;Data acquisition module, for collecting real-time monitoring data;Risk assessment module, for calculating various landslide risk indexes based on the three-dimensional parameter matrix and the real-time monitoring data;Data quality check and safety correction module is configured to perform physical consistency check, time series lag correlation analysis and spatial consistency check on the monitoring data;Early warning decision and state transition module is configured to determine the early warning level according to the modified comprehensive landslide early warning index, and drive the system to migrate between multiple preset working states;Closed-loop learning module is configured to iteratively optimize the three-dimensional parameter matrix and the weighted fusion and synergistic effect addition mechanism based on the comparison feedback of early warning results and actual geological events.The application solves the technical problem of early warning system self-learning evolution.

Description

Technical Field

[0001] This invention relates to the field of disaster early warning and monitoring technology, and in particular to an intelligent early warning system for shallow landslides. Background Technology

[0002] Landslides, a frequent geological disaster worldwide, are characterized by their suddenness, destructive power, and wide-ranging impact, seriously threatening people's lives and property and the stability of infrastructure. With the rapid development of the Internet of Things, sensor networks, and big data technologies, intelligent early warning systems based on multi-source monitoring data have become a key technological means for landslide disaster prevention and control. However, existing early warning technologies still face multiple technical bottlenecks in practical engineering applications, limiting their reliability and practicality.

[0003] Current mainstream landslide early warning technologies are mainly divided into two categories: one is the empirical discrimination method based on fixed thresholds, which triggers an early warning when a single parameter such as rainfall or displacement rate exceeds a preset threshold; the other is the risk prediction method based on physical models (such as the limit equilibrium method and numerical simulation) or data-driven models (such as machine learning). Although the above methods have achieved certain results in specific scenarios, they still have the following systemic defects: existing technologies generally use regionally unified thresholds or static statistical thresholds, failing to effectively integrate inherent site risk factors such as slope, lithology, historical landslide density, and vegetation cover, and also failing to consider dynamic environmental conditions such as seasonal changes and the cumulative effect of previous rainfall. The parameter settings of most early warning systems rely on human experience and cannot be autonomously optimized based on the comparison feedback between early warning results and actual disaster events; after the early warning is triggered, the system's monitoring frequency, data processing strategies, and other operational statuses fail to dynamically match the risk level; at the same time, the risk assessment process often ignores the nonlinear synergistic effects between multiple parameters, leading to distorted risk quantification.

[0004] In summary, existing landslide early warning technologies have significant shortcomings in terms of data reliability, site adaptability, system intelligence, and operation and maintenance friendliness, making it difficult to meet the engineering requirements for highly reliable and adaptable early warning in complex geological environments. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent early warning system for shallow landslides, aiming to solve the problems of existing technologies.

[0006] To achieve the above objectives, this application provides the following solution: This application provides an intelligent early warning system for shallow landslides, including a parameter management module for storing and updating a three-dimensional parameter matrix; a data acquisition module for acquiring real-time monitoring data; a risk assessment module for calculating various landslide risk indices based on the three-dimensional parameter matrix and the real-time monitoring data, and generating a comprehensive landslide early warning index through a weighted fusion and synergistic effect addition mechanism; a data quality verification and safety correction module configured to perform physical consistency verification, temporal lag correlation analysis, and spatial consistency verification on the monitoring data, generate data quality coefficients, and dynamically correct the comprehensive landslide early warning index by integrating site safety factors; an early warning decision and state transition module configured to determine the early warning level based on the corrected comprehensive landslide early warning index, and drive the system to transition between multiple preset working states, with each working state corresponding to differentiated monitoring strategies, parameter adjustment strategies, and information output strategies; and a closed-loop learning module configured to iteratively optimize the three-dimensional parameter matrix and the weighted fusion and synergistic effect addition mechanism based on the comparison feedback between the early warning results and actual geological events.

[0007] Through the above technical solutions, the beneficial effects of this invention are as follows: By employing a four-pronged mechanism of data verification, security correction, state transition, and closed-loop learning, the reliability and adaptability of landslide early warning are significantly improved. Multi-dimensional data quality verification suppresses false alarms at the source; dynamic fusion of site safety factors enables adaptive threshold regions, effectively reducing missed alarms; synergistic effects accurately capture the critical points of risk superposition; the state transition strategy intelligently allocates monitoring resources according to risk levels, optimizing operational efficiency; and closed-loop learning drives continuous parameter evolution, enhancing the system's long-term environmental adaptability. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention, and the embodiments in the accompanying drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the structure of a shallow landslide intelligent early warning system provided in an embodiment of this application; Figure 2 A flowchart illustrating the operation of a shallow landslide intelligent early warning system provided in this application embodiment; The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0011] The foregoing and other technical contents, features and effects of the present invention are described in conjunction with the appendix below. Figure 1-2 The detailed description of the embodiments will make this clear. All structural details mentioned in the following embodiments are based on the accompanying drawings.

[0012] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings.

[0013] In one exemplary embodiment, such as Figure 1 As shown, a shallow landslide intelligent early warning system is provided, the system comprising: The parameter management module is used to store and update the three-dimensional parameter matrix; The data acquisition module is used to collect real-time monitoring data; The risk assessment and fusion module is used to calculate various landslide risk indices based on the three-dimensional parameter matrix and the real-time monitoring data, and to generate a comprehensive landslide early warning index through weighted fusion and synergistic effect addition mechanism. The data quality verification and security correction module is configured to perform physical law consistency verification, temporal correlation analysis and spatial distribution rationality verification on the monitoring data, generate data quality assessment results, and dynamically correct the comprehensive landslide early warning index by integrating site safety assessment parameters. The early warning decision and state transition module is configured to determine the early warning level based on the modified comprehensive landslide early warning index and drive the system to transition between multiple preset working states. Each working state corresponds to a different monitoring strategy, parameter adjustment strategy and information output strategy. The closed-loop learning module is configured to iteratively optimize the three-dimensional parameter matrix and the weighted fusion and synergistic effect addition mechanism based on the comparison feedback between the early warning results and actual geological events.

[0014] In one specific embodiment, the parameter management module constructs and dynamically maintains a three-dimensional parameter matrix, which includes a basic parameter layer, a weighted parameter layer, and an algorithm parameter layer. The basic parameter layer of the three-dimensional parameter matrix includes soil tension threshold, runoff coefficient threshold, rainfall threshold, and interflow threshold; the weighted parameter layer includes the weighting coefficients and synergistic additive coefficients of each risk index; and the algorithm parameter layer includes attenuation coefficients, time windows, and sensitivity parameters. The three-dimensional parameter matrix table P is shown in Appendix Table 1.

[0015] Appendix 1: Three-dimensional parameter matrix table.

[0016] Example initial values ​​(based on a typical loam slope) are shown in Appendix 2. The system status control module sets the initial working state of the system to normal and loads the corresponding state configuration (such as monitoring frequency and enabled adjustment mechanism).

[0017]

[0018]

[0019] Appendix 2 Example initialization (values ​​based on typical loam slope).

[0020] The data acquisition module is used to collect rainfall data, multi-layer soil tension data, surface runoff data, 30cm interflow data, and soil volumetric water content in real time. Data sequence collected: Rainfall sequence P(t): (Rainfall intensity) and (Cumulative rainfall); Multi-layered soil tension sequence: Where i represents the sensor burial depth (e.g., 10cm, 30cm, 50cm); Surface runoff sequence: (t), and calculate the real-time runoff coefficient. ; 30cm depth interflow sequence (t); Pore water pressure (matrix suction) sequence at a depth of 30 cm: Soil volumetric water content sequence: (t).

[0021] The risk assessment and fusion module is used to calculate various landslide risk indices based on the three-dimensional parameter matrix and the real-time monitoring data, and to generate a comprehensive landslide early warning index through weighted fusion and synergistic effect addition mechanisms. The various landslide risk indices include: rainfall risk index. Soil stability risk index Runoff Risk Index Mid-flow risk index The synergistic effect enhancement mechanism nonlinearly amplifies the early warning index when multiple risk indices simultaneously exceed a preset threshold.

[0022] In one specific embodiment, the physical consistency verification includes: verifying rainfall and total runoff, verifying surface runoff and interflow, verifying surface and deep soil tension, verifying soil tension and soil volumetric water content, verifying runoff coefficient decomposition, and verifying rainfall-interflow hysteresis response.

[0023] The data quality verification and security correction module assesses the reliability of input data and, considering inherent site risks, corrects the early warning index. It performs data quality verification, including physical consistency verification (such as water balance and soil water potential gradient) and temporal lag correlation analysis (calculating rainfall-interflow lag time). Physical consistency verification includes spatial consistency checks (using a vertical moisture propagation model). Physical consistency verification involves real-time checks of the physical and logical relationships between parameters, including six core checks: Verification object: Rainfall (P) vs. Total runoff (P) Physical relationship (ideal): Tolerance range: / ≤ 0.95; Anomaly score: 0; Anomaly handling: Mass conservation check: Violation indicates sensor failure or severe data distortion, mark sensor failure, and start alternative algorithm.

[0024] Verification object: Surface runoff ( ) vs. the Flowing Water Physical relationship (ideal): Tolerance range: ≤ 0.7; Abnormal score: 0.5; Abnormal handling: Verification of soil drainage capacity. Interflow should not continuously exceed the allowable range of soil saturated hydraulic conductivity. Check for deep seepage and adjust the saturation zone assessment.

[0025] Verification object: Surface soil tension (ψ) vs. deep soil tension; Physical relationship (ideal): Tolerance range: Inverted pressure ≤ 10 kPa; Anomaly score: 0.5; Anomaly handling: Vertical gradient verification of soil-water potential. Under non-freezing conditions, gravitational potential leads to lower potential at deeper layers. Mark unsaturated flow anomalies and check sensors.

[0026] Verification object: Soil tension (ψ) vs. soil volumetric water content (θ); Physical relationship (ideal): conforms to the soil moisture characteristic curve specified on site; Tolerance range: deviation ≤15%; Anomaly score: 0.5; Anomaly handling: soil-water relationship verification. Used to identify sensor drift or structural changes in the soil such as cracking / shrinkage. Recalibrate the θ-ψ curve parameters.

[0027] Verification objects: runoff coefficient decomposition; physical relationships (ideal): Tolerance range: Anomaly score: 0.5; Anomaly handling: Runoff path rationality verification. The sum of surface and subsurface runoff coefficients should be basically matched with the total runoff coefficient; otherwise, it indicates an anomaly in the flow distribution measurement.

[0028] Verification object: Rainfall-interflow hysteresis response; Physical relationship (ideal): Tolerance range: typical lag of 30-90 minutes; Anomaly score: 0.5 / 1; Anomaly handling: hydrological process dynamics verification. A lag that is too short (<20 min) indicates that the soil is near saturation and the risk is high; a lag that is too long (>120 min) indicates that the monitoring system is slow to respond.

[0029] The time-series lag correlation analysis includes: calculating the cross-correlation function between the rainfall sequence and the interflow sequence at the maximum lag time to find the peak lag time. The time-series lag correlation analysis is performed by calculating the cross-correlation function between the rainfall sequence P(t) and the interflow sequence... At maximum lag time Find the peak lag time using the cross-correlation function within the function. :

[0030] Response pattern recognition: <20 minutes: Fast saturation response, LSI increases by 0.15; 20 minutes ≤ Normal response time: ≤ 120 minutes; > 120 minutes: Response delay, enable auxiliary diagnostics.

[0031] The spatial consistency verification includes: numerically simulating the moist front using the Richards equations and comparing the simulation results with the actual moist conditions. During spatial consistency verification, a simplified Richards equation is used to numerically simulate the propagation of the moist front, and the simulation results are compared with the actual moist conditions of the five layers of sensors. Severe inconsistencies will trigger "preferred flow anomaly" or "sensor failure" alarms.

[0032] Establish a vertical moisture propagation model and verify the consistency of data across layers: (This formula is used to simulate the propulsion of a moist front.)

[0033] Inconsistency handling: When The multi-sensor voting mechanism was initiated. Possible reasons include: priority flow, inaccurate parameters, and sensor malfunction.

[0034] After completing physical consistency verification, temporal lag correlation analysis, and spatial consistency verification, a score is assigned to each verification result. ∈{0, 0.5, 1} (Pass: 1 point; Partial pass: 0.5 points; Fail: 0 points).

[0035] The system generates a quantitative data quality coefficient by verifying data across three dimensions: physical laws, temporal correlation, and spatial distribution. This explicitly quantifies the impact of abnormal data caused by sensor malfunctions, data transmission anomalies, and environmental interference. When data quality falls below a preset threshold, the system automatically terminates the early warning process and triggers a structured manual review mechanism. This fundamentally avoids false alarms caused by "garbage in, garbage out," significantly improving the reliability of early warning decisions.

[0036] In one specific embodiment, dynamically correcting the comprehensive landslide early warning index includes: normalizing all verification results to generate the data quality coefficient; calculating the site safety factor based on inherent risk factors, including slope, historical landslide density, and vegetation coverage; and dynamically correcting the comprehensive landslide early warning index according to the data quality coefficient and the site safety factor.

[0037] Calculate the data quality coefficient All verification results are normalized to a data quality coefficient Q between 0.3 and 1.0. The final warning index is multiplied by Q. When Q is below 0.6, the system will refuse to issue an automatic warning and instead issue a "data quality alarm" requesting manual intervention. This fundamentally avoids the decision-making risk of "garbage in, garbage out".

[0038]

[0039] Variable description: : The score for the i-th verification item (pass: 1 point; partially pass: 0.5 points; fail: 0 points); : Total number of items verified. Decision threshold: If the value is ≥0.6, continue the early warning process; <0.6: Trigger manual review.

[0040] Determine the site safety factor Based on inherent risk factors such as slope, historical landslide density, and vegetation cover, calculations are performed. Slope factor: F=1.2 when slope > 25°; Historical landslide density: F=1.15 in high-density areas; Vegetation cover: F=1.1 in low-coverage areas; Overall safety factor: A dynamic correction mechanism for site safety factors is introduced, incorporating inherent geological risk factors such as slope characteristics, historical disaster distribution, and vegetation cover status into the calculation logic of the early warning index, enabling the early warning threshold to adapt to regional differences. This effectively addresses the technical shortcomings of traditional fixed threshold models, which are insufficiently sensitive in high-risk areas and overly sensitive in low-risk areas, achieving precise early warning tailored to each region.

[0041] Calculate the final early warning index :

[0042] Variable description: Final warning index; Data quality coefficient (0.3-1.0); Safety factor (based on inherent site risks); LSI is a preliminary comprehensive early warning index generated by fusing multiple risk indices. If... If the value falls below a set threshold (e.g., 0.6), automatic alerts will be paused and a data quality alarm will be issued.

[0043] In one specific embodiment, the early warning decision and state transition module includes: comparing the comprehensive landslide early warning index with a preset level threshold to determine the current early warning level, and driving the system to transition between multiple preset working states according to the early warning level; the multiple preset working states include normal state, attention state, alert state, emergency state, and post-disaster state.

[0044] according to Make early warning decisions and control the overall operational status of the system. Determine the early warning level and... The current warning level is determined by comparing it with a preset level threshold (e.g., ...). <0.4 green, ≥ 0.4 blue, 0.4 > ≥ 0.55 yellow ≥ 0.7 (Red); Warning information output: The warning decision and output module generates structured warning information and releases it through preset channels.

[0045] The system status is controlled according to the new warning level, triggering a status transition (e.g., "Normal" → "Attention"). Details of adaptive adjustments based on the new status are shown in the table below: monitoring frequency (e.g., from 10 minutes / time to 5 minutes / time); enabled adjustment mechanisms (e.g., enabling pre-condition adjustments for "Attention" status); and resource allocation calculations.

[0046]

[0047] Table 3. Resource Allocation Adjustment Table In one specific embodiment, the early warning decision and state transition module further includes: when the early warning level reaches the highest level (e.g., red), automatically triggering an external emergency response process through an emergency linkage interface.

[0048] The system synchronously pushes structured early warning information (including disaster location, risk basis, and recommended measures) to the local emergency management platform, geological disaster prevention center, and designated responsible personnel's terminals via standardized communication protocols (such as HTTPS / MQTT); it also activates on-site audible and visual alarm devices to trigger high-volume sirens and warning lights, and activates the emergency broadcast system to continuously broadcast evacuation instructions; simultaneously, it sends road control suggestions to the traffic management platform, pushes resource requests to the rescue dispatch system, and synchronizes the coordinates of dangerous areas to the electronic map service to generate dynamic electronic fences. The system has a built-in linkage status feedback mechanism; if no external confirmation response is received within the set time limit, it automatically escalates the notification method (such as adding a voice call) and fully records the linkage log for post-event traceability and optimization.

[0049] In one specific embodiment, the closed-loop learning module includes: comparing each early warning result with the actual geological event, and classifying the early warning event into true positive, false positive, false negative or true negative based on the comparison result; analyzing the source of early warning error based on the classification result, identifying the parameter subset that needs to be optimized, and using statistical learning methods to iteratively update the relevant parameters in the three-dimensional parameter matrix and the weighted fusion coefficients in the risk assessment module.

[0050] Once the landslide warning event has completely ended (i.e., the system confirms that the landslide body has stabilized or the risk has been eliminated), the closed-loop learning module automatically starts the full-process parameter optimization mechanism: First, the warning judgment result is accurately compared with the actual geological state confirmed by on-site verification, and strictly classified into four categories: True Positive (TP): Correct warning; False Positive (FP): False alarm; False Negative (FN): Missed alarm; True Negative (TN): Correct warning without warning.

[0051] If the event is classified as TP or TN, the system only archives the data for long-term statistical analysis. However, if FP or FN occurs, a deep optimization process is triggered to analyze the possible causes of false positives (FP) or false negatives (FN) and determine the subset of parameters that need optimization (usually from layer B or layer W). The system then retrieves the full-cycle data for this event (including rainfall time series, displacement monitoring curves, dynamic soil volumetric water content, previous condition parameters, and real-time adjustment records), and performs multi-dimensional root cause inference by combining the expert knowledge base and historical case database. For example, an FP event may be caused by an overreaction of the "previous condition adjustment coefficient" in layer B to short-term rainfall, or by a low setting of the "displacement mutation threshold" in layer W; an FN event may be caused by insufficient "real-time feedback gain" in layer W, or by the "soil saturation attenuation factor" in layer B not being adapted to local soil and rock characteristics.

[0052] Based on the attribution conclusions, the system intelligently selects a subset of parameters that need optimization (typically focusing on key parameters of the dynamic adjustment layer B or weight layer W in the risk calculation model) and constructs a Bayesian update framework for each parameter. For the selected parameter τ, the current value is used as the prior μ_prior, and a likelihood function is constructed using the observed value τ_obs of the current event and the outcome label L. The likelihood mean and variance are calculated, and then the posterior mean is precisely solved using the Bayesian formula. The posterior distribution is calculated according to the Bayesian formula, and the optimal estimate and confidence level (variance) of the parameter are updated.

[0053] The parameter iteration optimization process includes: data collection, collecting all observation data of the early warning event; result classification, determining whether the event belongs to TP / FP / FN / TN; cause analysis, analyzing possible causes of false alarms / missed alarms; parameter selection, determining the set of parameters that need to be adjusted; Bayesian update, performing the update calculation of the formula; effect verification, verifying the effect of the updated parameters; parameter confirmation, confirming the update or rollback.

[0054] In one specific embodiment, the closed-loop learning module further includes: periodically calculating the system's early warning performance indicators, and dynamically adjusting the learning strategy and optimization cycle based on the evaluation results.

[0055] A three-tiered learning evaluation system is constructed: Dynamic calculation of performance indicators: Based on the classification results of early warning events (TP / FP / FN / TN), periodic statistical reports on accuracy, precision, recall, and F1 score are generated to quantify the system's early warning effectiveness; Adaptive strategy adjustment: When any indicator continuously fluctuates beyond a preset threshold, parameter fluctuation analysis is automatically triggered, and the associated thresholds and weight coefficients in the three-dimensional parameter matrix are batch-calibrated, dynamically shortening the subsequent learning cycle. The learning cycle includes short-term (real-time updates after each early warning event), medium-term (monthly statistical evaluation), and long-term (annual comprehensive evaluation and recalibration). A statistical report on learning effectiveness is generated, and parameters with fluctuations exceeding the threshold are batch-calibrated. Learning effectiveness evaluation indicators include: ; Accuracy: ; Recall rate: ; F1 score: .

[0056] The confirmed optimized parameter values ​​are updated in the three-dimensional parameter matrix P, persistently stored by the parameter management module through a transactional database, and simultaneously pushed to all edge computing nodes and the cloud-based main control system to ensure network-wide parameter consistency. At this point, a single closed-loop learning cycle is successfully completed. The system, with more accurate perception capabilities and decision-making logic, enters the next monitoring phase, continuously implementing the intelligent operation and maintenance concept of practice feedback-driven model evolution, significantly improving the long-term adaptability and reliability of landslide risk early warning.

[0057] In one specific embodiment, the system constructs a complete closed-loop processing cycle of real-time perception, dynamic decision-making, intelligent execution, and continuous evolution. Based on the current operating state (e.g., normal, monitored, alert, emergency, or post-disaster), the system continuously cycles through five core stages using a monitoring frequency (dynamically adjustable from 10 minutes / time to 30 seconds / time): First, a multi-source sensor network collects raw data such as rainfall, surface displacement, pore water pressure, and soil volumetric water content at a state-driven frequency; second, a dynamic adjustment module adaptively activates corresponding strategies based on the system state, for example, in the "alert" state, encrypting sampling to 2 minutes / time and simultaneously enabling pre-condition adjustment and real-time feedback mechanisms; subsequently, a risk assessment and fusion module comprehensively calculates multi-dimensional indicators such as rainfall risk index, displacement risk index, and soil stability risk index. The system generates a comprehensive landslide risk index through a weighted fusion algorithm. Subsequently, the quality verification and safety correction module performs three-level verification on the fusion results (data consistency verification, physical logic verification, and historical trend comparison), smooths out abnormal fluctuations or reduces confidence weights to ensure robust and reliable output. Finally, the early warning decision and state transition module compares the comprehensive landslide risk early warning index with a preset threshold to determine the early warning level, generates structured early warning information containing location, risk level, and recommended measures, and releases it through multiple channels such as SMS, broadcast, and emergency platform. At the same time, it drives the system state transition (such as "normal" → "attention") and adjusts the monitoring frequency, computing resource allocation, and output mode accordingly.

[0058] In one specific embodiment, the mechanism for threshold determination includes: the geomechanical basis of soil tension threshold; the watershed hydrological basis of runoff threshold; and the statistical-mechanical dual basis of rainfall threshold.

[0059] The geomechanical basis of soil tension threshold is directly linked to the formula for the shear strength of unsaturated soil. By setting a critical safety factor slightly higher than 1.0 (e.g., 1.05), the critical matric suction (tension) at each potential sliding depth is derived. This gives the threshold a clear mechanical safety meaning. Based on the extended Mohr-Coulomb criterion, the critical matric suction of each layer is derived by setting a critical safety factor:

[0060] c' Effective cohesion (kPa), σ Total stress (kPa) Pore ​​gas pressure (kPa), Pore ​​water pressure (kPa) Effective internal friction angle (°), Internal friction angle (°) related to matrix suction. Angle of slope (°).

[0061] Physical meaning: The numerator represents the anti-slip force, the denominator represents the sliding force, and Fs>1 indicates stability; matrix attraction: ( (Soil Tension) Derivation of Critical Tension Threshold: By setting a critical safety factor Fs=1.05, the critical matrix suction of each layer is calculated. 10cm layer: =5 kPa (corresponding to FS≈1.05, most sensitive); 30cm layer: = 12 kPa (potential sliding surface location); 50cm layer: = 18 kPa (Deep stability control) The watershed hydrological basis for the runoff threshold combines the SCS-CN model and the Green-Ampt infiltration model. Through dynamically corrected curve numbers (CN values), the runoff generation capacity under different antecedent wetting conditions is theoretically calculated, providing a hydrological theoretical foundation for the runoff coefficient threshold, rather than just an empirical value. The critical runoff coefficient under different antecedent conditions is derived.

[0062] Dynamic CN value correction:

[0063] Runoff depth calculation:

[0064] Critical runoff coefficient:

[0065] Variable description: Potential maximum retention capacity; Dynamic CN value; Runoff depth; Critical runoff coefficient.

[0066] The theoretical threshold table is obtained through calculation:

[0067] Table 4. Theoretical Threshold Table The statistical-mechanistic dual-basis approach to rainfall thresholds employs a dual-track verification system of "statistical extreme value boundary (ID curve)" and "mechanistic inversion (saturation required rainfall)". The statistical threshold provides a macroscopic warning line, while the mechanistic threshold provides a dynamic and personalized critical rainfall estimate based on real-time soil water shortage status (calculated from tension data).

[0068] The statistical basis is the rainfall intensity-duration (ID) boundary curve of historical landslide events:

[0069] Parameter description: Critical rainfall intensity; D: Rainfall duration (h) Based on the mechanism and the infinite slope model, the critical rainfall that triggers the landslide is calculated:

[0070] Parameter description: Critical rainfall amount; The depth of the moistened front is calculated in real time from tension data; Initial volumetric water content of the soil; : Soil saturated volumetric water content.

[0071] In one specific embodiment, the system further includes: an anomaly handling module, configured to automatically trigger a corresponding processing mechanism when an anomaly is detected, the processing mechanism including marking sensor faults, initiating alternative algorithms, checking deep leakage, adjusting saturation zone judgment, and triggering a multi-sensor voting mechanism for data repair; a user interface, providing a visual operation platform for displaying real-time monitoring data, early warning information, system state transition processes, and historical data analysis reports; and a security priority adjudication unit, which makes decisions based on preset security priority rules when multi-source threshold adjustment instructions conflict.

[0072] The anomaly handling module is deeply coupled with the data quality verification process. For example, when physical consistency verification identifies a violation of the rainfall-runoff water balance, it automatically marks the rain gauge / flow meter as faulty and initiates an alternative algorithm based on spatial interpolation of nearby stations to generate corrected data; when time-series lag analysis detects... When the threshold is exceeded, the deep seepage diagnosis process is triggered, and the soil profile volumetric water content data is linked to verify the boundary of the saturated zone. When the spatial consistency check finds that the simulation deviation of the wetting front exceeds the limit, the multi-sensor voting mechanism is activated: the data of the five layers of tension sensors are combined, the repair value is generated by weighting according to the confidence level, and an anomaly tracing report (including anomaly type, confidence level, and disposal suggestions) is generated and pushed to the operation and maintenance terminal.

[0073] The user interface can adopt a multi-view collaborative design, with the main map view dynamically rendering a risk heat map and sensor status (green / yellow / red indicators); the warning dashboard displays real-time information. The system displays curves, current system status, and migration trajectory; the data quality diagnosis subpage highlights the location of abnormal sensors and provides interactive buttons for manual confirmation, ignoring, and marking for maintenance; the historical analysis module supports tracing back early warning events along the timeline, viewing parameter optimization comparison charts generated by closed-loop learning, and exporting PDF-format maintenance reports, thus achieving a closed loop of machine diagnosis, manual decision-making, and handling with traceability.

[0074] The safety priority adjudication unit has a built-in three-level conflict resolution rule base. When there is a conflict between the instructions from the closed-loop learning module (which suggests relaxing the threshold) and the threshold mechanism basis base (which suggests tightening the threshold based on the new geological survey), the conservative solution of lowering the threshold is adopted first (safety priority principle). If the conflict involves instructions to correct abnormal measured data and instructions for historical statistics, the measured data source is given higher weight (measurement priority principle). In emergency situations, the adjustment strategy that can shorten the response time is executed first (timeliness priority principle). The adjudication process generates a structured log (including conflict source, adjudication basis, and effective parameters), which is synchronously archived to the audit module to ensure that the decision-making process is traceable and reviewable.

[0075] In one exemplary embodiment, such as Figure 2 As shown, the system's operation flow includes: Three-dimensional parameter matrix loading stage: During system initialization, the pre-set three-dimensional parameter matrix is ​​fully loaded. The three-dimensional parameter matrix adopts a hierarchical architecture: the basic parameter layer stores soil tension threshold, runoff coefficient threshold, rainfall threshold, interflow threshold, etc.; the weight parameter layer defines the weights of each risk index, including the dynamic weight coefficients of the four risk factors (rainfall, soil, runoff, and interflow) in the comprehensive assessment; the algorithm parameter layer contains threshold calculation model coefficients, synergistic effect function parameters, Bayesian learning prior distribution, and other core algorithm parameters, including attenuation coefficient, time window, sensitivity parameters, etc. CRC verification is performed during the loading process to ensure parameter integrity and version consistency, providing a benchmark for subsequent dynamic calculations.

[0076] Multi-source monitoring data acquisition and time synchronization: The system collects key parameters in real time through a distributed sensor network: the total rainfall in the area is obtained through a rain gauge; soil tension is monitored by a multi-layer soil tensiometer (10cm / 30cm / 50cm depth); the flow velocity and flow rate of surface runoff are recorded by an ultrasonic surface runoff sensor; the dynamics of interflow in the soil are collected by a 30cm depth lateral seepage monitoring device; the soil tension is converted into soil pore water pressure by real-time soil tension sensor monitoring values; and the soil moisture meter continuously measures the soil volumetric water content.

[0077] A three-layer dynamic threshold collaborative adjustment mechanism: The system dynamically generates thresholds based on a three-dimensional parameter matrix and environmental conditions, including: Seasonal adjustment: Identifying the current season based on the system calendar and calling the seasonal correction coefficient in the basic parameter layer to macroscopically correct the basic threshold; Previous condition adjustment: Analyzing historical sequences such as past cumulative rainfall and soil volumetric water content decay curves, and using the exponential decay model in the algorithm parameter layer to calculate the previous humidity influence factor, dynamically correcting the threshold benchmark; Real-time feedback adjustment (Bayesian update): Introducing real-time data such as current soil volumetric water content and pore water pressure, identifying abnormal fluctuations through sliding window statistics, and adaptively fine-tuning the threshold at the microsecond level. The results of the three-layer adjustment are weighted and superimposed to generate a dynamic threshold set containing multiple dimensions such as rainfall threshold, soil tension threshold, and runoff threshold, and the effective time window is marked. The range of adjusted thresholds is controlled by adjustment priority conflict resolution rules.

[0078] Risk index calculation and synergistic fusion: Comparison of preprocessed monitoring data (after 3σ anomaly removal, linear interpolation, and unit normalization) with a dynamic threshold set: Rainfall risk index Based on the rainfall data, the critical rainfall intensity is calculated statistically, and the saturation rainfall requirement is calculated mechanistically. After normalizing the critical rainfall intensity and the saturation rainfall requirement, the maximum value of the two is taken to obtain the rainfall risk index. Soil stability index. Based on the soil tension data and an infinite slope stability model, the soil safety factor is inverted; the soil safety factor is then normalized to obtain the soil stability index. Surface runoff index. Risk index of soil midstream Based on the surface runoff data, the real-time runoff coefficient is normalized according to the dynamic flow coefficient threshold to calculate the runoff risk index. Based on the interflow data, a risk index based on flow rate and a risk index based on pore water pressure are calculated. The interflow risk index and the risk index based on pore water pressure are weighted and normalized to obtain the interflow risk index. After weighted summation of the four indices according to the coefficients set in the weight parameter layer, the synergistic effect is considered: when ≥2 indices simultaneously exceed the threshold, a preliminary comprehensive early warning index is generated based on the nonlinear enhancement function in the algorithm parameter layer to quantify the multi-factor coupling risk.

[0079] Multi-dimensional data quality verification and index correction: The system performs triple verification: Physical consistency verification: Real-time check of the physical logic relationship between parameters; Temporal lag analysis: Finding the peak lag time by calculating the cross-correlation function of the rainfall sequence and the interflow sequence at the maximum lag time; Spatial consistency verification: Numerical simulation of the wetting front advance is performed using the simplified Richards equation, and the simulation results are compared with the actual wetting conditions of the five-layer sensor. The data quality coefficient generated from the verification results. And the calculated site safety factor F, the final warning index =LSI (Preliminary Comprehensive Early Warning Index) × ×F effectively suppresses misjudgments caused by data noise and site-specific characteristics.

[0080] Five-level state transition and dynamic resource scheduling: The revised early warning index matches the preset threshold range to trigger state transition: Normal: 10-minute sampling frequency; Attention: 5-minute sampling, push patrol reminder; Alert: 2-minute sampling, emergency team on standby; Emergency: 30-second continuous sampling, triggering audible and visual alarms; Post-disaster: 10-minute sampling, special monitoring mode, initiating disaster reporting process. State transition logs are archived in real time, supporting retrospective analysis.

[0081] Tiered early warning information generation and multi-channel distribution: The system generates structured early warning messages based on the current status, including elements such as timestamp, geographical location, risk index, threshold parameters, and handling suggestions. These messages are simultaneously pushed to the management platform (visual dashboard), responsible personnel's mobile terminals (SMS / APP push), and public information release systems (broadcast / electronic screen) via the API gateway. According to the "Geological Disaster Early Warning Level Specification," the system automatically matches blue warnings to generate a regular report; yellow warnings to generate a reminder; orange warnings to generate a warning notice; and red warnings to trigger emergency response.

[0082] Closed-loop learning and parameter matrix iterative optimization: After the warning event ends, the system automatically activates the closed-loop learning module to evaluate the learning effect and determine whether recalibration is necessary. It collects full-cycle event data (original monitoring values, warning index trajectory, and manual review results); employs a Bayesian update algorithm: using the current parameters as the prior distribution and combining actual disaster feedback to calculate the posterior distribution, optimizing weight coefficients, threshold adjustment parameters, etc.; evaluates the optimization effect through k-fold cross-validation; valid parameters are written to memory and marked with a version number, while invalid updates trigger the manual review process.

[0083] Dual-thread collaborative operation and system self-evolution: The main loop continuously executes the core process of monitoring, evaluation, and early warning to ensure real-time response capabilities; the closed-loop learning module runs as a low-priority asynchronous task in the background between events, with parameter updates written through atomic operations to avoid blocking the main process. Through the dual-engine architecture of real-time decision-making and post-event optimization, the system accumulates knowledge in each early warning practice, gradually improving parameter adaptation accuracy and early warning reliability, and achieving intelligent and adaptive evolution capabilities for specific scenarios.

[0084] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, 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 this application.

Claims

1. A shallow landslide intelligent early warning system, characterized in that, The system includes: The parameter management module is used to store and update the three-dimensional parameter matrix; The data acquisition module is used to collect real-time monitoring data; The risk assessment module is used to calculate various landslide risk indices based on the three-dimensional parameter matrix and the real-time monitoring data, and to generate a comprehensive landslide early warning index through weighted fusion and synergistic effect addition mechanism; The data quality verification and security correction module is configured to perform physical consistency verification, temporal lag correlation analysis and spatial consistency verification on the monitoring data, generate data quality coefficients, and dynamically correct the comprehensive landslide early warning index by integrating site safety factors. The early warning decision and state transition module is configured to determine the early warning level based on the modified comprehensive landslide early warning index and drive the system to transition between multiple preset working states. Each working state corresponds to a different monitoring strategy, parameter adjustment strategy and information output strategy. The closed-loop learning module is configured to iteratively optimize the three-dimensional parameter matrix and the weighted fusion and synergistic effect addition mechanism based on the comparison feedback between the early warning results and actual geological events.

2. The system according to claim 1, characterized in that, The physical consistency verification includes: verifying rainfall and total runoff, verifying surface runoff and interflow, verifying surface and deep soil tension, verifying soil tension and soil volumetric water content, verifying runoff coefficient decomposition, and verifying rainfall-interflow hysteresis response.

3. The system according to claim 1, characterized in that, The time-series lag correlation analysis includes: finding the peak lag time by calculating the cross-correlation function between the rainfall sequence and the interflow sequence at the maximum lag time.

4. The system according to claim 1, characterized in that, The spatial consistency verification includes: performing a numerical simulation of the wetting front using the Richards equation, and comparing the simulation results with the actual wetting conditions.

5. The system according to any one of claims 1 to 4, characterized in that, The dynamic correction of the comprehensive landslide early warning index includes: Normalize all the verification results to generate the data quality coefficient; The site safety factor is calculated based on inherent risk factors, including slope, historical landslide density, and vegetation cover. The comprehensive landslide early warning index is dynamically corrected based on the data quality coefficient and the site safety coefficient.

6. The system according to claim 1, characterized in that, The early warning decision and state transition module includes: comparing the comprehensive landslide early warning index with a preset level threshold to determine the current early warning level, and driving the system to transition between multiple preset working states based on the early warning level; the multiple preset working states include normal state, attention state, alert state, emergency state, and post-disaster state.

7. The system according to claim 6, characterized in that, The early warning decision and state transition module also includes: when the early warning level reaches the highest level, automatically triggering an external emergency response process through an emergency linkage interface.

8. The system according to claim 1, characterized in that, The closed-loop learning module includes: Each warning result is compared with the actual geological event, and the warning event is classified as true positive, false positive, false negative or true negative based on the comparison results; Based on the classification results, the sources of early warning errors are analyzed, the subset of parameters that need to be optimized are identified, and statistical learning methods are used to iteratively update the relevant parameters in the three-dimensional parameter matrix and the weighted fusion coefficients in the risk assessment module.

9. The system according to claim 8, characterized in that, The closed-loop learning module also includes periodically calculating the system's early warning performance indicators and dynamically adjusting the learning strategy and optimization cycle based on the evaluation results.

10. The system according to claim 1, characterized in that, The system also includes: The anomaly handling module is configured to automatically trigger a corresponding processing mechanism when an anomaly is detected. The processing mechanism includes marking sensor failure, starting an alternative algorithm, checking deep leakage, adjusting the saturation zone judgment, and triggering a multi-sensor voting mechanism to repair the data. The user interface provides a visual operation platform for displaying real-time monitoring data, early warning information, system status transition processes, and historical data analysis reports. The security priority decision-making unit makes a decision based on preset security priority rules when multiple source threshold adjustment instructions conflict.

Images