Urban geological safety risk dynamic evaluation system and method based on multi-source data fusion

Abstract

The invention discloses an urban geological safety risk dynamic evaluation method based on multi-source data fusion, and relates to the technical field of geological monitoring, and the method comprises the following steps: S100, synchronously collecting multi-source geological environment data through a space-ground-air integrated monitoring network; s200, a dynamic risk probability is calculated based on the geologic feature vector, a short-term and temporary early warning subsystem outputs a minute-level risk probability, a medium and long term evaluation subsystem outputs an interannual scale risk level, and a graded early warning control chain is triggered according to the value of the risk probability; s300, calculating a comprehensive risk index based on short-term and temporary early warning output and a medium and long term risk evaluation result; and S400, monitoring the early warning accuracy of the S200 and the decision execution effect of the S300 through a feedback verification unit, triggering bidirectional optimization when the early warning false alarm rate exceeds a predetermined threshold or the decision deviation exceeds an allowable range, reversely starting the data reacquisition process of the S100, and synchronously adjusting the risk probability model of the S200 and the risk index calculation weight of the S300.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological monitoring, in particular to a city geological safety risk dynamic evaluation system and method based on multi-source data fusion. BACKGROUND

[0002] City geological safety risk refers to the possibility or uncertainty of natural or human factors leading to karst collapse, goaf collapse, flow collapse and ground subsidence and other city geological safety events and causing losses within a certain period under specific adverse geological conditions. City geological safety events are influenced by regional and human activities and show different characteristics. Natural disaster type city geological safety risk is mainly affected by region, and human activity influenced city geological safety risk mainly reflects ground subsidence, karst collapse, road collapse caused by excessive exploitation of groundwater, goaf collapse caused by mineral resource exploitation, and secondary disasters caused by shallow gas exploitation.

[0003] A method and system for intelligent monitoring and early warning of city geological safety risk are disclosed in Chinese patent No. 2024112735623. First, the monitoring area is divided into nine grids and monitoring points are set. Real-time collection of ground subsidence, collapse and landslide data is performed. Then, these data are input into a prediction model to generate the occurrence probability of various geological risks. The prediction results of each monitoring point are analyzed based on GIS technology to form a risk map and determine the first risk level. Finally, the second risk level is obtained by combining the building coverage rate of different risk areas and comprehensive evaluation. The risk map is updated accordingly and the corresponding warning scheme is started.

[0004] A complex mountainous area geological disaster risk evaluation method and system based on artificial intelligence are disclosed in Chinese patent No. 2024100972350. The data association storage results obtained by prior association can be used to execute mountainous area geological disaster risk evaluation instructions, which can simplify the data association processing process and improve the processing efficiency of mountainous area geological disaster risk evaluation instruction execution, thereby achieving accurate evaluation of geological disaster risk evaluation.

[0005] Similar to the existing technologies mentioned above, when assessing urban geological safety risks, the nine-square grid center point method theoretically pursues spatial uniformity, but seriously ignores the heterogeneity of the urban geological environment. Due to the high uncertainty in the distribution of geological hazards such as landslides, karst caves, and fault zones, the placement of the center point of the machinery is very likely to miss these key hidden danger points, leading to the fundamental failure of the monitoring system. Moreover, geological hazards are low-probability events, and the number of "not occurring" samples in historical data will be far greater than the number of "occurring" samples. Without special processing, the model is very likely to tend to predict "safety" indefinitely, losing its early warning significance. Furthermore, since the geological hazard environment is dynamic, such as groundwater level fluctuations, soil stress changes, and human engineering activities, the pre-correlated data results begin to "become outdated" the moment they are generated. If the system calls "old" correlation results to evaluate "new" risks, it will lead to serious misjudgments.

[0006] Given the aforementioned challenges, achieving accurate, rapid, dynamic assessment and intelligent early warning of urban geological safety risks through multi-source data fusion has become a core technological hurdle that needs to be overcome in urban geological safety risk management. 。

[0007] Therefore, it is necessary to invent a dynamic assessment system and method for urban geological safety risks based on multi-source data fusion to solve the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a dynamic assessment system and method for urban geological safety risks based on multi-source data fusion, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a dynamic assessment method for urban geological safety risks based on multi-source data fusion, comprising the following steps: S100: Multi-source geological environment data are collected synchronously through an integrated air-ground-ground monitoring network. Deformation and displacement data obtained by ground sensor array, deformation data obtained by aerial remote sensing equipment, and pore water pressure data obtained by underground monitoring equipment are fused through a spatiotemporal alignment mechanism to generate a geological feature vector. S200. Calculate the dynamic risk probability based on the geological feature vector, wherein the short-term early warning subsystem outputs the risk probability at the minute level, the medium- and long-term evaluation subsystem outputs the risk level at the annual scale, and triggers the graded early warning control chain according to the value of the risk probability. S300. A comprehensive risk index is calculated based on short-term early warning output and medium- and long-term risk assessment results. The calculation result of the risk index is directly related to the decision response chain. The comprehensive risk index is obtained by weighting the short-term early warning probability and the long-term risk level. S400, monitoring the early warning accuracy of S200 and the decision execution effect of S300 by the feedback verification unit, triggering bidirectional optimization when the early warning false alarm rate exceeds a predetermined threshold or the decision deviation exceeds the allowed range, reversely starting the data reacquisition process of S100, and synchronously adjusting the risk probability model of S200 and the risk index calculation weight of S300; S500, writing the optimization result of S400 into the digital twin platform, generating an evaluation report containing a risk zoning map and an emergency scheme suggestion, and completing the full-chain closed loop from data acquisition to decision optimization.

[0010] Preferably, in step S100, the space-time alignment mechanism comprises binding the ground displacement data, remote sensing deformation data and underground pore water pressure data to a unified geographic coordinate system by a space-time alignment module, wherein the ground sensor collects real-time data to generate a displacement point cloud at a fixed time interval; the aerial remote sensing device obtains regional deformation distribution during the data acquisition gap; when the spatial deviation of the point cloud and the deformation distribution exceeds a set threshold, the underground monitoring device is triggered to supplement deep data for calibration, wherein the calibration algorithm adopts a weighted least squares fitting method, and corresponding weight values are allocated according to the accuracy of different sensors.

[0011] Preferably, in step S200, the calculation of the dynamic risk probability is output by a short-term early warning model and a medium and long-term evaluation model in parallel, wherein the short-term early warning adopts an attention mechanism long and short term memory network model, the input includes parameters such as pore water pressure change rate, displacement acceleration and rainfall intensity, and the probability output is obtained through nonlinear transformation, the medium and long-term evaluation adopts a multi-index comprehensive evaluation model, the risk level is determined by combining the analytic hierarchy process and the entropy weight method to calculate the weight, and the activation condition and execution effect of the early warning control chain form a closed loop, the early warning level is ladder type increased with the increase of the risk probability, and the emergency response intensity is proportionally related to the risk level; When the short-term early warning probability exceeds a first threshold value, a multi-model joint verification mechanism is activated to recalibrate the input data; when the risk probability still exceeds the first threshold value after calibration, a high-level early warning instruction is executed and an emergency response device is started.

[0012] Preferably, in step S300, the comprehensive risk index is determined by the short-term early warning probability, the medium and long-term risk level and the geological environment parameters, and the weight distribution is dynamically adjusted based on the real-time risk change rate; when the comprehensive risk index is lower than a second threshold value, a regular monitoring mode is triggered and a low risk notice is issued, and the decision response chain comprises a double-layer feedback mechanism: First layer: calculating the emergency response range according to the risk index, the response range being positively correlated with the risk index; Second layer: dynamically adjusting the risk index calculation weight based on the actual disaster reduction effect, and adjusting the weight according to the change amount of the disaster reduction effect.

[0013] Preferably, the S400 comprises the following steps: S410, monitoring the early warning accuracy index in real time through the feedback verification engine, and the accuracy is calculated based on historical early warning and actual disaster occurrence; S420, when the early warning accuracy index is lower than the third threshold value, increasing the calculation weight of the real-time monitoring data in the geological feature vector, and synchronously monitoring the decision execution deviation data; S430, when the decision execution deviation exceeds the fourth threshold value, reducing the calculation weight of the historical data in the risk index; S440, writing the optimized weight parameters into the risk probability calculation rule of S200 and the risk index calculation rule of S300 in real time; S450, generating a new early warning instruction based on the updated calculation rule.

[0014] Preferably, in the step S500, the prediction model reliability is predicted by collecting system data quality indicators in real time, including signal-to-noise ratio, sampling completeness rate, timeliness delay and other parameters, when the reliability is lower than a preset value, a high-risk area is marked in the digital twin platform, and the monitoring network is optimized to preferentially cover the blind area, and a calibration alarm is generated and associated with a data quality report.

[0015] Preferably, the normalization processing of the short-term early warning probability and the medium and long-term risk level adopts a dynamic standardization method, the original probability and level are standardized based on the statistical characteristics of historical disaster data, the data window size used for statistical characteristic calculation is dynamically adjusted according to the real-time risk change rate, a smaller data window is used when the risk change rate is high to improve real-time performance, otherwise a larger data window is used to reduce the influence of noise, and the standardized comprehensive risk index is obtained by weighted calculation of the probability standardized value and the level standardized value, wherein the weight coefficient is dynamically allocated by the bidirectional optimization module.

[0016] Preferably, the compatibility of the data reacquisition and the original data is guaranteed by a space-time alignment engine, a global clock signal is injected into the reacquired data to realize time synchronization, the reacquired data is spatially aligned with the original point cloud coordinate system through the geological semantic label, the alignment algorithm adopts a matrix decomposition method to obtain an optimal transformation matrix, a data conflict resolution rule is established, when the risk probability difference is greater than a preset value, the digital twin model is updated according to the reacquired data, when the sensor data is not detected in the reacquisition, the historical data is used and marked with a confidence degree, and the confidence degree is calculated based on a probability reasoning model.

[0017] The application also provides a city geological safety risk dynamic evaluation system based on multi-source data fusion, which is used to realize the city geological safety risk dynamic evaluation method based on multi-source data fusion, and comprises a multi-source sensing unit, an intelligent analysis unit, a feedback verification unit and a digital twin management unit. The multi-source perception unit comprises a ground, air and underground integrated module, a global navigation satellite system receiver, a synthetic aperture radar sensor and a pore water pressure gauge co-substrate design, and outputs data streams bound to the same coordinate system, and an underground compensation module is activated when the ground and air data deviation is greater than a threshold value; The intelligent analysis unit comprises a risk control chain module and a decision response chain module, the risk control chain module is used for receiving a geological feature vector and outputting a warning instruction interlocked with a risk probability, and the decision response chain module is used for receiving a risk index and outputting a control signal interlocked with a response range, and interlocking logic is realized based on a rule engine.

[0018] Preferably, the feedback verification unit comprises a data quality monitoring module and a weight optimization module, the data quality monitoring module monitors a signal-to-noise ratio, a sampling rate and a timeliness index in real time, and triggers data reacquisition when the index is abnormal; the weight optimization module updates a calculation weight by using a dynamic adjustment algorithm, and visualizes and displays an optimization result by a digital twin management unit.

[0019] Technical effects and advantages of the present application: The present application combines sensor data fusion and underground compensation calibration, adopts dynamic weight distribution by an intelligent analysis unit, and improves model sensitivity by relying on a double-threshold activation mechanism, and triggers two-way optimization by a false alarm rate and a decision deviation by a feedback verification unit, reversely locates abnormal data reacquisition, and positively incrementally learns and adjusts weights, to form a “targeted perception-dynamic evaluation-closed-loop calibration” whole-chain prevention and control system, so as to realize precise and rapid dynamic evaluation and intelligent early warning of urban geological safety risk by multi-source data fusion. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is a flowchart of the method of the present application.

[0021] Figure 2 It is a system framework diagram of the method of the present application.

[0022] Figure 3 It is a logic control diagram of two-way optimization in the method of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] First embodiment Since geological disasters are small probability events, the number of "non-occurrence" samples in historical data will be much larger than that of "occurrence" samples, if not specially processed, the model will easily tend to predict "safety" forever, losing the significance of early warning, and since the geological disaster environment is dynamically changing, such as groundwater level fluctuation, soil stress change and human engineering activities, the data results of pre-association start to be "outdated" at the moment of generation, to solve the problem of serious misjudgment caused by calling "old" association results to evaluate "new" risks.

[0025] Therefore, to solve the above problems, as shown in the Figure 1 Figure 3 The present application provides a city geological safety risk dynamic evaluation system based on multi-source data fusion to overcome the above problems, which comprises a source perception unit, an intelligent analysis unit, a feedback verification unit and a digital twin management unit.

[0026] In this embodiment, the multi-source perception unit includes ground, air and underground integrated modules, which are designed with a global navigation satellite system receiver, a synthetic aperture radar sensor and a pore water pressure gauge on the same substrate, outputting data streams bound to the same coordinate system, and the underground compensation module is activated when the deviation between ground and air data is greater than the threshold.

[0027] In this embodiment, the intelligent analysis unit includes a risk control chain module and a decision response chain module, the risk control chain module is used to receive geological feature vectors and output early warning instructions interlocked with risk probability, and the decision response chain module is used to receive risk indexes and output control signals interlocked with response range, and the interlocking logic is realized based on a rule engine.

[0028] In this embodiment, the feedback verification unit includes a data quality monitoring module and a weight optimization module, the data quality monitoring module monitors the signal-to-noise ratio, sampling rate and timeliness indicators in real time, and triggers data reacquisition when the indicators are abnormal; the weight optimization module updates the calculation weight by using a dynamic adjustment algorithm, and visualizes and displays the optimization results through the digital twin management unit.

[0029] ​In use, the deformation displacement data of the ground sensor array, the deformation data of the aerial remote sensing device and the pore water pressure data of the underground monitoring device are synchronously collected by the multi-source perception unit, and the multi-source data is fused to generate a geological feature vector in a unified coordinate system by using a space-time alignment mechanism, and the initialization of the risk control chain module and the decision response chain module in the intelligent analysis unit is completed, the risk control chain module processes real-time data streams based on an attention mechanism long short-term memory network architecture, outputs minute-level risk probability to the decision response chain module, and receives data quality indicators from the feedback verification unit to dynamically adjust the early warning threshold parameters, the decision response chain module fuses the risk probability and the geological environment parameters by means of a multi-index comprehensive evaluation model, generates a comprehensive risk index, and optimizes the output effect of the emergency response strategy. The feedback verification unit adopts a real-time monitoring algorithm, calculates a weight adjustment value according to data quality indicators such as signal-to-noise ratio, sampling rate and timeliness, simultaneously feeds back the optimized parameters to the intelligent analysis unit, and monitors the early warning accuracy, decision deviation and system load through the digital twin management unit throughout the process, and once it is found that the indicators exceed the standard, for example, the false positive rate exceeds 5% or the decision deviation is greater than 0.2, the multi-module collaborative recalculation or the hierarchical degradation protocol is triggered, and finally a dynamic evaluation report with high risk area division accuracy, optimized response strategy and high data continuity is output.

[0030] Embodiment two Based on the above-mentioned system framework, the specific steps of the urban geological safety risk dynamic evaluation system based on multi-source data fusion proposed by the present application are elaborated in detail, so as to realize the accurate and rapid dynamic evaluation and intelligent early warning of urban geological safety risk based on multi-source data fusion.

[0031] The method comprises the following steps: S100, synchronously collecting multi-source geological environment data through a space-ground-air integrated monitoring network, wherein the deformation displacement data obtained by the ground sensor array, the deformation data obtained by the aerial remote sensing device and the pore water pressure data obtained by the underground monitoring device are fused to generate a geological feature vector by a space-time alignment mechanism.

[0032] In step S100 of the embodiment, the space-time alignment mechanism includes binding the ground displacement data, the remote sensing deformation data and the underground pore water pressure data to a unified geographic coordinate system through a space-time alignment module, wherein the ground sensor collects real-time data at a fixed time interval to generate a displacement point cloud; the aerial remote sensing device obtains regional deformation distribution during the data collection interval; when the spatial deviation of the point cloud and the deformation distribution exceeds a set threshold, the underground monitoring device is triggered to supplement deep data for calibration, wherein the calibration algorithm adopts a weighted least squares fitting method, and different sensor accuracies are assigned with corresponding weight values.

[0033] It should be noted that, for ease of understanding, the weight of GNSS data is set to 0.8, the weight of InSAR data is set to 0.6, and the weight of underground data is set to 0.9. The GNSS data comes from global navigation satellite systems such as GPS, BeiDou, and GLONASS, and mainly includes high-precision location information, displacement vectors, velocity changes, and temporal deformation data of ground monitoring points. In this method, GNSS sensors are deployed in the ground monitoring network to collect displacement data with millimeter-level accuracy, such as horizontal displacement and vertical settlement, in real time. The sampling frequency is set to 1Hz to 100Hz to track dynamic changes on the ground surface.

[0034] InSAR data, or Synthetic Aperture Radar Interferometry data, is acquired by radar satellites and used to generate surface deformation maps, phase information, and elevation changes. By providing large-scale, periodic deformation monitoring, it can detect slow or regional land subsidence and fault activity. It supplements the spatial coverage of GNSS data, and is particularly suitable for remote areas or areas where ground sensors are difficult to deploy, thus providing the system with macroscopic deformation trends and helping to identify large-area risk areas such as land subsidence funnels. Its weight is set to 0.6 because although InSAR data has a wide coverage, its accuracy is relatively low due to atmospheric interference, spatiotemporal resolution limitations, and data processing delays.

[0035] Underground data, such as pore water pressure and soil stress, come from directly buried sensors. They are highly accurate and sensitive to changes in deep geology, so they have a weight close to 1 and are used to calibrate other data.

[0036] S200. Calculate the dynamic risk probability based on the geological feature vector, wherein the short-term early warning subsystem outputs the risk probability at the minute level, the medium- and long-term evaluation subsystem outputs the risk level at the annual scale, and triggers the graded early warning control chain according to the value of the risk probability.

[0037] In step S200 of this embodiment, the calculation of dynamic risk probability is output in parallel by the short-term early warning model and the medium- and long-term evaluation model. The short-term early warning adopts a long short-term memory network model with attention mechanism. The input includes parameters such as pore water pressure change rate, displacement acceleration, and rainfall intensity. The probability output is obtained through nonlinear transformation. The medium- and long-term evaluation adopts a multi-index comprehensive evaluation model. The risk level is obtained by combining the analytic hierarchy process and the entropy weight method to determine the weight. The activation conditions and execution effects of the early warning control chain form a closed loop. The early warning level increases stepwise with the increase of risk probability. The emergency response intensity is proportionally related to the risk level. When the probability of a short-term warning exceeds the first threshold, the multi-model joint verification mechanism is activated to recalibrate the input data; if the risk probability still exceeds the first threshold after calibration, an advanced warning instruction is executed and the emergency response device is activated.

[0038] It should be noted that the short-term early warning adopts an Attention-LSTM model, and the input includes the change rate of pore water pressure (ΔP / Δt), displacement acceleration (d2s / dt2), and rainfall intensity (I), and the output = σ( · [LSTM_output] + ), wherein σ is a Sigmoid function; the medium and long-term evaluation adopts a multi-index comprehensive evaluation model, and the risk level = Σ( · ), the weight is determined by an AHP-entropy weight method.

[0039] S300, calculate a comprehensive risk index based on the short-term early warning output and the medium and long-term risk evaluation result, and the calculation result of the risk index is directly related to a decision response chain, wherein the comprehensive risk index is obtained by weighted calculation of the short-term early warning probability and the long-term risk level.

[0040] It should be noted that the short-term early warning subsystem output minute-level risk probability is defined as , the medium and long-term evaluation subsystem output annual-scale risk level , and the comprehensive risk index is obtained by weighted calculation of and .

[0041] In step S300 of the embodiment, the comprehensive risk index is determined by the short-term early warning probability, the medium and long-term risk level, and the geological environment parameter, and the weight distribution is dynamically adjusted based on the real-time risk change rate; when the comprehensive risk index is lower than a second threshold value, a conventional monitoring mode is triggered and a low-risk notice is issued, and the decision response chain includes a double-layer feedback mechanism: First layer: calculate an emergency response range according to the risk index, and the response range is positively correlated with the risk index.

[0042] Second layer: dynamically adjust the risk index calculation weight based on the actual disaster reduction effect, and the weight is adjusted according to the change amount of the disaster reduction effect.

[0043] It should be noted that the weight distribution is dynamically adjusted based on the real-time risk change rate, and the calculation formula is:

[0044] , wherein is the short-term early warning probability, and the value range is [0, 1]; is the medium and long-term risk level, and the value range is [0, 1]; and are dynamic weight coefficients, and satisfy the constraint condition + =1. The output range is normalized to [0, 1] for the comprehensive risk index.

[0045] The risk index calculation weight is dynamically adjusted based on the actual disaster reduction effect, and the weight adjustment formula is:

[0046] wherein, is a learning rate coefficient, and the value range is (0, 1), is the normalized disaster reduction effect change, is a time decay factor.

[0047] S400, the pre-warning accuracy of S200 and the decision execution effect of S300 are monitored by a feedback verification unit, and when the pre-warning false alarm rate exceeds a predetermined threshold or the decision deviation exceeds the allowed range, bidirectional optimization is triggered, and the data reacquisition process of S100 is started in reverse, and the risk probability model of S200 and the risk index calculation weight of S300 are adjusted synchronously.

[0048] It should be noted that, for the sake of understanding, the preset threshold of the pre-warning false alarm rate is set to 5%, and the allowed range of the decision deviation is 0.2, that is, when the pre-warning false alarm rate exceeds 5% or the decision deviation δ> 0.2, bidirectional optimization is triggered.

[0049] S500, the optimization result of S400 is written into the digital twin platform, an evaluation report containing the risk zoning map and the emergency scheme suggestion is generated, and the whole chain closed loop from data acquisition to decision optimization is completed.

[0050] In step S500 of the embodiment, the system data quality indicators are collected in real time, including signal-to-noise ratio, sampling completeness, timeliness delay and other parameters, and the reliability of the prediction model is predicted. When the reliability is lower than the preset value, mark the high-risk area in the digital twin platform, and optimize the monitoring network to cover the blind area preferentially, and generate a calibration alarm and associate a data quality report.

[0051] In this embodiment, the normalization processing of short-term early warning probability and medium and long-term risk level adopts a dynamic standardization method, the original probability and level are standardized based on the statistical characteristics of historical disaster data, and the data window size used for statistical characteristic calculation is dynamically adjusted according to the real-time risk change rate. When the risk change rate is high, a smaller data window is used to improve real-time performance, otherwise a larger data window is used to reduce noise influence. The standardized comprehensive risk index is obtained by weighted calculation of the probability standardized value and the level standardized value, wherein the weight coefficient is dynamically allocated by the bidirectional optimization module.

[0052] In the embodiment, the compatibility of data reacquisition and original data is guaranteed by a space-time alignment engine, a global clock signal is injected into the reacquired data to realize time synchronization, the reacquired data is spatially aligned with the original point cloud coordinate system through a geological semantic label, an optimal transformation matrix is solved by a matrix decomposition method, a data conflict resolution rule is established, the digital twin model is updated according to the reacquired data when the risk probability difference is greater than a preset value, historical data is used when sensor data is not detected in reacquisition and is marked with confidence, and the confidence is calculated based on a probability reasoning model.

[0053] In the embodiment, S400 further includes the following steps: S410, real-time monitoring of early warning accuracy indicators by a feedback verification engine, the accuracy being calculated based on historical early warnings and actual disaster occurrence.

[0054] S420, when the early warning accuracy indicators are lower than a third threshold value, increasing the calculation weight of real-time monitoring data in the geological feature vector, and synchronously monitoring decision execution deviation data.

[0055] S430, when the decision execution deviation exceeds a fourth threshold value, reducing the calculation weight of historical data in the risk index.

[0056] S440, real-time writing of the optimized weight parameters into the risk probability calculation rule of S200 and the risk index calculation rule of S300.

[0057] S450, generating of new early warning instructions based on the updated calculation rules.

[0058] It should be noted that in step S420, the real-time monitoring data weight adjustment is:

[0059] wherein, the basic weight coefficient is represented by, the adjustment amplitude coefficient is represented by, the time attenuation coefficient is represented by, and the risk probability change amount is represented by.

[0060] In step S430, the historical data weight reduction amount formula is:

[0061] wherein, the deviation sensitivity coefficient is represented by, the current historical data weight is represented by, and the decision execution deviation amount is represented by.

[0062] The real-time monitoring data weight is processed by combining the basic value with the dynamic adjustment term, a fixed basic weight is set to ensure the minimum participation degree, the real-time data proportion is increased according to the risk fluctuation, and the weight is prevented from oscillating sharply through the time decay factor, and the historical data weight adjustment adopts a deviation grading response strategy, which is divided into three response levels according to the decision execution deviation degree, i.e. mild deviation conservative adjustment, moderate deviation standard adjustment and severe deviation intensive adjustment, and the weight reduction amount is proportional to the current historical data weight, so as to avoid excessive punishment for small deviation, while fully responding to major deviation.

[0063] It should be noted that the feedback engine realizes collaborative work, so that the real-time data weight ensures that the risk calculation is synchronized with the current situation, the historical data weight adjustment ensures that the system finely responds to the decision deviation, and through mathematical constraints and decay control, the system maintains the stability of operation while making the risk calculation rule have self-adaptability, logical interpretability and operation efficiency, thereby forming a complete risk calculation optimization closed loop, which not only meets the real-time requirement, but also ensures the reliability of the system in long-term operation.

[0064] In use, the geological environment data is synchronously collected through the integrated monitoring network of space, time and space, when the spatial deviation of displacement point cloud and deformation distribution exceeds the preset value, the underground compensation module is automatically activated, and the weighted least squares method is used for calibration, the geological feature vector in the unified coordinate system is output, and the signal-to-noise ratio, sampling completeness and timeliness are checked in real time through the data quality monitoring module.

[0065] If the index is abnormal, the data reacquisition process is triggered, and the global clock signal is injected through the space-time alignment engine to realize time synchronization of new and old data, then the geological feature vector is input into the intelligent analysis unit, the short-term and short-impending warning subsystem analyzes the parameters such as the change rate of pore water pressure, displacement acceleration and rainfall intensity in real time, and outputs the minute-level risk probability, if the risk probability exceeds the set value, the system starts the multi-model joint verification mechanism, and the medium and long-term evaluation subsystem generates the annual scale risk grade based on the analytic hierarchy process-entropy weight method, and the static parameters such as geological structure and historical disaster records, and the two are weighted to generate a comprehensive risk index, when the comprehensive risk index exceeds the set value, a red warning is triggered and traffic control and personnel evacuation are activated.

[0066] The decision response chain module executes graded response according to the comprehensive risk index value, if it is low risk, a regular monitoring notice is issued, if it is high risk, the municipal departments are linked to start traffic control, personnel evacuation and engineering reinforcement, and the weight is optimized in real time through the disaster reduction effect feedback, the digital twin management unit synchronously generates a dynamic risk heat map, marks the high-risk blind area with a settlement rate exceeding the preset value, optimizes the monitoring network coverage, and outputs the grouting path planning and other emergency schemes.

[0067] When the false alarm rate of the short-term early warning model exceeds the set value, it indicates that the model may be overfitting or have data bias. When the decision execution deviation, i.e., the difference between the actual disaster loss and the predicted loss, is greater than the preset value, it indicates that the decision response chain is out of touch with the actual needs. At this time, the feedback verification unit continuously calculates these indicators. Once the standard is exceeded, the optimization process is immediately activated. The optimization procedure includes two directions: forward optimization and backward optimization. Among them, backward optimization corrects the input bias by re-collecting data. First, it locates the abnormal data area. If the false alarm rate of S200 increases, the feedback verification unit identifies the data of a specific geographical area, such as a subway construction area, through spatiotemporal correlation analysis. It uses isolated forest to detect abnormal data points and combines the visualization tools of the digital twin platform to highlight the problem area. It dispatches mobile monitoring equipment, and its deployment density is dynamically adjusted based on the risk level. The data of this area is re-collected. The re-collected data is integrated with the existing data through a spatiotemporal alignment mechanism. The calibration algorithm adopts weighted least squares fitting to ensure the consistency between the new data and the historical data. After re-collection, the data layer performs a quality check.

[0068] Forward optimization focuses on the model layer, improving prediction and decision-making accuracy by adjusting parameters and weights. Based on re-collected data, it optimizes the input weights of short-term early warning models and medium- to long-term evaluation models. When the false alarm rate remains high, it triggers an incremental learning process to fine-tune model parameters using new data, avoiding the computational overhead of global retraining. The re-collected data is immediately used for model adjustment, while the model output guides the next data collection focus, such as prioritizing re-collection of high-risk areas identified by the model. Through the interaction and closed-loop control of forward and backward optimization, it solves the problem that data collection and model analysis often operate independently, causing model updates to lag behind environmental changes. This enables accurate, rapid, dynamic evaluation and intelligent early warning of urban geological safety risks through multi-source data fusion.

[0069] It should be noted that this invention eliminates missed detections due to spatial heterogeneity by fusing data from air, ground, and space sensors and using an underground compensation calibration mechanism. Furthermore, it employs a dynamic weight allocation through an intelligent analysis unit to suppress safety biases caused by sample imbalance. It also enhances model sensitivity through a dual-threshold activation mechanism. The feedback verification unit triggers bidirectional optimization based on false alarm rate and decision bias, reversely locates abnormal data for re-collection, and adjusts weights through forward incremental learning. This forms a full-chain prevention and control system of "targeted perception – dynamic assessment – ​​closed-loop calibration," thereby achieving accurate, rapid, dynamic evaluation and intelligent early warning of urban geological safety risks through multi-source data fusion.

[0070] Finally, it should be noted that the above description is only preferred embodiments of the present application, and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, modifications or equivalent replacements of the technical solutions described in the foregoing embodiments, or equivalent replacements of part of the technical features, can still be made by those skilled in the art within the spirit and principle of the present application, and any modifications, equivalent replacements, improvements, etc. made within the scope of the present application should be included in the protection scope of the present application.

[0071] Finally, it should be noted that the above description is only preferred embodiments of the present application, and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, modifications or equivalent replacements of the technical solutions described in the foregoing embodiments, or equivalent replacements of part of the technical features, can still be made by those skilled in the art within the spirit and principle of the present application, and any modifications, equivalent replacements, improvements, etc. made within the scope of the present application should be included in the protection scope of the present application.

Claims

1. A dynamic assessment method for urban geological safety risks based on multi-source data fusion, characterized in that, Includes the following steps: S100: Multi-source geological environment data are collected synchronously through an integrated air-ground-ground monitoring network. Deformation and displacement data obtained by ground sensor array, deformation data obtained by aerial remote sensing equipment, and pore water pressure data obtained by underground monitoring equipment are fused through a spatiotemporal alignment mechanism to generate a geological feature vector. S200. Calculate the dynamic risk probability based on the geological feature vector, wherein the short-term early warning subsystem outputs the risk probability at the minute level, the medium- and long-term evaluation subsystem outputs the risk level at the annual scale, and triggers the graded early warning control chain according to the value of the risk probability. S300. A comprehensive risk index is calculated based on short-term early warning output and medium- and long-term risk assessment results. The calculation result of the risk index is directly related to the decision response chain. The comprehensive risk index is obtained by weighting the short-term early warning probability and the long-term risk level. S400 monitors the early warning accuracy of S200 and the decision execution effect of S300 through the feedback verification unit. When the false alarm rate of the early warning exceeds the predetermined threshold or the decision deviation exceeds the allowable range, it triggers bidirectional optimization, reverses the data re-acquisition process of S100, and synchronously adjusts the risk probability model of S200 and the risk index calculation weight of S300. The S500 incorporates the optimization results of the S400 into the digital twin platform, generating an assessment report that includes a risk zoning map and emergency response plan recommendations, thus completing a closed loop from data collection to decision optimization.

2. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, In step S100, the spatiotemporal alignment mechanism includes binding ground displacement data, remote sensing deformation data, and underground pore water pressure data to a unified geographic coordinate system through a spatiotemporal alignment module. Ground sensors collect real-time data at fixed time intervals to generate displacement point clouds; aerial remote sensing equipment acquires regional deformation distribution during data acquisition intervals; when the spatial deviation between the point cloud and the deformation distribution exceeds a set threshold, underground monitoring equipment is triggered to supplement deep data for calibration. The calibration algorithm adopts a weighted least squares fitting method, assigning corresponding weight values ​​according to the accuracy of different sensors.

3. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, In step S200, the calculation of dynamic risk probability is output in parallel by the short-term early warning model and the medium- and long-term evaluation model. The short-term early warning adopts a long short-term memory network model with attention mechanism. The input includes parameters such as pore water pressure change rate, displacement acceleration, and rainfall intensity. The probability output is obtained through nonlinear transformation. The medium- and long-term evaluation adopts a multi-index comprehensive evaluation model. The risk level is obtained by combining the analytic hierarchy process and the entropy weight method to determine the weights. The activation conditions and execution effects of the early warning control chain form a closed loop. The early warning level increases stepwise with the increase of risk probability. The emergency response intensity is proportionally related to the risk level. When the probability of a short-term warning exceeds the first threshold, the multi-model joint verification mechanism is activated to recalibrate the input data; if the risk probability still exceeds the first threshold after calibration, an advanced warning instruction is executed and the emergency response device is activated.

4. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, In step S300, the comprehensive risk index is jointly determined by the short-term early warning probability, the medium- and long-term risk level, and geological environmental parameters, and the weight allocation is dynamically adjusted based on the real-time risk change rate. When the comprehensive risk index falls below the second threshold, the regular monitoring mode is triggered and a low-risk announcement is issued. The decision-response chain includes a two-layer feedback mechanism: The first layer: Calculate the emergency response range based on the risk index, and the response range is positively correlated with the risk index; The second layer: The risk index calculation weights are dynamically adjusted based on the actual disaster reduction effect, and the weights are adjusted accordingly based on the changes in the disaster reduction effect.

5. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, The S400 includes the following steps: S410: Real-time monitoring of early warning accuracy indicators through a feedback verification engine, with accuracy calculated based on historical early warnings and actual disaster occurrences; S420. When the early warning accuracy rate index is lower than the third threshold, increase the calculation weight of real-time monitoring data in the geological feature vector and simultaneously monitor decision execution deviation data. S430. When the decision execution deviation exceeds the fourth threshold, reduce the weight of historical data in the risk index. S440. Write the optimized weight parameters into the risk probability calculation rule of S200 and the risk index calculation rule of S300 in real time. S450: Generate new warning instructions based on the updated calculation rules.

6. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, In step S500, the reliability of the model is predicted by real-time collection of system data quality indicators, including parameters such as signal-to-noise ratio, sampling completeness rate, and timeliness delay. When the reliability is lower than the preset value, high-risk areas are marked in the digital twin platform, and the monitoring network is optimized to prioritize coverage of blind spots. At the same time, a calibration alarm is generated and associated with a data quality report.

7. The method for dynamic assessment of urban geological safety risks according to claim 3, characterized in that, The normalization of short-term warning probability and medium-to-long-term risk level adopts a dynamic standardization method. Based on the statistical characteristics of historical disaster data, the original probability and level are standardized. The size of the data window used for statistical feature calculation is dynamically adjusted according to the real-time risk change rate. When the risk change rate is high, a smaller data window is used to improve real-time performance; otherwise, a larger data window is used to reduce noise impact. The standardized comprehensive risk index is obtained by weighted calculation of probability standardization value and level standardization value, where the weight coefficient is dynamically allocated by the bidirectional optimization module.

8. The method for dynamic assessment of urban geological safety risks according to claim 1, characterized in that, The compatibility between the reacquired data and the original data is ensured by a spatiotemporal alignment engine. A global clock signal is injected into the reacquired data to achieve time synchronization. Geological semantic tags are used to spatially align the reacquired data with the original point cloud coordinate system. The alignment algorithm uses matrix decomposition to find the optimal transformation matrix and establishes data conflict resolution rules. When the risk probability difference is greater than a preset value, the digital twin model is updated based on the reacquired data. When sensor data is not detected in the reacquired data, historical data is used and the confidence level is marked. The confidence level is calculated based on a probabilistic inference model.

9. A dynamic assessment system for urban geological safety risks based on multi-source data fusion, the system being used to implement the dynamic assessment method for urban geological safety risks based on multi-source data fusion as described in any one of claims 1-8, characterized in that, include: Multi-source sensing unit, intelligent analysis unit, feedback verification unit, digital twin management unit; The multi-source sensing unit includes ground, air and underground integrated modules. It uses a global navigation satellite system receiver, synthetic aperture radar sensor and pore water pressure gauge on a common substrate design to output a data stream bound to the same coordinate system. The underground compensation module is activated when the deviation between the ground and air data is greater than a threshold. The intelligent analysis unit includes a risk control chain module and a decision response chain module. The risk control chain module is used to receive geological feature vectors and output early warning instructions that are interlocked with risk probabilities. The decision response chain module is used to receive risk indices and output control signals that are interlocked with response ranges. The interlocking logic is implemented based on a rule engine.

10. The urban geological safety risk dynamic assessment system according to claim 9, characterized in that, The feedback verification unit includes a data quality monitoring module and a weight optimization module. The data quality monitoring module monitors the signal-to-noise ratio, sampling rate, and timeliness indicators in real time, and triggers data re-collection when the indicators are abnormal. The weight optimization module uses a dynamic adjustment algorithm to update the calculated weights and visualizes the optimization results through a digital twin management unit.

Images