A tunnel portal slope early warning method and system based on multi-source monitoring fusion

Abstract

This invention relates to the field of signal devices and early warning systems, specifically a method and system for early warning of tunnel entrance slopes based on multi-source monitoring fusion. The method includes: acquiring data through a three-dimensional monitoring network; predicting future deformation trends using a long short-term memory network model combined with displacement and rainfall factors; evaluating the slope evolution stages using finite element numerical analysis; constructing a dynamic fact base for an expert system based on the evaluation results; using an inference engine to perform pattern matching between the fact base and a rule base containing slope cutting, surcharge counterpressure, slope protection, and drainage measures to generate suggestions; activating an alarm mechanism when a graded threshold is triggered and displaying the path and suggestions using a three-dimensional visualization platform. This invention, through multi-source monitoring fusion and expert decision-making linkage, achieves a predictive alarm system with the ability to predict disaster mechanisms and intelligently generate prevention and control measures.

Description

Technical Field

[0001] This invention relates to the field of signal devices and early warning systems, specifically to a method and system for early warning of tunnel entrance slopes based on multi-source monitoring fusion. Background Technology

[0002] Due to the complex geological structure of tunnel entrance slopes, they are highly susceptible to instability when exposed to external environmental disturbances (such as rainfall, earthquakes, or human activities). Such incidents can lead to severe damage to facilities and operational safety accidents. Traditional slope monitoring methods often rely on manual inspections or single displacement monitoring devices. This "point-based monitoring" model is inadequate and limited when dealing with highly nonlinear and non-uniform geological formations like tunnel entrance slopes. Furthermore, the limited monitoring indicators result in high false alarm and false negative rates, and the systems fail to reflect the evolution of the slope's internal mechanical state, lacking effective extrapolation and prediction of future trends. Moreover, existing systems are mostly variable measurement tools, leading to a disconnect between monitoring information and subsequent preventative measures, making it difficult to automatically generate targeted prevention strategies or emergency plans based on the characteristics of the damage.

[0003] To address this, a method and system for early warning of tunnel entrance slopes based on multi-source monitoring fusion is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a tunnel entrance slope early warning method and system based on multi-source monitoring fusion. Through multi-dimensional perception, time-series prediction and expert decision-making linkage, it can realize the prediction of disaster mechanism and intelligent generation of prevention and control measures.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A tunnel entrance slope early warning method based on multi-source monitoring fusion includes: A three-dimensional monitoring network deployed on the slope of the tunnel entrance is used to acquire multi-source monitoring data in real time. The multi-source monitoring data is input into a long short-term memory network model. Using displacement time series as a feature and combining it with rainfall intensity synergy factors, the future deformation trend is predicted through nonlinear extrapolation calculation. Based on the aforementioned future deformation trend, combined with the slope failure mode evolution mechanism obtained from finite element numerical analysis, the current evolution stage of the slope is dynamically assessed, and the slope is identified as being in the initial creep stage, the accelerated deformation stage, or the critical slip stage. Based on the assessment results, disease type, location of occurrence, numerical descriptive parameters and textual descriptive parameters are extracted to construct a dynamic fact base for the JESS expert system. Using the inference engine of the JESS expert system, the dynamic fact base is matched with the rule base containing measures such as slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage to generate prevention suggestions. When the deformation trend and prevention suggestions trigger the preset red, orange, yellow, and blue graded early warning thresholds, the graded alarm mechanism is activated, and the disaster simulation path and corresponding prevention strategy suggestions are displayed simultaneously through the three-dimensional visualization platform.

[0006] Preferably, the multi-source monitoring data acquisition and preprocessing process includes: acquiring in real time the absolute coordinate flow of surface displacement of the tunnel entrance slope, the change value of the inclination angle of each depth node of the deep soil, the relative displacement of the cracks in the slope protection skeleton, the small tilt angle of the retaining wall, the pore water pressure and water level, the real-time and cumulative rainfall, and the soil moisture content through the three-dimensional monitoring network; converting the absolute coordinate flow of surface displacement, the change value of the inclination angle of the depth node, the pore water pressure, and the cumulative rainfall into independent evidence sources, and calculating the basic probability allocation value for the judgment result of each independent evidence source on the four risk states of safety, concern, warning, and danger; introducing improved evidence theory to calculate the conflict coefficient between each independent evidence source and identifying the monitoring data with conflict; using the credibility and weighted average rules, correcting the conflict coefficient exceeding the preset threshold, and outputting the multi-source monitoring data after eliminating the conflict.

[0007] Preferably, the process of the long short-term memory network model predicting future deformation trends includes: reconstructing multi-source monitoring data according to time series, extracting deformation feature vectors containing the absolute coordinates of the surface displacement, the inclination change values ​​of each depth node of the deep soil, and the relative displacement of the cracks in the slope protection skeleton, and simultaneously extracting the cumulative rainfall and the soil moisture content as co-feature factors; using the forget gate, input gate, and output gate of the long short-term memory network model, performing memory and forgetting processing on the historical time series features of the deformation feature vector and co-feature factors, and updating the cell state of the long short-term memory network model; based on the cell state, establishing a nonlinear mapping relationship between displacement and time, performing displacement trend extrapolation for the next 24 to 48 hours, and calculating the cumulative displacement and displacement rate at future times; calculating the variation law of the reciprocal of the displacement rate with time, and outputting the future deformation trend based on the rate trend of the reciprocal of the displacement rate approaching zero.

[0008] Preferably, the process of identifying the current evolution stage of the slope includes: using finite element numerical analysis to simulate the displacement of the tunnel entrance slope during the strength reduction process, and establishing a mapping relationship between the evolution of the plastic zone of the displacement cloud map and the evolution stage; When the future deformation trend shows linear growth and the cross-correlation function between displacement and rainfall intensity is within a safe range, the evolution stage is identified as the initial creep stage; when the displacement rate curve shows an inflection point and the tangent angle increases, accompanied by a rise in groundwater level exceeding a preset environmental threshold, the evolution stage is identified as entering the accelerated deformation stage; based on the trend line where the reciprocal of the displacement rate approaches zero, the evolution stage is identified as the critical slip stage.

[0009] Preferably, the process of constructing the dynamic fact base of the JESS expert system includes: mapping the disease type, evolution stage, and causative parameters of the tunnel entrance slope to the input items of the dynamic fact base based on the evaluation results; storing the disease type and the occurrence location using single-string attribute slots, and storing the numerical description parameters using multi-floating-point numerical attribute slots; storing the identifier code array corresponding to the text description parameters generated by the database index using multi-string attribute slots; and aggregating the contents of the single-string attribute slots, the multi-floating-point numerical attribute slots, and the multi-string attribute slots into the dynamic fact base, which serves as the working memory.

[0010] Preferably, the specific steps for generating prevention recommendations include: using the inference engine to perform pattern matching between the disease type, the location of occurrence, and the numerical class description parameters in the dynamic fact base and the triggering conditions of various prevention and control measures in the preset rule base; constructing an inference network containing root nodes, univariate check nodes, and multivariate connection nodes using the Rete algorithm, performing conflict resolution on multiple successfully matched prevention and control measures, and determining the prevention and control instruction with the highest priority; and selecting matching items from the slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage measures according to the severity of the evolution stage to form the prevention recommendations.

[0011] Preferably, the specific steps for activating the graded alarm mechanism and synchronously displaying it through a 3D visualization platform include: comparing the cumulative displacement prediction value and displacement rate prediction value in the future deformation trend with preset four-level warning thresholds (red, orange, yellow, and blue) to determine the real-time monitoring level of the tunnel entrance slope; automatically matching preset prevention measures based on the determined real-time monitoring level to generate standardized emergency procedure instructions including personnel evacuation, traffic control, and material dispatch; synchronously marking the implementation location of prevention recommendations on the 3D visualization platform, and demonstrating the reinforcement effects of slope cutting, slope bottom counter-pressure, rebar and mesh slope protection, and slope top interception and drainage measures through a 3D model; and synchronously pushing the generated warning level, emergency procedure instructions, and prevention strategies to mobile terminals and audible and visual alarms through the early warning cloud platform.

[0012] A tunnel entrance slope early warning system based on multi-source monitoring fusion includes: Three-dimensional monitoring module: Real-time acquisition of multi-source monitoring data through a three-dimensional monitoring network deployed on the slope of the tunnel entrance; Trend prediction module: Input the multi-source monitoring data into the long short-term memory network model, use displacement time series as features and combine rainfall intensity synergy factor, and predict future deformation trends through nonlinear extrapolation calculation; State assessment module: Based on the future deformation trend and combined with the slope failure mode evolution mechanism obtained by finite element numerical analysis, dynamically assess the current evolution stage of the slope and identify whether the slope is in the initial creep stage, the accelerated deformation stage, or the critical slip stage. Fact base construction module: Based on the assessment results, extract disease types, occurrence locations, numerical descriptive parameters, and textual descriptive parameters to construct the dynamic fact base of the JESS expert system; Intelligent Decision Module: Utilizing the inference engine of the JESS expert system, the dynamic fact base is matched with a rule base containing measures such as slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage to generate prevention suggestions. Early warning response module: When the deformation trend and prevention suggestions trigger the preset red, orange, yellow and blue graded early warning thresholds, the graded alarm mechanism is activated, and the disaster simulation path and corresponding prevention strategy suggestions are displayed synchronously through the three-dimensional visualization platform.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention acquires multi-dimensional spatial, mechanical, and environmental data through a three-dimensional monitoring network and incorporates improved evidence theory for preprocessing. By utilizing evidence source transformation and conflict resolution logic, it solves the problems of asynchrony and semantic conflict in heterogeneous sensor data under complex tunnel entrance environments. Compared to traditional single-index early warning, multi-source data fusion improves the accuracy of identifying evolutionary stages such as initial creep, accelerated deformation, and critical slip, effectively reducing the system's false alarm and false negative rates.

[0014] 2. This invention combines finite element method (FEM) numerical analysis with a long short-term memory (LSTM) network model. It utilizes a deep learning model to capture the nonlinear correlation between displacement and rainfall synergy factors, and combines this with the mechanical law that the reciprocal of displacement rate approaches zero, to achieve extrapolation prediction of deformation trends over the next 24 to 48 hours. This technology transcends the threshold-triggered mode of traditional passive monitoring, enabling earlier identification of instability precursors and securing a crucial prevention window for rail transit emergency response.

[0015] 3. This invention utilizes the Rete algorithm to infer prevention and control recommendations. Based on the real-time forecast of the evolution stage and disease characteristics, the system can automatically recommend combined prevention and control strategies such as slope cutting, counterweighting, and drainage, and conduct simulation exercises using a three-dimensional visualization platform. Attached Figure Description

[0016] Figure 1 This is a flowchart of a tunnel entrance slope early warning method based on multi-source monitoring fusion proposed in this invention; Figure 2 This is a flowchart illustrating a tunnel entrance slope early warning method based on multi-source monitoring fusion proposed in this invention. Figure 3 This is a flowchart illustrating the method for identifying the current evolution stage of a slope according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 Please see Figures 1 to 3 This invention provides a method for early warning of tunnel entrance slopes based on multi-source monitoring fusion, the technical solution of which is as follows: A method for early warning of tunnel entrance slopes based on multi-source monitoring fusion, such as Figures 1-2 As shown, it includes: A three-dimensional monitoring network deployed on the slope of the tunnel entrance is used to acquire multi-source monitoring data in real time. The multi-source monitoring data is input into a long short-term memory network model. Using displacement time series as a feature and combining it with rainfall intensity synergy factors, the future deformation trend is predicted through nonlinear extrapolation calculation. Based on the aforementioned future deformation trend, combined with the slope failure mode evolution mechanism obtained from finite element numerical analysis, the current evolution stage of the slope is dynamically assessed, and the slope is identified as being in the initial creep stage, the accelerated deformation stage, or the critical slip stage. Based on the assessment results, disease type, location of occurrence, numerical descriptive parameters and textual descriptive parameters are extracted to construct a dynamic fact base for the JESS expert system. Using the inference engine of the JESS expert system, the dynamic fact base is matched with the rule base containing measures such as slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage to generate prevention suggestions. When the deformation trend and prevention suggestions trigger the preset red, orange, yellow, and blue graded early warning thresholds, the graded alarm mechanism is activated, and the disaster simulation path and corresponding prevention strategy suggestions are displayed simultaneously through the three-dimensional visualization platform.

[0019] Furthermore, the multi-source monitoring data acquisition and preprocessing process includes: acquiring in real time the absolute coordinate flow of surface displacement of the tunnel entrance slope, the inclination angle change values ​​of each depth node of deep soil, the relative displacement of cracks in the slope protection skeleton, the slight tilt angle of the retaining wall, pore water pressure and water level, real-time and cumulative rainfall, and soil moisture content through the three-dimensional monitoring network; converting the absolute coordinate flow of surface displacement, the inclination angle change values ​​of depth nodes, the pore water pressure, and the cumulative rainfall into independent evidence sources, and calculating the basic probability allocation value for the judgment results of the four risk states of safety, concern, warning, and danger for each independent evidence source; introducing improved evidence theory to calculate the conflict coefficient between each independent evidence source and identifying conflicting monitoring data; using credibility and weighted average rules to correct the conflict coefficient exceeding the preset threshold, and outputting multi-source monitoring data after conflict elimination.

[0020] Specifically, the process of calculating the basic probability allocation value includes: establishing the membership function of each independent evidence source in advance through finite element numerical analysis, inputting the physical quantities monitored in real time into the membership function, and obtaining the original probability allocation of the four risk states of safety, concern, warning, and danger.

[0021] The process of introducing and modifying the improved evidence theory includes: assigning an initial weighting factor to each independent evidence source based on the historical measurement accuracy and current operating status of each sensor in the three-dimensional monitoring network; calculating the pairwise conflict degree between each independent evidence source using the distance metric formula in the evidence theory to generate a conflict matrix; calculating the global conflict coefficient based on the conflict matrix; when the global conflict coefficient exceeds a preset threshold, reducing the initial weighting factor of conflicting evidence using the weighted average rule and correspondingly increasing the weight of consistent evidence, and reallocating the basic probability allocation value; specifically, the preset threshold ranges from 0.7 to 0.9. The modified basic probability allocation value is then fused and calculated using the Dempster combination rule to output multi-source monitoring data after conflict elimination.

[0022] This invention enhances the reliability of multi-source monitoring data by introducing improved evidence theory and conflict resolution mechanisms. By constructing membership functions and conflict matrices, the system can quantify the deviations between different sensors and accurately identify anomalous evidence caused by local interference. Combining confidence weight correction and the Dempster combination rule eliminates false alarms caused by environmental temperature drift or local sensor malfunctions, ensuring high consistency and authenticity of the fused monitoring data input into the prediction model, thus laying a solid data foundation for the identification of slope evolution stages.

[0023] Furthermore, the process of the long short-term memory network model predicting future deformation trends includes: reconstructing multi-source monitoring data according to time series, extracting deformation feature vectors containing the absolute coordinates of the surface displacement, the inclination change values ​​of each depth node of the deep soil, and the relative displacement of the cracks in the slope protection skeleton, and simultaneously extracting the cumulative rainfall and the soil moisture content as co-feature factors; using the forget gate, input gate, and output gate of the long short-term memory network model, performing memory and forgetting processing on the historical time series features of the deformation feature vector and co-feature factors, and updating the cell state of the long short-term memory network model; based on the cell state, establishing a nonlinear mapping relationship between displacement and time, performing displacement trend extrapolation for the next 24 to 48 hours, and calculating the cumulative displacement and displacement rate at future times; calculating the variation law of the reciprocal of the displacement rate with time, and outputting the future deformation trend based on the rate trend of the reciprocal of the displacement rate approaching zero.

[0024] Specifically, the process of reconstructing according to the time series includes: using the sliding window technique to extract a historical monitoring sequence of length T, and concatenating the deformation feature vector with the collaborative feature factor in spatial dimension to construct a multidimensional spatiotemporal input matrix.

[0025] The specific parameter configuration for updating the cell state of the Long Short-Term Memory (LSTM) network model includes: setting the number of hidden layers in the LTM network model to 2 to 4, with each layer containing 64 to 128 hidden units, and using a linear rectified function as the activation function to prevent gradient vanishing; during the model training phase, multi-source monitoring data from the historical monitoring period are selected as training samples, the multi-dimensional spatiotemporal input matrix extracted by the sliding window is used as the network input, the cumulative displacement and displacement rate at the corresponding time are used as supervision labels, the mean squared error is used as the loss function, and the network weights are updated through backpropagation and gradient descent algorithms until the validation set error converges; during the online prediction phase, the multi-source monitoring data is used to construct the input matrix in the same way as in the training phase, the network parameters after training are loaded, and the cumulative displacement prediction value and displacement rate prediction value for multiple future time steps are output.

[0026] Furthermore, the process of extrapolating the displacement trend for the next 24 to 48 hours includes: The predicted displacement value output by the Long Short-Term Memory Network model at the current moment is used as the input for the next moment. Combined with a preset future rainfall scenario factor, the extrapolation calculation for continuous time steps is achieved through iterative recursion. A first-order forward difference operation is performed on the predicted continuous cumulative displacement sequence to extract the displacement rate at each step. The reciprocal sequence of the displacement rate is linearly fitted using the least squares method. When the absolute value of the slope of the fitted line exceeds a preset mutation threshold and the intercept approaches the horizontal axis of the time axis, it is determined that the reciprocal of the displacement rate approaches zero, and the future deformation trend is output accordingly. The preset mutation threshold ranges from 1.5 to 3.0.

[0027] This invention achieves high-precision early warning by coupling deep learning models with physical criteria. It utilizes long short-term memory networks to capture the nonlinear correlation between deformation features and rainfall synergy factors, and combines multi-layer hidden layer configuration and recursive multi-step prediction to accurately extrapolate displacement trends for the next 24 to 48 hours. By fitting the reciprocal of the displacement rate using the least squares method, it can scientifically capture the mechanical nodes at which slopes transition from accelerated deformation to critical instability, providing physically supported countdown criteria for emergency response in rail transit.

[0028] Furthermore, such as Figure 3 As shown, the process of identifying the current evolution stage of the slope includes: using finite element numerical analysis to simulate the displacement of the tunnel entrance slope during the strength reduction process, and establishing a mapping relationship between the evolution of the plastic zone of the displacement cloud map and the evolution stage; When the future deformation trend shows linear growth and the cross-correlation function between displacement and rainfall intensity is within a safe range, the evolution stage is identified as the initial creep stage; when the displacement rate curve shows an inflection point and the tangent angle increases, accompanied by a rise in groundwater level exceeding a preset environmental threshold, the evolution stage is identified as entering the accelerated deformation stage; based on the trend line where the reciprocal of the displacement rate approaches zero, the evolution stage is identified as the critical slip stage.

[0029] Specifically, the process of establishing the mapping relationship includes: extracting the evolution process of the plastic zone in the displacement cloud map from local cracking above the tunnel lining to the bulging of the soil in front of the retaining wall corner, until the plastic zone penetrates the entire sliding surface, and defining the strength reduction coefficient corresponding to the penetration of the plastic zone as the initiation threshold of the critical slip period; Furthermore, the specific criteria for identifying the evolutionary stage include: The safe interval is defined as the value of the cross-correlation function being between 0.1 and 0.3, indicating that the displacement response to rainfall input has lag and the amplitude is weak; the increase in tangent angle refers to the real-time tangent angle of the displacement rate curve being greater than 45°, and the preset environmental threshold is defined as the cumulative rise in groundwater level exceeding 500 mm within 24 hours; the reciprocal of the displacement rate is fitted using the least squares method, and when the predicted intercept of the trend line with the horizontal axis of the time axis is less than 24 hours, and the value of the reciprocal of the displacement rate is less than the reciprocal of 0.05 mm / day, it is determined that the critical slip period has been entered.

[0030] This invention solves the technical challenge of ambiguous identification of slope evolution stages by deeply integrating finite element mechanical characteristics with quantitative monitoring indicators. By setting specific cross-correlation function intervals, displacement tangent angles, and groundwater level rise thresholds, and combining this with intercept prediction of the inverse of the displacement rate, it achieves a characterization of the slope from creep to critical slip state. This method not only improves the scientific rigor of the early warning system but also provides a definite triggering basis for the precise deployment of subsequent intelligent prevention strategies.

[0031] Furthermore, the process of constructing the dynamic fact base of the JESS expert system includes: mapping the disease type, evolution stage, and causative parameters of the tunnel entrance slope to the input items of the dynamic fact base based on the evaluation results; storing the disease type and the occurrence location using single-string attribute slots, and storing the numerical description parameters using multi-floating-point numerical attribute slots; storing the identifier code array corresponding to the text description parameters generated by the database index using multi-string attribute slots; and aggregating the contents of the single-string attribute slots, the multi-floating-point numerical attribute slots, and the multi-string attribute slots into the dynamic fact base, which serves as the working memory.

[0032] Specifically, the detailed implementation of constructing the dynamic fact base of the JESS expert system is as follows: In the JESS expert system, a fact template named "SlopeStatus" is predefined. The fact template includes the single-string attribute slot, the multi-floating-point numeric attribute slot, and the multi-string attribute slot. Based on the evolution stage in the evaluation results, the "accelerated deformation period" or "critical slip period" is encapsulated as a state attribute in the single-string attribute slot, and the displacement rate, the reciprocal of the displacement rate, and the rise in groundwater level are encapsulated into the multi-floating-point numerical attribute slot in a preset order. The database index pre-stores the correspondence between disease characteristics and identification codes. By querying the mapping table, unstructured descriptive text is converted into integer identification codes and filled into the multi-string attribute slots in the form of an array, thereby reducing the memory usage of the working memory. Using JESS assertion instructions, the encapsulated fact template instance is injected into the working memory in real time, making it a dynamic fact that the inference engine can call.

[0033] This invention addresses the technical obstacle of directly involving unstructured monitoring data in intelligent decision-making by refining the fact template definition and mapping encapsulation logic of the JESS expert system. It classifies and stores disease characteristics using different types of attribute slots and transforms textual descriptions into identifier code arrays using a database indexing mechanism, improving the retrieval efficiency of the working memory and the inference response speed. This method ensures that key information such as slope evolution stages can be injected into the inference network in a standardized factual form, providing definitive data support for matching prevention recommendations.

[0034] Furthermore, the specific steps for generating prevention recommendations include: using the inference engine to perform pattern matching between the disease type, the location of occurrence, and the numerical class description parameters in the dynamic fact base and the triggering conditions of various prevention and control measures in the preset rule base; constructing an inference network containing root nodes, univariate check nodes, and multivariate connection nodes using the Rete algorithm, performing conflict resolution on multiple successfully matched prevention and control measures, and determining the prevention and control instruction with the highest priority; and selecting matching items from the slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage measures according to the severity of the evolution stage to form the prevention recommendations.

[0035] Specifically, the preset rule base is constructed using IF-THEN logic based on production rules, and the triggering condition corresponds to the value range of the attribute slot in the fact template.

[0036] Furthermore, the process of constructing the inference network and resolving conflicts using the Rete algorithm includes: The univariate check node is set to perform equal-value filtering on the evolution stage and the disease type, forming an Alpha memory block for preliminary screening of prevention and control measures that meet the basic conditions; the multivariate connection node is set to perform a composite logic judgment on the displacement rate and the groundwater level rise, forming a Beta memory block through connection operations to achieve a comprehensive assessment of the severity of the deformation trend. When multiple prevention and control measures simultaneously meet the triggering conditions, the conflict resolution is performed using the priority function P=f(S,E), where S is the risk weight of the evolutionary stage (the critical slip stage has a higher risk weight than the accelerated deformation stage), and E is the degree of influence of the environmental factor. Based on the priority calculation results, the rule at the top of the agenda is activated, and specific engineering parameters are extracted from the slope cutting, slope bottom counter-pressure, rebar and wire mesh slope protection, and slope top interception and drainage measures through the RHS action section, and combined to form the prevention recommendations.

[0037] This invention addresses the technical challenges of low inference efficiency and difficulty in resolving rule conflicts in expert systems when handling multi-source heterogeneous monitoring data by refining the node mapping and conflict resolution logic of the Rete algorithm in slope early warning. It utilizes a hierarchical filtering of monitoring indicators via an Alpha / Beta network, combined with a priority function based on risk weights, to achieve scientific selection of prevention and control measures such as slope cutting and counterweighting. This method improves the system's decision-making response speed under emergency conditions such as critical slippage.

[0038] Furthermore, the specific steps for activating the graded alarm mechanism and synchronously displaying it through the 3D visualization platform include: comparing the cumulative displacement prediction value and displacement rate prediction value in the future deformation trend with preset four-level warning thresholds (red, orange, yellow, and blue) to determine the real-time monitoring level of the tunnel entrance slope; automatically matching preset prevention measures based on the determined real-time monitoring level to generate standardized emergency procedure instructions including personnel evacuation, traffic control, and material dispatch; synchronously marking the implementation location of the prevention recommendations on the 3D visualization platform, and demonstrating the reinforcement effects of the slope cutting, slope bottom counter-pressure, rebar and mesh slope protection, and slope top drainage measures through a 3D model; and synchronously pushing the generated warning level, emergency procedure instructions, and prevention strategies to mobile terminals and audible and visual alarms through the early warning cloud platform.

[0039] Specifically, the process of determining the real-time monitoring level follows the "strictest principle", that is, when the warning level corresponding to the cumulative displacement prediction value and the displacement rate prediction value is inconsistent, the higher level is taken as the final real-time monitoring level.

[0040] Furthermore, the specific implementation methods for displaying and demonstrating on the 3D visualization platform include: Based on the determined real-time monitoring level, a preset color value table is used to color-render the slope model in the 3D visualization platform, with red, orange, yellow, and blue corresponding to visual indicators of instability risk from high to low. A mapping table is established between the prevention recommendations and the 3D component library. When the slope cutting or slope bottom counter-pressure recommendation is generated, the system automatically generates a geometric component model at the corresponding coordinates on the 3D visualization platform through a parametric modeling interface. The reinforcement effect is demonstrated by calling the finite element numerical analysis module to compare the slope stability safety factor before and after implementing the prevention recommendations. The trend of change is observed, and the increment of the slope stability safety factor is superimposed on the side of the three-dimensional model in the form of a dynamic bar chart or contour cloud map; the movement trajectory of the landslide body is simulated using a particle system, and the spatial overlap analysis of the trajectory with the blockade area in the traffic control instruction is performed to intuitively show the impact range of the disaster simulation path on driving safety.

[0041] This invention solves the technical challenges of abstract early warning information and poor intuitiveness of emergency response by refining the "strictest principle" of graded early warning and using dynamic mapping logic of three-dimensional visualization. By leveraging the real-time linkage between parametric modeling and finite element mechanical calculations, it achieves a quantitative demonstration of the reinforcement effect of prevention and control measures, transforming complex mechanical assessment results into perceptible spatial information. This method improves decision-making efficiency and ensures the precise implementation of prevention recommendations and emergency procedures in the spatial dimension.

[0042] This invention enhances the intelligence level of tunnel entrance slope safety monitoring by constructing a closed-loop early warning system encompassing "perception-prediction-judgment-prevention-contingency planning." By deeply integrating a three-dimensional monitoring network with a long short-term memory network model, it achieves a leap from single-point passive monitoring to multi-dimensional proactive trend prediction, ensuring the early identification of disaster precursors. Simultaneously, the automated generation of prevention and control suggestions is achieved through the JESS expert system and the Rete algorithm, combined with a three-dimensional visualization platform to intuitively display the decision-making path. This effectively solves the technical challenge of the disconnect between monitoring data and emergency response, enhancing the scientific rigor and collaborative efficiency of disaster prevention and mitigation in rail transit.

[0043] Example 2 Based on the tunnel entrance slope early warning method based on multi-source monitoring fusion described in Example 1, the following describes the detailed workflow of a tunnel entrance slope early warning system based on multi-source monitoring fusion through a specific application scenario of a tunnel entrance slope on a rail transit line under heavy rainfall conditions: During periods of heavy rainfall, a three-dimensional monitoring network deployed on the slope of the tunnel entrance acquires multi-source monitoring data in real time. The cumulative rainfall reached 150 mm, and the groundwater level rose by 550 mm cumulatively within 24 hours. At this point, the cumulative rainfall, pore water pressure, and absolute coordinate flow of surface displacement were converted into independent evidence sources. Due to hysteresis interference caused by local disturbances in the deep soil sensors, the conflict coefficient between the deep soil sensors and the surface displacement evidence source was calculated to be 0.75 using an improved evidence theory, exceeding a preset threshold. Subsequently, the initial weighting factor of the conflicting evidence was reduced using credibility and weighted averaging rules, and the conflict-free multi-source monitoring data was output. The conflict-free data is reconstructed according to a time series, and a historical monitoring sequence of length T is extracted using a sliding window technique to construct a multi-dimensional spatiotemporal input matrix, which is then input into a Long Short-Term Memory (LSTM) network model. The LTM network model uses forgetting, input, and output gate logic to process the historical temporal features of deformation feature vectors and cooperating feature factors, updating the cell states of the LTM network model. The displacement trend extrapolation for the next 24 to 48 hours is performed iteratively to calculate the cumulative displacement and displacement rate at future times. The reciprocal sequence of the displacement rate is linearly fitted using the least squares method. In this embodiment, the absolute slope of the fitted line is identified as 2.5, exceeding a preset mutation threshold, and the predicted intercept is less than 24 hours, indicating that the reciprocal of the displacement rate is close to 0. This embodiment combines the slope failure mode evolution mechanism obtained from finite element numerical analysis to perform state identification. Because the displacement rate curve shows an inflection point and the real-time tangent angle increases to 55°, far exceeding the 45° judgment standard, and accompanied by a groundwater level rise exceeding a preset environmental threshold of 500 mm, this embodiment identifies the evolution stage as transitioning from the initial creep stage to the accelerated deformation stage. Based on the evaluation results, this embodiment begins to construct the dynamic fact base of the JESS expert system. In the JESS expert system, a fact template named "SlopeStatus" is invoked to encapsulate the identified "accelerated deformation period" as a status attribute in a single-string attribute slot, and the displacement rate and its reciprocal are encapsulated in a multi-floating-point numerical attribute slot. By querying the database index mapping table, textual descriptions such as "rainfall-induced accelerated deformation" are converted into integer identifiers and filled into multi-string attribute slots. Assertion instructions are then used to inject the encapsulated fact instances into the working memory in real time. The inference engine initiates the Rete algorithm to construct an inference network, performs equal-value filtering of the evolution stage and the disease type in the Alpha network, and uses the Beta network to perform composite logic judgment on the displacement rate and the groundwater level rise. When multiple prevention and control measures simultaneously meet the triggering conditions, the system uses a priority function to perform conflict resolution. Since this is during the accelerated deformation period and rainfall is continuous, this embodiment determines the highest priority prevention and control instruction to be the combination of slope bottom surcharge counterpressure and slope top interception and drainage measures, thereby forming the prevention recommendation.

[0044] In this embodiment, the cumulative displacement prediction value in the future deformation trend is compared with the preset four-level early warning threshold, and the real-time monitoring level is determined to be an orange warning based on the "strictest principle". On the 3D visualization platform, the system calls the color value table to render the slope model in orange, and automatically generates the geometric component model of the surcharge counterpressure at the corresponding coordinates at the bottom of the slope through the parametric modeling interface. By calling the finite element numerical analysis module, the platform dynamically demonstrates the increasing trend of the slope stability safety factor after implementing counterpressure reinforcement. Finally, the system pushes the generated early warning level, standardized emergency procedure instructions, and the prevention strategy to the mobile terminal through the early warning cloud platform.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A tunnel portal slope pre-warning method based on multi-source monitoring fusion, characterized in that, include: A three-dimensional monitoring network deployed on the slope of the tunnel entrance is used to acquire multi-source monitoring data in real time. The multi-source monitoring data is input into a long short-term memory network model. Using displacement time series as a feature and combining it with rainfall intensity synergy factors, the future deformation trend is predicted through nonlinear extrapolation calculation. Based on the aforementioned future deformation trend, combined with the slope failure mode evolution mechanism obtained from finite element numerical analysis, the current evolution stage of the slope is dynamically assessed, and the slope is identified as being in the initial creep stage, the accelerated deformation stage, or the critical slip stage. Based on the assessment results, disease type, location of occurrence, numerical descriptive parameters and textual descriptive parameters are extracted to construct a dynamic fact base for the JESS expert system. Using the inference engine of the JESS expert system, the dynamic fact base is matched with the rule base containing measures such as slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage to generate prevention suggestions. When the deformation trend and prevention suggestions trigger the preset red, orange, yellow, and blue graded early warning thresholds, the graded alarm mechanism is activated, and the disaster simulation path and corresponding prevention strategy suggestions are displayed simultaneously through the three-dimensional visualization platform. 2.The tunnel portal slope pre-warning method based on multi-source monitoring fusion of claim 1, wherein The process of multi-source monitoring data acquisition and preprocessing includes: acquiring in real time the absolute coordinate flow of surface displacement of the tunnel entrance slope, the change value of the inclination angle of each depth node of the deep soil, the relative displacement of the cracks in the slope protection skeleton, the small tilt angle of the retaining wall, the pore water pressure and water level, the real-time and cumulative rainfall, and the soil moisture content through the three-dimensional monitoring network; converting the absolute coordinate flow of surface displacement, the change value of the inclination angle of the depth node, the pore water pressure, and the cumulative rainfall into independent evidence sources, and calculating the basic probability allocation value for the judgment result of each independent evidence source on the four risk states of safety, concern, warning, and danger; introducing improved evidence theory to calculate the conflict coefficient between each independent evidence source and identifying conflicting monitoring data; using the credibility and weighted average rules, correcting the conflict coefficient exceeding the preset threshold, and outputting the multi-source monitoring data after eliminating the conflict. 3.The tunnel portal slope pre-warning method based on multi-source monitoring fusion of claim 2, characterized in that, The process of predicting future deformation trends using the Long Short-Term Memory (LSTM) network model includes: reconstructing multi-source monitoring data according to time series; extracting deformation feature vectors containing the absolute coordinates of the surface displacement, the inclination changes of each depth node in the deep soil, and the relative displacement of the cracks in the slope protection skeleton; simultaneously extracting the cumulative rainfall and the soil moisture content as co-feature factors; using the forget gate, input gate, and output gate of the LTM network model to perform memory and forgetting processing on the historical time series features of the deformation feature vector and co-feature factors, and updating the cell state of the LTM network model; based on the cell state, establishing a nonlinear mapping relationship between displacement and time, performing displacement trend extrapolation for the next 24 to 48 hours, and calculating the cumulative displacement and displacement rate at future times; calculating the variation law of the reciprocal of the displacement rate with time, and outputting the future deformation trend based on the rate trend of the reciprocal of the displacement rate approaching zero.

4. The tunnel portal slope early warning method based on multi-source monitoring fusion according to claim 3, characterized in that, The process of identifying the current evolution stage of the slope includes: using finite element numerical analysis to simulate the displacement of the tunnel entrance slope during the strength reduction process, and establishing a mapping relationship between the evolution of the plastic zone of the displacement cloud map and the evolution stage; When the future deformation trend shows linear growth and the cross-correlation function between displacement and rainfall intensity is within a safe range, the evolution stage is identified as the initial creep stage; when the displacement rate curve shows an inflection point and the tangent angle increases, accompanied by a rise in groundwater level exceeding a preset environmental threshold, the evolution stage is identified as entering the accelerated deformation stage; based on the trend line where the reciprocal of the displacement rate approaches zero, the evolution stage is identified as the critical slip stage.

5. The tunnel portal slope early warning method based on multi-source monitoring fusion according to claim 4, characterized in that, The process of constructing the dynamic fact base of the JESS expert system includes: mapping the disease type, evolution stage, and causative parameters of the tunnel entrance slope to the input items of the dynamic fact base based on the evaluation results; storing the disease type and occurrence location using single-string attribute slots, and storing the numerical description parameters using multi-floating-point numerical attribute slots; storing the identifier code array corresponding to the text description parameters generated by the database index using multi-string attribute slots; and aggregating the contents of the single-string attribute slots, the multi-floating-point numerical attribute slots, and the multi-string attribute slots into the dynamic fact base, which serves as the working memory.

6. The tunnel portal slope pre-warning method based on multi-source monitoring fusion according to claim 5, characterized in that, The specific steps for generating prevention recommendations include: using the inference engine to perform pattern matching between the disease type, occurrence location, and numerical description parameters in the dynamic fact base and the triggering conditions of various prevention and control measures in the preset rule base; constructing an inference network containing root nodes, univariate check nodes, and multivariate connection nodes using the Rete algorithm, performing conflict resolution on multiple successfully matched prevention and control measures, and determining the prevention and control instruction with the highest priority; and selecting matching items from the slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage measures according to the severity of the evolution stage to form the prevention recommendations.

7. The tunnel entrance slope early warning method based on multi-source monitoring fusion according to claim 6, characterized in that, The specific steps for activating the graded alarm mechanism and synchronously displaying it through a 3D visualization platform include: comparing the cumulative displacement prediction value and displacement rate prediction value in the future deformation trend with preset four-level warning thresholds (red, orange, yellow, and blue) to determine the real-time monitoring level of the tunnel entrance slope; automatically matching preset prevention measures based on the determined real-time monitoring level to generate standardized emergency procedure instructions including personnel evacuation, traffic control, and material dispatch; synchronously marking the implementation location of prevention recommendations on the 3D visualization platform, and demonstrating the reinforcement effects of slope cutting, slope bottom counter-pressure, rebar and mesh slope protection, and slope top drainage measures through a 3D model; and synchronously pushing the generated warning level, emergency procedure instructions, and prevention strategies to mobile terminals and audible and visual alarms through the early warning cloud platform.

8. A tunnel entrance slope early warning system based on multi-source monitoring fusion, characterized in that, include: Three-dimensional monitoring module: Real-time acquisition of multi-source monitoring data through a three-dimensional monitoring network deployed on the slope of the tunnel entrance; Trend prediction module: Input the multi-source monitoring data into the long short-term memory network model, use displacement time series as features and combine rainfall intensity synergy factor, and predict future deformation trends through nonlinear extrapolation calculation; State assessment module: Based on the future deformation trend and combined with the slope failure mode evolution mechanism obtained by finite element numerical analysis, dynamically assess the current evolution stage of the slope and identify whether the slope is in the initial creep stage, the accelerated deformation stage or the critical slip stage. Fact base construction module: Based on the assessment results, extract disease types, occurrence locations, numerical descriptive parameters, and textual descriptive parameters to construct the dynamic fact base of the JESS expert system; Intelligent Decision Module: Utilizing the inference engine of the JESS expert system, the dynamic fact base is matched with a rule base containing measures such as slope cutting, slope bottom surcharge counterpressure, rebar and wire mesh slope protection, and slope top interception and drainage to generate prevention suggestions. Early warning response module: When the deformation trend and prevention suggestions trigger the preset red, orange, yellow and blue graded early warning thresholds, the graded alarm mechanism is activated, and the disaster simulation path and corresponding prevention strategy suggestions are displayed synchronously through the three-dimensional visualization platform.

Images