Intelligent early warning method and system for high slope disasters based on cloud computing, medium and product

Abstract

The invention discloses an intelligent early warning method and system for high slope disasters based on cloud computing, a medium and a product, and relates to the technical field of disaster early warning. The method comprises the following steps: constructing a case evolution graph at a cloud based on a historical landslide case database; matching the multi-dimensional real-time feature vector of the target slope with nodes in a case evolution graph, determining a target node with the highest similarity, searching and generating a plurality of first potential evolution paths leading to a known instability case in the case evolution graph, and evaluating the risk level of each first potential evolution path; early warning information is generated according to the risk level, virtual engineering intervention measures are simulated and applied according to the early warning information, the modification effect of the virtual engineering intervention measures on the case evolution graph is evaluated, and the relieving effect of each intervention strategy is evaluated based on the modification effect; and selecting an optimal intervention strategy for deployment according to the relieving effect. By implementing the technical scheme, high slope disasters can be early warned accurately.

Description

Technical Field

[0001] This application relates to the field of disaster early warning, and in particular to a cloud computing-based intelligent early warning method, system, medium and product for high slope disasters. Background Technology

[0002] With rapid economic development and continuous infrastructure construction, numerous high slope projects have emerged, including roads, bridges, and water conservancy projects. The stability of high slopes is crucial not only for the safety and normal operation of these projects, but also for the accurate and timely early warning of high slope disasters, which can effectively reduce losses.

[0003] Traditional methods for early warning of high slope disasters typically include on-site monitoring and experience-based judgment. On-site monitoring mainly involves deploying various sensors on the high slope, such as displacement sensors, stress sensors, and piezometers, to acquire real-time data on changes in the slope's physical parameters. Staff regularly collect sensor data on-site and assess slope stability based on trends in the data. Experience-based judgment relies on the extensive experience of engineers and experts, combined with factors such as on-site geological conditions, topography, and meteorological conditions, to evaluate the stability of the high slope. Furthermore, it involves referencing past cases of similar projects to compare the current condition of the high slope and make early warning decisions.

[0004] However, on-site monitoring methods rely on the accuracy of sensors and the real-time nature of data. Sensor malfunctions or data transmission interruptions can lead to inaccurate or delayed early warning information. Furthermore, sensors can only acquire limited physical parameters, making it difficult to comprehensively reflect the complex state of high slopes. Experience-based judgment methods, on the other hand, are subjective; different engineers and experts may reach different conclusions due to differences in experience and judgment criteria. This lack of scientific quantitative analysis and systematic evaluation makes it impossible to accurately and promptly provide early warnings of high slope hazards. Summary of the Invention

[0005] This application provides a cloud-based intelligent early warning method, system, medium, and product for high slope disasters, which can accurately predict high slope disasters, evaluate the mitigation effect of intervention strategies, select the optimal strategy for deployment, and continuously update the case evolution map to improve the accuracy of early warning and intervention.

[0006] In the first aspect, this application provides a cloud computing-based intelligent early warning method for high slope disasters. The method includes: constructing a case evolution map in the cloud based on a historical landslide case database, wherein the nodes in the case evolution map are historical normal cases and historical landslide cases represented by multi-dimensional feature vectors, the directed edges between nodes represent the evolution path between cases, and the weight of the directed edges is assigned based on the feature similarity and spatiotemporal correlation between cases. The real-time monitoring data of the target slope is quantified into a multi-dimensional real-time feature vector. The multi-dimensional real-time feature vector is matched with the nodes in the case evolution map to determine the target node with the highest similarity. Starting from the target node, multiple first potential evolution paths leading to known instability cases are searched and generated in the case evolution map, and the risk level of each first potential evolution path is evaluated. When a risk level exceeds a preset threshold, an early warning is generated based on the risk level. In the digital twin environment of the target slope, virtual engineering intervention measures are simulated and applied based on the early warning information. The modification effect of the virtual engineering intervention measures on the case evolution map is evaluated, and the mitigation effect of each intervention strategy is evaluated based on the modification effect. Based on the mitigation effect, the optimal intervention strategy is selected and deployed, and the actual slope state evolution sequence and the effect feedback of the actual intervention measures are used as new nodes and added to the case evolution map.

[0007] By employing the aforementioned technical solutions, and through intelligent matching and path searching of real-time monitoring data with evolutionary maps constructed based on historical cases, the potential paths and risk levels of slopes evolving towards instability can be dynamically assessed. This represents a leap from static threshold alarms to dynamic trend prediction, significantly improving the accuracy and foresight of early warnings. The introduction of digital twin technology allows for the pre-simulation and evaluation of the effects of different engineering interventions in a virtual environment, thereby assisting in the selection of the optimal treatment plan. This represents a shift from experience-driven to simulation-driven decision-making, enhancing the scientific rigor and effectiveness of risk response measures. A closed-loop learning feedback mechanism has been established, feeding real slope state evolution and intervention effect results into the evolutionary map as new case knowledge. This enables the system to continuously accumulate experience, self-improve, and possess dynamic growth and continuous optimization capabilities. Relying on a cloud computing platform, a series of functions, including multi-source data fusion, risk assessment, simulation, and strategy optimization, have been integrated, achieving fully automated and intelligent management of the entire process from monitoring and early warning to decision-making and feedback. This greatly improves the efficiency and intelligence level of high slope safety operation and maintenance.

[0008] In some embodiments, quantifying the real-time monitoring data of the target slope into a multi-dimensional real-time feature vector, and matching the multi-dimensional real-time feature vector with nodes in the case evolution map to determine the target node with the highest similarity, specifically includes: Multiple feature factors corresponding to the historical landslide case database are extracted from the real-time monitoring data. These feature factors include landslide pattern, sliding surface angle, sliding surface grade, angle between the main structural plane and the slope, rock mass quality index, slope angle, unloading loosening zone depth, groundwater level, and rainfall. According to the preset quantification rules, each feature factor is mapped to a numerical value or category identifier, and then combined in a predetermined dimensional order to generate the multidimensional real-time feature vector of the target slope. Calculate the weighted Euclidean distance and cosine similarity between the multidimensional real-time feature vector and the multidimensional feature vector of each node in the case evolution graph; The weighted Euclidean distance and the cosine similarity are combined using a preset fusion function to obtain a comprehensive matching score; The node with the highest comprehensive matching score is determined as the target node that is most similar to the current state of the target slope.

[0009] By employing the aforementioned technical solution, multiple key geological and working condition characteristics corresponding to the case library are extracted from real-time data and quantified into multi-dimensional feature vectors in a unified format according to preset rules. This transforms complex and heterogeneous field monitoring data into structured data that the system can recognize and compute, laying the foundation for subsequent accurate comparisons. A comprehensive matching algorithm combining weighted Euclidean distance and cosine similarity is used, which can measure both the absolute numerical differences of feature vectors and assess the similarity of their spatial distribution. This dual-measurement mechanism overcomes the limitations of a single similarity index, making the judgment of the current slope state more accurate and reliable. A preset fusion function combines the results of the two similarity measurements, and the target node is determined based on the principle of the highest comprehensive score. This allows the system to intelligently and objectively select historical scenarios most similar to the current target slope state from a vast number of historical cases, providing a highly relevant and reliable starting point for subsequent risk simulation. By quickly and accurately linking the real-time perceived slope condition with historical experience accumulated in the case library, the system no longer views real-time data in isolation, but interprets it within a rich knowledge network composed of historical cases, greatly enhancing the depth and insight of the condition assessment.

[0010] In some embodiments, the step of searching and generating multiple first potential evolutionary paths leading to known instability cases in the case evolution graph, starting from the target node, specifically includes: Starting from the target node, search all reachable subsequent nodes along the directed edge direction to form multiple first paths. Starting from all known unstable case nodes, search all possible predecessor nodes in the reverse directed edge direction to form multiple second paths. Maintain a corresponding path state tuple for each search path. The path state tuple is used to record the cumulative transition probability product of the path and the number of path nodes. The search path is either the first path or the second path. If the cumulative transition probability product of the search path is lower than a preset probability threshold, or if the number of path nodes of the search path exceeds a preset maximum path depth, then the search path is terminated. When the first path and the second path have overlapping nodes, the first path and the second path are spliced ​​together to form a first potential evolutionary path, and the first potential evolutionary paths are sorted from largest to smallest according to the cumulative transition probability product.

[0011] By employing the aforementioned technical solution and combining a forward search strategy from the target node with a reverse tracing strategy from the instability case node, the system can more comprehensively and proactively explore all possible paths for the slope to evolve from its current state to a failure state, achieving proactive discovery and forward-looking prediction of potential risks. By setting a cumulative transition probability threshold and a maximum path depth as search termination conditions, ineffective searches for infinitely deep or extremely low-probability paths within the graph are effectively avoided. This ensures that computational resources are concentrated on high-probability, short-path key risk evolution scenarios, improving the computational efficiency and engineering practicality of the search process. Through the splicing of bidirectional paths, a first potential evolution path can be systematically constructed. Subsequently, these paths are sorted according to their cumulative transition probabilities, allowing the system to automatically select the core risk paths with the highest probability of occurrence, providing a clear and focused basis for subsequent risk assessment and decision-making.

[0012] In some embodiments, assessing the risk level of each of the first potential evolutionary paths includes: Based on the quantification results of rock mass quality indicators, unloading loosening zone depth and slope angle in the multidimensional feature vectors of each node on the first potential evolution path, the local instability potential energy of each node is dynamically calculated. Based on the historical fluctuations of the directed edge weights in the first potential evolution path and the differences in the multidimensional feature vectors of adjacent nodes, the transition vulnerability of each directed edge is calculated. The local instability potential energy and the transfer vulnerability are used as inputs and sequentially input into a preset risk transmission chain model. The risk transmission chain model multiplies the output potential energy of the previous node with the transfer vulnerability of the current edge to obtain the input potential energy transmitted to the next node. The final output potential energy of the risk transmission chain model is defined as the comprehensive risk index of the first potential evolution path, and the comprehensive risk index is mapped to the corresponding risk level according to a preset numerical range.

[0013] By employing the aforementioned technical solution, and comprehensively considering the local instability potential energy of nodes along the path (reflecting static geological conditions) and the transfer vulnerability of edges (reflecting the difficulty of dynamic evolution), the originally qualitative and vague risk assessment is transformed into quantitative indicators based on clear physical meaning and statistical data, making risk assessment more objective and accurate. A risk transmission chain model is introduced to simulate the dynamic process of risk propagation from one node to the next along the path and its potential amplification. This not only identifies high-risk nodes but, more importantly, enables risk tracking throughout the entire disaster evolution process. Local instability potential energy is dynamically calculated based on real-time quantitative characteristics of nodes, while transfer vulnerability considers historical fluctuations and state differences. This allows risk assessment to closely integrate with the current specific state of the slope and its evolutionary context, rather than relying on static thresholds, significantly improving the dynamic adaptability and contextual relevance of risk warnings.

[0014] In some embodiments, simulating the application of virtual engineering intervention measures based on the early warning information includes: The key instability factor with the highest contribution is identified from the multidimensional feature vectors of the nodes of the first potential evolution path associated with the warning information using principal component analysis. Based on the key instability factors, engineering measures with inhibitory relationships are retrieved from the preset intervention measure knowledge graph. A virtual intervention action vector with three dimensions, namely support resistance, drainage efficiency, and slope adjustment, is constructed according to the engineering measures. The intensity of each dimension is weighted according to the contribution of the key instability factors. In the digital twin model, the virtual intervention vector is used as the boundary condition to calculate the stress field redistribution, displacement field evolution and dynamic changes of potential slip surfaces of the target slope after intervention, and a snapshot of the state after intervention is generated based on the calculation results.

[0015] By employing the aforementioned technical solution, principal component analysis is used to identify the key instability factors with the highest contribution from numerous characteristic factors, enabling interventions to directly address the main contradictions leading to instability. Based on this, measures with specific inhibitory relationships are retrieved from a knowledge graph, ensuring the highly targeted nature of the constructed virtual intervention plan. In the digital twin model, virtual interventions are applied in vector form, and numerical calculations are used to solve for the stress field, displacement field, and dynamic response of the potential slip surface of the slope after intervention. This is equivalent to conducting a comprehensive and visualized simulation before actual construction, allowing decision-makers to intuitively and quantitatively predict the potential effects of different intervention strategies. The constructed virtual intervention action vector covers the three core engineering dimensions of slope reinforcement: support resistance, drainage efficiency, and slope adjustment. This method encourages the generation of comprehensive treatment plans, rather than single measures, enabling a more systematic and comprehensive approach to complex slope instability problems and improving the rationality and effectiveness of the overall intervention strategy.

[0016] In some embodiments, evaluating the effect of the virtual engineering intervention on the modification of the case evolution map includes: The features in the post-intervention state snapshot are mapped to virtual state nodes in the case evolution graph. Starting from the virtual state nodes, all second potential evolutionary paths leading to known unstable cases are searched again in the case evolution graph. The first potential evolutionary path is compared with the second potential evolutionary path, and the original high-risk path that does not appear in the second potential evolutionary path is identified as the high-risk path that is cut off. The path cutting rate is calculated as the proportion of the number of high-risk paths that have been cut off to the total number of paths in the first potential evolution path. The average comprehensive risk index of all paths in the second potential evolution path is calculated and compared with the average comprehensive risk index of the first potential evolution path to obtain the risk decay degree. Based on the path interruption rate and the risk attenuation, the topology reconstruction intensity value of the virtual engineering intervention measures on the case evolution map is obtained by weighted calculation. The topology reconstruction intensity value is used to characterize the intensity of the modification effect.

[0017] The aforementioned technical solution not only focuses on changes in local conditions but also assesses the intervention's effectiveness at the topological level of the entire case evolution map. By identifying severed high-risk paths, it reveals how intervention measures fundamentally alter the potential evolutionary pattern of the disaster, disrupting the key transmission chain from the current state to instability, thus achieving an effective assessment of the macroscopic and structural effects of the intervention measures. A comprehensive topological reconstruction intensity value is obtained by weighted fusion of the path interruption rate (reflecting the reduction of risky paths) and the risk decay degree (reflecting the risk level of remaining paths). This indicator avoids the limitations of single evaluations and comprehensively measures the combined effect of intervention measures in both reducing risk value and interrupting risky paths.

[0018] In some embodiments, evaluating the mitigation effect of each intervention strategy based on the modification includes: Each intervention strategy is compared with predefined engineering constraints, and intervention strategies that meet the engineering constraints are selected to form a set of feasible strategies. The engineering constraints include an implementation cost threshold and a construction period upper limit. For each intervention strategy in the set of feasible strategies, the topology reconstruction intensity values ​​of all the virtual engineering intervention measures included are weighted and summed to calculate the comprehensive mitigation effectiveness score of the intervention strategy, and the intervention strategy with the highest comprehensive mitigation effectiveness score is selected as the optimal intervention strategy. The optimal intervention strategy is pushed to the engineering management terminal, and a monitoring task is started simultaneously to track the slope status feedback data after the optimal intervention strategy is actually deployed.

[0019] By employing the aforementioned technical solution, and comparing the intervention strategy with predefined engineering constraints such as implementation cost thresholds and construction period limits, it is ensured that the selected solution is not only technically effective but also economically and temporally feasible. This avoids the dilemma of selecting the optimal technical solution but being unable to implement the project, making the decision-making results more practically instructive. For multiple virtual engineering intervention measures included in an intervention strategy, the overall mitigation effectiveness score of the strategy is calculated by weighted summation of their respective topology reconstruction intensity values. This method can quantitatively evaluate the overall effect of composite intervention measures, reflecting the synergistic effect between different measures, and providing a scientific basis for selecting the optimal comprehensive governance solution.

[0020] In a second aspect, embodiments of this application provide a computer system including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the method described in any possible implementation of the first aspect.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0022] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0023] It is understood that the computer system provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By constructing a case evolution map, historical landslide cases and normal cases are linked in the form of multi-dimensional feature vectors, forming a disaster evolution knowledge network with spatiotemporal correlation. This enables the system to proactively deduce multiple potential paths for the current slope state to develop into an unstable state, breaking through the limitations of traditional single-threshold early warning and realizing dynamic simulation of the disaster evolution process and risk path identification; 2. It not only matches the current state with historical cases but also further assesses the risk level of each potential evolutionary path, thus identifying high-risk paths in the long-term evolutionary trend even before obvious signs of slope instability appear. This assessment mechanism based on path probability and state transmission significantly improves the scientific rigor and lead time of early warnings. 3. By simulating the application of virtual engineering interventions in a digital twin environment, the modifying effect of various measures on the case evolution graph can be quantitatively analyzed, such as cutting off high-risk paths and reducing the risk level of paths. This allows users to simulate and compare the mitigation effects of different intervention strategies before implementing actual engineering projects, thereby selecting the optimal solution and improving the reliability and economy of engineering decisions; 4. The actual slope condition evolution and the effects of intervention measures are fed back as new nodes, continuously enriching the case evolution map. This mechanism enables the system to learn and improve online: as data accumulates, the coverage and prediction accuracy of the map continuously improve, forming a disaster early warning and decision support system that becomes increasingly intelligent with use. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a cloud-based intelligent early warning method for high slope disasters in an embodiment of this application. Figure 2 This is a schematic diagram illustrating the process of simulating the application of virtual engineering intervention measures in the embodiments of this application; Figure 3 This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application. Detailed Implementation

[0026] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0028] The following is combined with Figure 1 The method of the embodiments of this application will be described below.

[0029] Figure 1 This is a flowchart illustrating a cloud-based intelligent early warning method for high slope disasters, as described in an embodiment of this application. Figure 1 As shown, the method includes the following steps: S101. Based on the historical landslide case database, a case evolution graph is constructed in the cloud. The nodes in the case evolution graph are historical normal cases and historical landslide cases represented by multi-dimensional feature vectors. The directed edges between nodes represent the evolution path between cases. The weight of the directed edges is assigned based on the feature similarity and spatiotemporal correlation between cases. S102. Quantify the real-time monitoring data of the target slope into a multi-dimensional real-time feature vector, match the multi-dimensional real-time feature vector with the nodes in the case evolution map, determine the target node with the highest similarity, and search and generate multiple first potential evolution paths leading to known instability cases in the case evolution map starting from the target node, and evaluate the risk level of each first potential evolution path. S103. When the risk level is greater than a preset threshold, an early warning message is generated based on the risk level. In the digital twin environment of the target slope, virtual engineering intervention measures are simulated and applied based on the early warning message. The modification effect of the virtual engineering intervention measures on the case evolution map is evaluated. The mitigation effect of each intervention strategy is evaluated based on the modification effect. S104. Based on the mitigation effect, select the optimal intervention strategy for deployment, and add the actual slope state evolution sequence and the effect feedback of the actual intervention measures as new nodes to the case evolution map.

[0030] Through engineering case studies and literature searches, a structured database of historical landslide cases was constructed. Each case (whether a stable slope or an unstable landslide) was meticulously recorded, detailing nine core characteristic factors: landslide pattern, slip surface angle, slip surface grade, angle between the main structural plane and the slope, rock quality index (RQD), slope angle, unloading loosened zone depth, groundwater level, and rainfall. To achieve quantitative description and comparability of landslide cases, these nine characteristic factors underwent standardized quantification. For different types of data (e.g., continuous values, grades, categories), normalization, grade assignment, and one-hot encoding methods were employed to uniformly map all characteristic values ​​to comparable numerical ranges. The processed nine values ​​were combined in a predetermined order, transforming each case into a standardized 9-dimensional feature vector. In the cloud-based map, each such multi-dimensional feature vector constitutes a node, categorized into two types: historical normal cases representing stable states and historical landslide cases representing failed states. Directed edges between nodes: These represent the potential evolutionary path from one slope state (node) to another. For example, a node pointing from a "stable state with a certain combination of features" to a node pointing from a "critical state where parameters deteriorate due to rainfall." Weight assignment: The weights of the edges are not fixed; they comprehensively reflect the probability of evolution and the strength of the correlation between states. Weights are assigned based on feature similarity and spatiotemporal correlation. Feature similarity: The more similar the feature vectors of two nodes, the higher the probability that there may be an evolutionary relationship between them. Spatiotemporal correlation: If cases have a historical temporal sequence (such as a monitoring data sequence) or spatial proximity (such as different stages of a slope), the weights of the directed edges between them will be enhanced.

[0031] Nine feature data points corresponding to the case library are extracted from the real-time monitoring system of the target slope (such as deep-hole inclinometers and ground-based radar mentioned in the document). Following the same quantization rules, a multi-dimensional real-time feature vector of the target slope is generated. This real-time vector is then matched with the vectors of all historical nodes in the graph to find the target node with the highest similarity. The target node represents the historical state most similar to the current state of the target slope. Starting from this target node, all paths leading to known instability case nodes are searched along directed edges in the graph; these paths are the first potential evolution paths. For each found potential path, the system assesses its risk level. The assessment criteria may include: the severity of geological conditions at each node along the path, the weight of the directed edges traversed by the path, and the path length (the steps required to evolve to instability).

[0032] When the risk level of a potential evolution path exceeds a preset threshold, the system automatically generates an early warning and triggers a digital twin process. In the digital twin environment of the target slope (i.e., a virtual slope model built based on real data), the system simulates virtual engineering interventions based on the risk path characteristics associated with the early warning information. For example, for paths with low drainage efficiency, it simulates adding drainage holes; for paths with insufficient support, it simulates adding anchor bolts. These interventions change the slope's state, which is reflected in the digital twin as a snapshot of the post-intervention state. The system maps this new state back to the case evolution map and observes the changes. For example, whether the previously high-risk first potential evolution path has been cut off, and whether the overall risk level of the path has decreased. The system quantifies the effects of these modifications, calculating indicators such as path cutoff rate and risk attenuation, and then comprehensively evaluates the mitigation effect of each intervention strategy. The system compares the mitigation effects of all virtual intervention strategies, and, combined with engineering constraints (such as implementation cost and construction period), automatically selects the optimal intervention strategy. This strategy is then pushed to engineering managers for actual deployment. The system will continuously monitor the actual slope condition evolution sequence and the effectiveness of actual intervention measures. This real, new data (including successful or unsuccessful experiences) will be added as a new node to the case evolution map.

[0033] In some embodiments, quantifying the real-time monitoring data of the target slope into a multi-dimensional real-time feature vector, and matching the multi-dimensional real-time feature vector with nodes in the case evolution map to determine the target node with the highest similarity, specifically includes: Multiple feature factors corresponding to the historical landslide case database are extracted from the real-time monitoring data. These feature factors include landslide pattern, sliding surface angle, sliding surface grade, angle between the main structural plane and the slope, rock mass quality index, slope angle, unloading loosening zone depth, groundwater level, and rainfall. According to the preset quantification rules, each feature factor is mapped to a numerical value or category identifier, and then combined in a predetermined dimensional order to generate the multidimensional real-time feature vector of the target slope. Calculate the weighted Euclidean distance and cosine similarity between the multidimensional real-time feature vector and the multidimensional feature vector of each node in the case evolution graph; The weighted Euclidean distance and the cosine similarity are combined using a preset fusion function to obtain a comprehensive matching score; The node with the highest comprehensive matching score is determined as the target node that is most similar to the current state of the target slope.

[0034] Nine core characteristic factors, completely consistent with the definitions in the historical case database, were identified and extracted from the real-time monitoring data stream of the target slope. These data originated from various sensors deployed on the slope (such as deep-hole inclinometers, groundwater pressure gauges, and rain gauges) and preliminary geological survey data. Landslide pattern: identified through geological models and real-time deformation data. Slip surface angle and slope angle: obtained through mapping and 3D modeling. Rock mass quality indicators and unloading loosened zone depth: derived from geological survey reports, serving as relatively static initial data. Groundwater level and rainfall: transmitted in real-time by sensors. Slip surface grade and the angle between the main structural planes and the slope: obtained through geological logging and back-analysis of monitoring data. Ensuring consistency in content and dimensions between real-time data and historical case data lays the foundation for subsequent quantification and comparison. Based on pre-defined unified rules for the case database, the extracted characteristic factors were mapped to numerical values. Continuous numerical values ​​(such as slip surface angle and groundwater level): directly used measured values ​​and normalized to eliminate dimensional influences. For example, divide all slope angles by 90 degrees. For graded rating types (e.g., RQD, slip surface grade): assign values ​​and normalize them according to preset grade standards (e.g., RQD 0-100%, or surface roughness grades 1-5). For category enumeration types (e.g., landslide pattern): use "one-hot encoding" to assign a physically meaningful sequential score. Combine the nine processed values ​​according to a fixed dimensional order (e.g., landslide pattern, slip surface angle, ..., rainfall) to generate a standardized multidimensional real-time feature vector representing the current state of the target slope. Compare the generated real-time feature vector with the feature vector of each historical case node in the case evolution graph. To improve matching accuracy, use two complementary similarity metrics: weighted Euclidean distance and cosine similarity. Weighted Euclidean distance: calculates the straight-line distance between two vectors in nine-dimensional space, but assigns different weights to each dimension. The weights are pre-set based on the degree of influence of the feature factor on slope stability (e.g., RQD and groundwater level may have higher weights). Weighted Euclidean distance measures the overall similarity of two states in absolute values; the smaller the distance, the more similar the absolute values. Cosine similarity: calculates the cosine of the angle between two vectors, ignoring their absolute values. Cosine similarity measures the similarity of two states in terms of change trends and morphology. For example, even if two slopes have different absolute displacements, if their displacement rates, water level changes, and other indicators have consistent change patterns, the cosine similarity will be high. This dual measurement overcomes the limitations of a single indicator, capturing both differences in absolute values ​​and similarities in morphological patterns, making the matching results more comprehensive and reliable. A pre-set fusion function (e.g., weighted average, harmonic mean, etc.) combines the weighted Euclidean distance (which needs to be converted to a similarity score, such as 1 / (1+distance)) and cosine similarity into a comprehensive matching score.The algorithm iterates through all nodes in the case evolution graph, calculating the comprehensive matching score between each node and the multidimensional real-time feature vector. Finally, it selects the historical case node with the highest comprehensive matching score and designates it as the target node.

[0035] In some embodiments, the step of searching and generating multiple first potential evolutionary paths leading to known instability cases in the case evolution graph, starting from the target node, specifically includes: Starting from the target node, search all reachable subsequent nodes along the directed edge direction to form multiple first paths. Starting from all known unstable case nodes, search all possible predecessor nodes in the reverse directed edge direction to form multiple second paths. Maintain a corresponding path state tuple for each search path. The path state tuple is used to record the cumulative transition probability product of the path and the number of path nodes. The search path is either the first path or the second path. If the cumulative transition probability product of the search path is lower than a preset probability threshold, or if the number of path nodes of the search path exceeds a preset maximum path depth, then the search path is terminated. When the first path and the second path have overlapping nodes, the first path and the second path are spliced ​​together to form a first potential evolutionary path, and the first potential evolutionary paths are sorted from largest to smallest according to the cumulative transition probability product.

[0036] Forward Search (Extending Future from Current State): Starting from the target node, all subsequent nodes are visited layer by layer along the direction indicated by the directed edges in the case evolution graph. Each visit to a new node forms a first path from the target node to that node. This simulates a series of potential state changes that a slope might undergo from its current state. Reverse Search (Tracing Precursors from Instability): Simultaneously, starting from each known instability case node (i.e., the node marked as having experienced a landslide in the graph), all predecessor nodes are visited layer by layer against the direction indicated by the directed edges. Each visit to a new node forms a second path from that node to the instability case node. For each path being searched (whether the first or second path), the system maintains a dynamically updated path state tuple. The path state tuple contains at least two key pieces of information: the cumulative transition probability product and the number of path nodes. Cumulative Transition Probability Product: The product of the weights of all directed edges traversed along the path. It represents the overall probability of evolution along this path, and the probability product decreases as the path lengthens. Number of path nodes: The total number of nodes currently contained in the path, i.e., the length or depth of the path. During the search process, the state tuple of each path is checked in real time. Further expansion of the path is terminated immediately if any of the following conditions are met: (1) The cumulative transition probability product is lower than the preset probability threshold, which means that the probability of the path occurring is very low, belonging to a low-probability event, and there is no need to continue tracking; (2) The number of path nodes exceeds the preset maximum path depth, which prevents the search from consuming too much computing resources on excessively long paths, ensuring the efficiency and timeliness of the search. After both the forward search and the reverse search are completed, the system will detect whether there are overlapping nodes in the path sets explored by both. When an overlapping node is found, a first path from the target node to the overlapping node is spliced ​​with a second path from the overlapping node to the unstable case node, thereby forming a complete first potential evolution path from the current state to the unstable state. All generated first potential evolution paths are sorted from largest to smallest according to their cumulative transition probability product. Prioritizing the most likely path enables assessors and decision-makers to immediately focus on the most critical and pressing risk evolution patterns, thus highlighting key areas and prioritizing risk management.

[0037] In some embodiments, assessing the risk level of each of the first potential evolutionary paths includes: Based on the quantification results of rock mass quality indicators, unloading loosening zone depth and slope angle in the multidimensional feature vectors of each node on the first potential evolution path, the local instability potential energy of each node is dynamically calculated. Based on the historical fluctuations of the directed edge weights in the first potential evolution path and the differences in the multidimensional feature vectors of adjacent nodes, the transition vulnerability of each directed edge is calculated. The local instability potential energy and the transfer vulnerability are used as inputs and sequentially input into a preset risk transmission chain model. The risk transmission chain model multiplies the output potential energy of the previous node with the transfer vulnerability of the current edge to obtain the input potential energy transmitted to the next node. The final output potential energy of the risk transmission chain model is defined as the comprehensive risk index of the first potential evolution path, and the comprehensive risk index is mapped to the corresponding risk level according to a preset numerical range.

[0038] For each node on the first potential evolution path, three key geological features are quantitatively extracted from its multidimensional feature vector: Rock Quality Index (RQD), unloading loosening zone depth, and slope angle. The local instability potential energy of this node is dynamically calculated using a pre-defined formula. The Rock Quality Index directly reflects the integrity and strength of the rock mass; the unloading loosening zone depth characterizes the degree of fracturing and loosening of the shallow rock mass—the greater the depth, the worse the stability; the slope angle is crucial to the slope geometry—the larger the angle, the more unstable the slope. The pre-defined formula can be, for example: Local Instability Potential Energy = f(RQD, Unloading Loosening Zone Depth, Slope Angle). This formula ensures that the lower the RQD value, the greater the unloading zone depth, and the larger the slope angle, the higher the calculated potential energy value. This vividly represents the node's state as if it has accumulated high gravitational potential energy, with an inherent tendency to become unstable and slide down at any time. For each directed edge on the path (i.e., the evolutionary step from one node to the next), its transition vulnerability is calculated, based on two aspects: the historical fluctuation of the directed edge weights and the differences in the multidimensional feature vectors of adjacent nodes. Historical fluctuation of directed edge weights: If the historical probability (i.e., weight) of the state evolution represented by this edge fluctuates greatly, it indicates that the evolutionary process is unstable and highly uncertain, and its vulnerability is higher. Differences in the multidimensional feature vectors of adjacent nodes: The degree of difference between the feature vectors of two nodes is calculated. If multiple feature factors require drastic changes to transition from one state to another, it indicates that this transition is difficult and has a high threshold, and its vulnerability may also be higher (or conversely, if slight changes lead to state transitions, it indicates that the system is very sensitive and fragile). Transition vulnerability describes the difficulty and reliability of the process of transitioning from one state to another. Higher vulnerability means that this evolutionary step is more likely to occur, or that this step is a weak link in the chain. This introduces dynamic and uncertain considerations, rather than relying solely on the static weights of the edges. The calculated local instability potential and transition vulnerability are then used as inputs to a risk transmission chain model. The operational rules of the risk transmission chain model are as follows: Starting from the initial node (i.e., the target node) of the path, the local instability potential energy of that node is used as the initial output potential energy. When the risk is transmitted to the next node, the output potential energy of the previous node is multiplied by the transfer vulnerability of the directed edge connecting the current node and the previous node to obtain the input potential energy transmitted to the current node. For example, it can be calculated using the following formula: Output potential energy of the current node = Local instability potential energy of the current node + Transmitted input potential energy. Repeat the above steps until the risk is transmitted to the unstable case node. The multiplication operation demonstrates that the risk may be amplified during the transmission process. If the transfer vulnerability of an edge is high, it will significantly amplify the risk transmitted from the previous link. Each node will add its own inherent risk (local instability potential energy) to the input risk, causing the risk to accumulate continuously as the path progresses.This allows for a dynamic and chain-like simulation of the transmission, amplification, and accumulation of risk throughout the entire evolutionary path. The final output potential energy when the chain model is transmitted to the instability case node is defined as the comprehensive risk index of the first potential evolutionary path. Based on a predefined numerical range, this comprehensive risk index is mapped to the corresponding risk level (e.g., low risk, medium risk, high risk, extremely high risk).

[0039] Figure 2 This is a schematic diagram of the process of simulating the application of virtual engineering intervention measures in the embodiments of this application, such as... Figure 2 As shown, the simulated virtual engineering intervention measures based on the early warning information include: S201. Using principal component analysis, identify the key instability factor with the highest contribution from the multidimensional feature vectors of the nodes of the first potential evolution path associated with the warning information. S202. Based on the key instability factor, retrieve engineering measures with inhibitory relationships from the preset intervention measure knowledge graph, and construct a virtual intervention action vector including three dimensions: support resistance, drainage efficiency, and slope adjustment according to the engineering measures. The intensity of each dimension is weighted according to the contribution of the key instability factor. S203. In the digital twin model, the virtual intervention vector is used as the boundary condition to calculate the stress field redistribution, displacement field evolution and dynamic change process of potential slip surfaces of the target slope after intervention, and a snapshot of the state after intervention is generated based on the calculation results.

[0040] The multidimensional feature vectors of all nodes along all potential evolutionary paths associated with the early warning information are extracted to form a data matrix. Principal Component Analysis (PCA) is then performed on this data matrix. PCA is a statistical method for data dimensionality reduction and feature extraction. Its purpose is to transform multiple correlated original variables (9 features) into a few independent composite variables, i.e., principal components, while retaining most of the original information. Each principal component is a linear combination of the original 9 features, but the weights (i.e., loadings) of the combination are mathematically optimized so that the first principal component can explain the largest proportion of variance (i.e., variation) in the data. The system selects the principal component with the highest variance contribution rate (usually the first principal component) because it captures the most important evolutionary pattern on the current risk path. The loading matrix of this principal component is analyzed. The loadings reflect the magnitude and direction of the contribution of each original feature to the principal component. The original features with the largest absolute values ​​of loadings (usually 1-3) are the key instability factors. For example, the analysis results show that in the first principal component (contribution rate 70%), the "groundwater level" load is 0.92 and the "RQD" load is -0.85, significantly higher than other features. This indicates that the main driving forces behind the evolution of the current risk path are the rise in groundwater level and the deterioration of RQD. Therefore, these two are identified as key instability factors. The system uses the identified key instability factors as query conditions to retrieve specific engineering measures with inhibitory or ameliorative relationships from a pre-defined intervention knowledge graph. The intervention knowledge graph is a database storing knowledge in the engineering field, clarifying the logical relationship between "geological factors" and "intervention measures." For example, if the key factor is "groundwater level," the knowledge graph may return measures such as "deep drainage holes" and "surface intercepting ditches"; if the key factor is "rock mass quality," it may return measures such as "system anchors" and "prestressed anchor cables." These specific, discrete engineering measures are abstracted and categorized into three fundamental engineering dimensions based on their mechanical mechanisms, constructing a virtual intervention vector V = (Support Resistance, Drainage Efficiency, Slope Adjustment). Support Resistance: This encompasses the reinforcement effect of all measures that enhance the strength of the soil and rock mass (such as anchor bolts, anchor cables, and anti-slide piles). Drainage Efficiency: This encompasses the drainage effect of all measures that reduce pore water pressure and descal the slope. Slope Adjustment: This encompasses the effect of all measures that improve mechanical balance by altering the geometric shape (such as slope cutting and toeing). The strength value of each dimension of the virtual intervention vector is weighted according to the contribution of key instability factors (i.e., the absolute value of their loads). For example, in the above example, the contribution of "groundwater level" (load 0.92) is slightly higher than that of "RQD" (load 0.85), so when constructing the vector, the strength assignment of the "Drainage Efficiency" dimension will be slightly higher than that of the "Support Resistance" dimension. The constructed virtual intervention vector V is used as a new set of boundary and load conditions and applied to the digital twin model of the target slope.A digital twin model is a high-fidelity three-dimensional numerical model (e.g., constructed based on the finite element method or finite difference method) that matches the geometry and geomechanical properties of a real slope. For example, adding a "drainage efficiency" vector value is equivalent to lowering the groundwater level in a specific area in the model; adding a "support resistance" vector value is equivalent to activating support structural elements with specific stiffness and strength at corresponding locations. Using a numerical simulation solver, the slope model with the intervention measures are recalculated to find its new mechanical equilibrium state. The core physical fields of the solution include stress field redistribution, displacement field evolution, and dynamic changes of potential slip surfaces. Stress field redistribution: the redistribution of stress within the slope after intervention, and whether high stress concentration areas are reduced or shifted. Displacement field evolution: whether the slope's deformation trend, displacement amount, and displacement rate are effectively controlled. Dynamic changes of potential slip surfaces: changes in the location, shape, and safety factor of the most dangerous slip surface. After the solution converges, the system captures the stability state of the slope at that moment and extracts a series of key state parameters (such as maximum displacement, plastic zone range, and overall safety factor) to form a snapshot of the state after intervention. This snapshot is a complete and quantitative description of the final state of the slope under the virtual intervention strategy.

[0041] In some embodiments, evaluating the effect of the virtual engineering intervention on the modification of the case evolution map includes: The features in the post-intervention state snapshot are mapped to virtual state nodes in the case evolution graph. Starting from the virtual state nodes, all second potential evolutionary paths leading to known unstable cases are searched again in the case evolution graph. The first potential evolutionary path is compared with the second potential evolutionary path, and the original high-risk path that does not appear in the second potential evolutionary path is identified as the high-risk path that is cut off. The path cutting rate is calculated as the proportion of the number of high-risk paths that have been cut off to the total number of paths in the first potential evolution path. The average comprehensive risk index of all paths in the second potential evolution path is calculated and compared with the average comprehensive risk index of the first potential evolution path to obtain the risk decay degree. Based on the path interruption rate and the risk attenuation, the topology reconstruction intensity value of the virtual engineering intervention measures on the case evolution map is obtained by weighted calculation. The topology reconstruction intensity value is used to characterize the intensity of the modification effect.

[0042] The slope features (such as displacement field, stress field, safety factor, and other back-calculated characteristic factor values ​​after stabilization) contained in the post-intervention state snapshot are mapped and a new multidimensional feature vector is generated according to the same quantification rules as historical cases. This new multidimensional feature vector is temporarily added to the case evolution map as a virtual state node. This virtual state node represents the new stable state reached by the slope after the intervention measures are implemented. Using this virtual state node as a new starting point, all paths leading to known instability cases are re-searched in the case evolution map according to the same search algorithm as before (such as bidirectional search). These newly found paths are called second potential evolution paths. The first potential evolution path found before the intervention (representing the original risk) is compared with the second potential evolution path found after the intervention (representing the residual risk) across the entire set. Paths in the first potential evolution path whose risk level is assessed as "high risk" or "very high risk" are the original high-risk paths. Specifically, the original high-risk paths that disappear in the second potential evolution paths are identified, that is, the high-risk paths that have been cut off. For example, before intervention, the system predicted 5 possible landslide paths. After intervention, it was found that 2 of these paths were completely disconnected, indicating that the measures precisely addressed these two specific instability mechanisms. The path cutoff rate is calculated as follows: Path Cutoff Rate = (Number of cut high-risk paths) / (Total number of first potential evolutionary paths). This indicator reflects the effectiveness of the intervention in eliminating the diversity of risky paths. A higher rate indicates that the measures fundamentally eliminate more instability possibilities. The risk decay rate is calculated as follows: Risk Decay Rate = (Average comprehensive risk index of the first potential evolutionary path) - (Average comprehensive risk index of the second potential evolutionary path). This indicator reflects the effectiveness of the intervention in reducing the risk level of the remaining paths. A positive decay rate and a larger value indicate that even for those paths that were not completely cut off, their overall risk level was significantly reduced. The calculated path cutoff rate (e.g., 0.7) and risk decay rate (e.g., 35%) are fused using a pre-defined weighted formula to obtain a single scalar value—the topology remodeling intensity value. The formula may take the form: Topology Reconstruction Strength Value = α·Path Cutoff Rate + β·Risk Decay (where α and β are preset weighting coefficients, and α + β = 1). The Topology Reconstruction Strength Value comprehensively reflects the intensity of the modification of the case evolution graph topology by the virtual engineering intervention measures. The higher the value, the better the effect of the measure on the original risk network, reshaping a safer risk evolution pattern.

[0043] In some embodiments, evaluating the mitigation effect of each intervention strategy based on the modification includes: Each intervention strategy is compared with predefined engineering constraints, and intervention strategies that meet the engineering constraints are selected to form a set of feasible strategies. The engineering constraints include an implementation cost threshold and a construction period upper limit. For each intervention strategy in the set of feasible strategies, the topology reconstruction intensity values ​​of all the virtual engineering intervention measures included are weighted and summed to calculate the comprehensive mitigation effectiveness score of the intervention strategy, and the intervention strategy with the highest comprehensive mitigation effectiveness score is selected as the optimal intervention strategy. The optimal intervention strategy is pushed to the engineering management terminal, and a monitoring task is started simultaneously to track the slope status feedback data after the optimal intervention strategy is actually deployed.

[0044] Each virtual intervention strategy is automatically compared with predefined engineering constraints. These constraints primarily include an implementation cost threshold and a construction period cap. The implementation cost threshold (including materials, labor, machinery, etc.) must not exceed a certain budget limit. The construction period cap (the time required to complete the strategy) must be within the required timeframe. All intervention strategies that simultaneously meet these constraints are included in the feasible strategy set. Any strategy exceeding the cost or time budget is directly eliminated at this stage. For each intervention strategy in the feasible strategy set, the system performs an in-depth quantitative evaluation of its effectiveness. An intervention strategy may include multiple specific engineering measures (such as simultaneously using anchor bolts and drainage holes). For each virtual engineering intervention measure included in the strategy, the system takes its previously calculated topology reconstruction strength value (which quantifies the effect of a single measure), weights and sums these strength values ​​to calculate the overall mitigation effectiveness score of the entire strategy. The system directly compares the overall mitigation effectiveness scores of all strategies in the feasible strategy set and automatically selects the strategy with the highest score as the final recommended optimal intervention strategy. The selected optimal intervention strategy (including specific measures, construction drawings, budget, and schedule) is directly pushed to the project management terminal, notifying relevant project management personnel and construction units to guide on-site construction. Simultaneously, the system initiates a dedicated monitoring task to track the feedback data on slope condition after the actual deployment of this optimal intervention strategy. By tracking the evolution data of the slope condition after actual intervention and comparing it with the predicted effects in the digital twin, the system can determine the degree of agreement between the simulation prediction and the actual situation. This real data is used as new cases and fed into the case evolution graph to optimize future early warning and decision-making models, enabling the entire system to continuously learn from actual engineering results, becoming increasingly intelligent and accurate.

[0045] The above describes a cloud computing-based intelligent early warning method for high slope disasters in the embodiments of this application. The computer system in the embodiments of this application will be described in detail below in conjunction with the above-mentioned cloud computing-based intelligent early warning method for high slope disasters.

[0046] Please see Figure 3This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application.

[0047] In some embodiments, the computer system 300 includes a computer device, which may be a terminal device. The computer device includes a processor 301, a memory 302, a sensor module 303, a communication module 304, an input device 305, and an output device 306 connected via a system bus. The processor 301 of the computer device provides computing and control capabilities. The memory 302 of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database is used to store data.

[0048] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0049] In some embodiments of this application, a computer-readable storage medium is provided, including instructions that, when executed on the computer system 300, cause the computer system 300 to execute a cloud computing-based intelligent early warning method for high slope disasters according to an embodiment of this application.

[0050] In some embodiments of this application, a computer program product is also provided, which, when run on a computer system 300, causes the computer system 300 to execute a cloud computing-based intelligent early warning method for high slope disasters according to an embodiment of this application.

[0051] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0052] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0053] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A cloud computing-based intelligent early warning method for high slope disasters, characterized in that, include: Based on a historical landslide case database, a case evolution graph is constructed in the cloud. The nodes in the case evolution graph are historical normal cases and historical landslide cases represented by multi-dimensional feature vectors. The directed edges between nodes represent the evolution paths between cases. The weights of the directed edges are assigned based on the feature similarity and spatiotemporal correlation between cases. The real-time monitoring data of the target slope is quantified into a multi-dimensional real-time feature vector. The multi-dimensional real-time feature vector is matched with the nodes in the case evolution map to determine the target node with the highest similarity. Starting from the target node, multiple first potential evolution paths leading to known instability cases are searched and generated in the case evolution map, and the risk level of each first potential evolution path is evaluated. When a risk level exceeds a preset threshold, an early warning is generated based on the risk level. In the digital twin environment of the target slope, virtual engineering intervention measures are simulated and applied based on the early warning information. The modification effect of the virtual engineering intervention measures on the case evolution map is evaluated, and the mitigation effect of each intervention strategy is evaluated based on the modification effect. Based on the mitigation effect, the optimal intervention strategy is selected and deployed, and the actual slope state evolution sequence and the effect feedback of the actual intervention measures are used as new nodes and added to the case evolution map.

2. The intelligent early warning method for high slope disasters based on cloud computing according to claim 1, characterized in that, The process of quantifying real-time monitoring data of the target slope into a multi-dimensional real-time feature vector, and matching the multi-dimensional real-time feature vector with nodes in the case evolution map to determine the target node with the highest similarity, specifically includes: Multiple feature factors corresponding to the historical landslide case database are extracted from the real-time monitoring data. These feature factors include landslide pattern, sliding surface angle, sliding surface grade, angle between the main structural plane and the slope, rock mass quality index, slope angle, unloading loosening zone depth, groundwater level, and rainfall. According to the preset quantification rules, each feature factor is mapped to a numerical value or category identifier, and then combined in a predetermined dimensional order to generate the multidimensional real-time feature vector of the target slope. Calculate the weighted Euclidean distance and cosine similarity between the multidimensional real-time feature vector and the multidimensional feature vector of each node in the case evolution graph; The weighted Euclidean distance and the cosine similarity are combined using a preset fusion function to obtain a comprehensive matching score; The node with the highest comprehensive matching score is determined as the target node that is most similar to the current state of the target slope.

3. The intelligent early warning method for high slope disasters based on cloud computing according to claim 1, characterized in that, Starting from the target node, the process of searching and generating multiple first potential evolutionary paths leading to known instability cases in the case evolution graph specifically includes: Starting from the target node, search all reachable subsequent nodes along the directed edge direction to form multiple first paths. Starting from all known unstable case nodes, search all possible predecessor nodes in the reverse directed edge direction to form multiple second paths. Maintain a corresponding path state tuple for each search path. The path state tuple is used to record the cumulative transition probability product of the path and the number of path nodes. The search path is either the first path or the second path. If the cumulative transition probability product of the search path is lower than a preset probability threshold, or if the number of path nodes of the search path exceeds a preset maximum path depth, then the search path is terminated. When the first path and the second path have overlapping nodes, the first path and the second path are spliced ​​together to form a first potential evolutionary path, and the first potential evolutionary paths are sorted from largest to smallest according to the cumulative transition probability product.

4. The intelligent early warning method for high slope disasters based on cloud computing according to claim 3, characterized in that, The assessment of the risk level for each of the first potential evolutionary paths includes: Based on the quantification results of rock mass quality indicators, unloading loosening zone depth and slope angle in the multidimensional feature vectors of each node on the first potential evolution path, the local instability potential energy of each node is dynamically calculated. Based on the historical fluctuations of the directed edge weights in the first potential evolution path and the differences in the multidimensional feature vectors of adjacent nodes, the transition vulnerability of each directed edge is calculated. The local instability potential energy and the transfer vulnerability are used as inputs and sequentially input into a preset risk transmission chain model. The risk transmission chain model multiplies the output potential energy of the previous node with the transfer vulnerability of the current edge to obtain the input potential energy transmitted to the next node. The final output potential energy of the risk transmission chain model is defined as the comprehensive risk index of the first potential evolution path, and the comprehensive risk index is mapped to the corresponding risk level according to a preset numerical range.

5. The intelligent early warning method for high slope disasters based on cloud computing according to claim 1, characterized in that, The simulated application of virtual engineering intervention measures based on the early warning information includes: The key instability factor with the highest contribution is identified from the multidimensional feature vectors of the nodes of the first potential evolution path associated with the warning information using principal component analysis. Based on the key instability factors, engineering measures with inhibitory relationships are retrieved from the preset intervention measure knowledge graph. A virtual intervention action vector with three dimensions, namely support resistance, drainage efficiency, and slope adjustment, is constructed according to the engineering measures. The intensity of each dimension is weighted according to the contribution of the key instability factors. In the digital twin model, the virtual intervention vector is used as the boundary condition to calculate the stress field redistribution, displacement field evolution and dynamic changes of potential slip surfaces of the target slope after intervention, and a snapshot of the state after intervention is generated based on the calculation results.

6. The intelligent early warning method for high slope disasters based on cloud computing according to claim 5, characterized in that, The assessment of the effects of the virtual engineering intervention on the case evolution map includes: The features in the post-intervention state snapshot are mapped to virtual state nodes in the case evolution graph. Starting from the virtual state nodes, all second potential evolutionary paths leading to known unstable cases are searched again in the case evolution graph. The first potential evolutionary path is compared with the second potential evolutionary path, and the original high-risk path that does not appear in the second potential evolutionary path is identified as the high-risk path that is cut off. The path cutting rate is calculated as the proportion of the number of high-risk paths that have been cut off to the total number of paths in the first potential evolution path. The average comprehensive risk index of all paths in the second potential evolution path is calculated and compared with the average comprehensive risk index of the first potential evolution path to obtain the risk decay degree. Based on the path interruption rate and the risk attenuation, the topology reconstruction intensity value of the virtual engineering intervention measures on the case evolution map is obtained by weighted calculation. The topology reconstruction intensity value is used to characterize the intensity of the modification effect.

7. The intelligent early warning method for high slope disasters based on cloud computing according to claim 6, characterized in that, The evaluation of the mitigation effects of each intervention strategy based on the modifications includes: Each intervention strategy is compared with predefined engineering constraints, and intervention strategies that meet the engineering constraints are selected to form a set of feasible strategies. The engineering constraints include an implementation cost threshold and a construction period upper limit. For each intervention strategy in the set of feasible strategies, the topology reconstruction intensity values ​​of all the virtual engineering intervention measures included are weighted and summed to calculate the comprehensive mitigation effectiveness score of the intervention strategy, and the intervention strategy with the highest comprehensive mitigation effectiveness score is selected as the optimal intervention strategy. The optimal intervention strategy is pushed to the engineering management terminal, and a monitoring task is started simultaneously to track the slope status feedback data after the optimal intervention strategy is actually deployed.

8. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

Images