A Real-time Monitoring and Early Warning Method and System for Geological Hazards Based on AI Algorithms

Abstract

This application provides a method and system for real-time monitoring and early warning of geological disaster hazards based on AI algorithms, belonging to the field of geological disaster monitoring technology. The method includes: acquiring first hazard monitoring data from multiple hazard monitoring equipment groups and constructing a hazard monitoring dataset according to preset geological disaster types; determining the first hazard risk sequence corresponding to each preset hazard monitoring area based on each hazard monitoring dataset and a geological disaster prediction model, and dividing multiple hazard joint monitoring area sets according to each first hazard risk sequence; determining whether there is a hazard state deviation area based on second hazard monitoring data, each first hazard risk sequence, and a preset hazard evolution identification model. If so, determining the range of the hazard state deviation area based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard state deviation area, generating hazard evolution early warning information based on the range of the hazard state deviation area, and sending it to the user terminal.

Description

Technical Field

[0001] This application relates to the field of geological disaster monitoring technology, and in particular to a method and system for real-time monitoring and early warning of potential geological disaster hazards based on AI algorithms. Background Technology

[0002] Geological disasters are geological processes or phenomena caused by natural or human factors that can damage human life, property, and the environment, such as landslides, mudslides, and road collapses. Their spatial and temporal distribution is influenced by both the natural environment and human activities. Currently, monitoring of geological disaster risks mainly relies on manual on-site patrols and drone inspections. With the development of artificial intelligence, using AI to analyze data from on-site monitoring equipment to achieve monitoring and early warning has become an important means.

[0003] When monitoring potential geological hazards, existing methods often rely on single-point, timed monitoring due to limitations in on-site monitoring equipment hardware (such as power and network). This includes independent operation of GNSS displacement monitors and tilt sensors, using statistical thresholds for hazard warning and monitoring. This approach struggles to capture the interconnected evolution of regional hazards and fails to monitor geological hazards with evolutionary correlations in real time and flexibly. This results in untimely and incomplete hazard monitoring and warning at the regional level, increasing the false alarm rate and impacting the efficient execution of downstream geological hazard monitoring tasks. It may even lead to serious geological disaster consequences. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a method and system for real-time monitoring and early warning of potential geological hazards based on AI algorithms, which is used to monitor and warn of potential geological hazards in a real-time and comprehensive manner.

[0005] On the one hand, embodiments of this application provide a method for real-time monitoring and early warning of geological disaster hazards based on AI algorithms, the method comprising:

[0006] Acquire first-hazard monitoring data from multiple hazard monitoring equipment groups, and construct a hazard monitoring dataset according to preset geological hazard types; wherein, the hazard monitoring equipment group includes at least a GNSS receiver, a fiber optic earth pressure gauge, a water level gauge, and a rain gauge; the preset geological hazard types include at least one or more of the following: landslide, collapse, debris flow, and crack;

[0007] Based on the aforementioned hidden danger monitoring datasets and pre-trained geological disaster prediction models, the first hidden danger risk sequence corresponding to each preset hidden danger monitoring area is determined, and multiple sets of joint hidden danger monitoring areas are divided according to each first hidden danger risk sequence.

[0008] Based on the second hidden danger monitoring data corresponding to each of the joint monitoring areas for hidden dangers, the risk sequences of each of the first hidden dangers, and the preset hidden danger evolution identification model, it is determined whether there is a hidden danger state deviation area; wherein, the second hidden danger monitoring data includes at least real-time first hidden danger monitoring data and pre-correlated external activity data;

[0009] If so, based on the spatial distribution characteristics of the real-time first hidden danger monitoring data and the hidden danger status offset area, the range of the hidden danger status offset area is determined, so as to generate hidden danger evolution early warning information based on the range of the hidden danger status offset area and send it to the user terminal.

[0010] In one implementation of this application, first hazard monitoring data from multiple hazard monitoring equipment groups are obtained, and a hazard monitoring dataset is constructed according to a preset geological hazard type, specifically including:

[0011] After receiving the first hidden danger monitoring data from multiple hidden danger monitoring equipment groups, the feature attributes of each of the first hidden danger monitoring data from each of the hidden danger monitoring equipment groups corresponding to each of the preset hidden danger monitoring areas are extracted; wherein, the feature attributes include at least displacement, stress, groundwater level, and meteorological parameters;

[0012] According to the list of preset related attributes corresponding to each preset geological hazard type, each of the preset characteristic attributes and each preset geological environment parameter are added to the regional monitoring data subset in sequence according to the preset hidden danger monitoring area, and the regional monitoring data subsets corresponding to the same preset geological hazard type are combined into the hidden danger monitoring dataset.

[0013] In one implementation of this application, based on the aforementioned hazard monitoring datasets and a pre-trained geological disaster prediction model, a first hazard risk sequence corresponding to each preset hazard monitoring area is determined, specifically including:

[0014] Based on the preset geological hazard types of each of the aforementioned hazard monitoring datasets, a risk prediction sub-model is determined in the geological hazard prediction model corresponding to each of the aforementioned hazard monitoring datasets; wherein, the geological hazard prediction model is a pre-trained hybrid expert model;

[0015] Each of the aforementioned hazard monitoring datasets is input into the corresponding risk prediction sub-model to determine one or more hazard risk values ​​corresponding to the same preset hazard monitoring area; the multiple hazard risk values ​​correspond to different preset geological hazard types.

[0016] The first hidden danger risk sequence corresponding to the preset hidden danger monitoring area is generated in descending order.

[0017] In one implementation of this application, multiple sets of joint monitoring areas for hidden dangers are divided according to each of the first hidden danger risk sequences, specifically including:

[0018] Calculate the risk similarity between each pair of the first hidden danger risk sequences; wherein, the risk similarity is calculated based on the sum of the maximum hidden danger risk value and the sum of the minimum hidden danger risk value corresponding to each preset geological disaster type in the two first hidden danger risk sequences;

[0019] Based on the preset spatial dependencies of each preset hidden danger monitoring area and the risk similarity of each, a corresponding risk correlation matrix is ​​constructed;

[0020] Based on the risk correlation matrix and spectral clustering algorithm, an undirected weighted graph is constructed, and the corresponding Laplace matrix is ​​calculated. Based on the eigenvalue decomposition result of the Laplace matrix, an eigenvector matrix is ​​obtained so as to cluster multiple clusters, which serve as the set of joint monitoring areas for each initial hidden danger.

[0021] Based on the first average risk similarity and risk similarity variance among all pairs of regions within each initial joint monitoring region set, the second average risk similarity among any recombined subsets of the initial joint monitoring region set, and a preset threshold group, each initial joint monitoring region set is merged and split, with the merged and split results serving as each joint monitoring region set.

[0022] In one implementation of this application, based on the second hidden danger monitoring data corresponding to each of the joint hidden danger monitoring area sets, each of the first hidden danger risk sequences, and a preset hidden danger evolution identification model, it is determined whether there is a hidden danger state shift region, specifically including:

[0023] Using time as the row and monitoring data features as the column, a monitoring data feature matrix is ​​established corresponding to the second hidden danger monitoring data and the first hidden danger risk sequence; wherein, the pre-correlated external activity data in the second hidden danger monitoring data is obtained by marking and collecting associated external activities for each preset geological disaster type through an expert system; the associated external activities include at least one or more of the following: rainfall, earthquake, construction activities, mining activities, and agricultural activities;

[0024] Based on the preset geological environment parameters and pre-stored historical disaster data, a hazard status evolution network is determined among the preset hazard monitoring areas; wherein, the hazard status evolution network is a node network constructed according to the hazard-hazard correlation strength; the hazard-hazard correlation strength is obtained by weighted calculation based on the preset geological environment parameters and pre-stored historical disaster data;

[0025] The monitoring data feature matrix and the hidden danger state evolution network are input into the preset hidden danger evolution identification model so that the preset hidden danger evolution identification model determines the hidden danger state offset probability of each preset hidden danger monitoring area and compares the hidden danger state offset probability with a preset offset probability threshold.

[0026] And determine the preset hidden danger monitoring areas where the probability of the hidden danger state shift is greater than the preset shift probability threshold, and find that there is a hidden danger state shift evolution, so as to determine each preset hidden danger monitoring area with related synchronous hidden danger evolution as the hidden danger state shift area based on each preset hidden danger monitoring area with the hidden danger state shift evolution and the hidden danger state evolution network.

[0027] In one implementation of this application, based on the preset geological environment parameters and pre-stored historical disaster data, a network for the evolution of hazard states among the preset hazard monitoring areas is determined, specifically including:

[0028] Based on the pre-set geological environment parameters and the historical disaster data, a set of three correlation indicators is determined between each pair of the pre-set hidden danger monitoring areas; the set of three correlation indicators includes spatial correlation, geological similarity, and historical disaster synchronization frequency.

[0029] Based on the weighted sum of the preset weighted triplet and the associated index triplet, the correlation strength between the corresponding two preset hidden danger monitoring areas is determined.

[0030] Based on the correlation strength of each hidden danger and the existence condition of the preset edge, each preset hidden danger monitoring area is used as a node to construct the hidden danger state evolution network; wherein, the existence condition of the preset edge is that the correlation strength of the hidden danger is greater than the preset correlation strength threshold.

[0031] In one implementation of this application, the method further includes:

[0032] If it is determined that there is no such hazard state offset area, the monitoring data change curves corresponding to each of the preset hazard monitoring areas are determined;

[0033] By using a preset sliding window and a preset anomaly detection algorithm, the monitoring data change curve is iterated to check for any abnormal points.

[0034] If such a situation exists, a corresponding independent early warning message is generated and sent to the user terminal, and the corresponding first hidden danger risk sequence is updated to further divide the set of joint monitoring areas for each hidden danger.

[0035] In one implementation of this application, the range of the hazard state offset region is determined based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard state offset region, specifically including:

[0036] Based on the spatial distribution characteristics of the real-time first hidden danger monitoring data, interpolation processing is performed on the real-time first hidden danger monitoring data to determine the spatial distribution field of the hidden danger monitoring data;

[0037] Based on the spatial distribution field of the hazard monitoring data and the preset risk level range, the contour lines of different hazard risk levels are determined to obtain the hazard status offset area for each risk level.

[0038] The hotspot areas obtained by spatial hotspot analysis of the real-time first hidden danger monitoring data and the hidden danger status offset areas of each risk level are merged to obtain the range of the hidden danger status offset area.

[0039] In one implementation of this application, generating hazard evolution early warning information based on the range of the hazard status offset area specifically includes:

[0040] Based on the risk level corresponding to the range of the hazard status offset area, a warning information template in the preset warning information list is matched, and the preset geological disaster type and impact range that have undergone synchronous hazard evolution are added to the warning information template to obtain the hazard evolution warning information.

[0041] On the other hand, this application also provides a real-time monitoring and early warning system for geological disaster hazards based on AI algorithms, the system comprising:

[0042] The acquisition module is used to acquire first hazard monitoring data from multiple hazard monitoring equipment groups and construct a hazard monitoring dataset according to preset geological hazard types; wherein, the hazard monitoring equipment group includes at least a GNSS receiver, a fiber optic earth pressure gauge, a water level gauge, and a rain gauge; the preset geological hazard types include at least one or more of the following: landslide, collapse, debris flow, and crack;

[0043] The first determining module is used to determine the first hidden danger risk sequence corresponding to each preset hidden danger monitoring area based on each hidden danger monitoring dataset and a pre-trained geological disaster prediction model, and to divide multiple sets of joint hidden danger monitoring areas according to each first hidden danger risk sequence.

[0044] The second determining module is used to determine whether there is a hazard state deviation area based on the second hazard monitoring data corresponding to each of the hazard joint monitoring area sets, each of the first hazard risk sequences, and the preset hazard evolution identification model; wherein, the second hazard monitoring data includes at least real-time first hazard monitoring data and pre-correlated external activity data;

[0045] The third determining module is used to determine the range of the hazard status offset area based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard status offset area, so as to generate hazard evolution early warning information based on the hazard status offset area range and send it to the user terminal.

[0046] Compared with the prior art, the significant advantages of this application are as follows:

[0047] This application, through the aforementioned scheme, rationally establishes a correlation between multiple pre-set hazard monitoring areas, forming a joint hazard monitoring area set. This set can be used to identify the synchronous hazard evolution between at least two pre-set hazard monitoring areas, reveal the risk transmission mechanism between regions, and avoid blind spots in local monitoring. Simultaneously, it generates the hazard status offset area range, which can be effectively used for real-time joint monitoring and early warning of multiple pre-set hazard monitoring areas, and capture cross-regional, multi-factor hazard evolution. This application realizes the entire process from multi-source data acquisition and model analysis to early warning information generation and transmission, breaking through the limitations of single-point timed monitoring, and enabling real-time monitoring of geological disaster hazards and timely early warning. Attached Figure Description

[0048] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0049] Figure 1 This is a flowchart illustrating a method for real-time monitoring and early warning of geological disaster hazards based on AI algorithms, as described in this application.

[0050] Figure 2 This is a schematic diagram of the structure of a real-time monitoring and early warning system for geological disaster hazards based on AI algorithms, as described in an embodiment of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] This application provides a method and system for real-time monitoring and early warning of potential geological hazards based on AI algorithms, which can be used to monitor and warn of potential geological hazards in real time and comprehensively.

[0053] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0054] This application provides a method for real-time monitoring and early warning of geological disaster hazards based on AI algorithms. The method is applied to geological disaster hazard monitoring areas where hazard monitoring equipment groups have been pre-deployed. Figure 1 As shown, the method may include steps S101-S104:

[0055] S101, the server obtains the first hidden danger monitoring data from multiple hidden danger monitoring equipment groups and constructs a hidden danger monitoring dataset according to the preset geological disaster type.

[0056] The hazard monitoring equipment group includes at least a GNSS receiver, a fiber optic earth pressure gauge, a water level gauge, and a rain gauge; the preset geological hazard types include at least one or more of the following: landslide, collapse, debris flow, and crack.

[0057] It should be noted that the server, as the executing entity of the AI ​​algorithm-based real-time monitoring and early warning method for geological disaster hazards, is merely an example. The executing entity is not limited to the server; it can also be edge computing devices corresponding to various hazard monitoring equipment groups deployed on a pre-defined geological disaster monitoring object. This application does not make specific limitations in this regard. The geological disaster monitoring object can be a river basin, a mountain, etc., and the geological disaster monitoring object can be divided into several pre-defined hazard monitoring areas. A pre-defined hazard monitoring area can be understood as an area where hazard monitoring equipment groups have been set up by designated personnel, such as an area prone to landslides or with landslide risks. Hazard monitoring equipment groups can also specifically include, but are not limited to, GNSS receivers, crack gauges, inclinometers, water level gauges, piezometers, rain gauges, cameras, pressure sensors, etc. The hazard monitoring equipment groups corresponding to different pre-defined hazard monitoring areas can include different types of monitoring equipment, which are selected by the on-site personnel according to the actual geological disaster type and current situation. This application does not make specific limitations in this regard.

[0058] In this embodiment of the application, the acquisition of first hazard monitoring data from multiple hazard monitoring equipment groups and the construction of a hazard monitoring dataset according to preset geological hazard types specifically include:

[0059] After receiving monitoring data for each of the first hidden dangers from multiple hidden danger monitoring equipment groups, the characteristic attributes of each monitoring data for each of the first hidden dangers from each hidden danger monitoring equipment group corresponding to each preset hidden danger monitoring area are extracted. These characteristic attributes include at least displacement, stress, groundwater level, and meteorological parameters. Based on the pre-set list of relevant attributes corresponding to each preset geological hazard type, each characteristic attribute and each pre-set geological environment parameter are sequentially added to the regional monitoring data subset according to the preset hidden danger monitoring area. The regional monitoring data subsets corresponding to the same preset geological hazard type are then combined into a hidden danger monitoring dataset.

[0060] In other words, the server extracts corresponding attribute values ​​from the first hazard monitoring data sent by each hazard monitoring equipment group according to the pre-set monitoring data characteristics, thus obtaining various characteristic attributes. Using a pre-stored list of pre-set related attributes, the characteristic attributes corresponding to different pre-set geological hazard types in the same pre-set hazard monitoring area and the pre-set geological environment parameters of the pre-set hazard monitoring area are added to the regional monitoring data subset. Since each first hazard monitoring data collected in a pre-set hazard monitoring area can be used to assess the occurrence risk of different pre-set geological hazard types in that area, this application may generate regional monitoring data subsets corresponding to different pre-set geological hazard types for the same pre-set hazard monitoring area. For example, if the characteristic attributes of the pre-set hazard monitoring area include (displacement, stress, groundwater level), the regional monitoring data subset may include: a regional monitoring data subset of {displacement, stress, pre-set geological environment parameters} for the first geological hazard type, and a regional monitoring data subset of {stress, groundwater level, pre-set geological environment parameters} for the second geological hazard type. Among them, the pre-set geological environment parameters can be obtained by on-site investigation by specialists, including but not limited to topographic parameters, soil and rock properties, etc.

[0061] Subsequently, the server will statistically analyze the monitoring data subsets of each region corresponding to each preset hidden danger monitoring area within the geological disaster monitoring object, and obtain the hidden danger monitoring dataset corresponding to the preset geological disaster type according to the rule of merging the data of the same preset geological disaster type.

[0062] By dividing the hazard monitoring datasets related to geological hazard types, it is possible to efficiently analyze and process geological hazard-related data from multiple preset hazard monitoring areas.

[0063] S102, the server determines the first hidden danger risk sequence corresponding to each preset hidden danger monitoring area based on each hidden danger monitoring dataset and the pre-trained geological disaster prediction model, and divides multiple sets of joint hidden danger monitoring areas according to each first hidden danger risk sequence.

[0064] In this embodiment of the application, based on each hidden danger monitoring dataset and a pre-trained geological disaster prediction model, the first hidden danger risk sequence corresponding to each preset hidden danger monitoring area is determined, specifically including:

[0065] Based on the preset geological hazard types of each hazard monitoring dataset, risk prediction sub-models are determined within the geological hazard prediction model corresponding to each dataset. These geological hazard prediction models are pre-trained hybrid expert models. Each hazard monitoring dataset is input into its corresponding risk prediction sub-model to determine one or more hazard risk values ​​corresponding to the same preset hazard monitoring area. Multiple hazard risk values ​​correspond to different preset geological hazard types. The first hazard risk sequence corresponding to the preset hazard monitoring area is generated in descending order of risk.

[0066] In other words, this application deploys a pre-trained geological hazard prediction model to analyze the hazard monitoring dataset and obtain the hazard risk values ​​of a preset hazard monitoring area under different preset geological hazard types. Specifically, this application uses a hybrid expert model as the geological hazard prediction model, trained using several historical hazard monitoring dataset samples labeled with hazard risk values. This model has different risk prediction sub-models specifically for predicting risk values ​​for different preset geological hazard types. After the server processes each hazard monitoring dataset through its respective risk prediction sub-model, it can obtain the hazard risk value for each preset hazard monitoring area for each preset geological hazard type. It is possible that a preset hazard monitoring area has only one preset geological hazard type. Based on the preset hazard monitoring areas, a first hazard risk sequence bound to each area is generated. In the first hidden danger risk sequence, the risk values ​​of each hidden danger are arranged from large to small. The geological disaster types can include: landslide, collapse, debris flow, crack, etc. For example, the first hidden danger risk sequence of the preset hidden danger monitoring area A is [landslide: 0.9, collapse: 0.8, debris flow: 0, crack: 0], and the first hidden danger risk sequence of the preset hidden danger monitoring area B is [debris flow: 0.7, collapse: 0.4, debris flow: 0.3, crack: 0.1].

[0067] Furthermore, in one embodiment of this application, after obtaining the aforementioned first hidden danger risk sequence, in order to better address the shortcomings of single-point real-time monitoring, the above-mentioned division of multiple hidden danger joint monitoring area sets according to each first hidden danger risk sequence specifically includes:

[0068] Calculate the risk similarity between each pair of the first hidden danger risk sequences. The risk similarity is calculated based on the sum of the maximum and minimum hidden danger risk values ​​corresponding to each preset geological hazard type in the two first hidden danger risk sequences. Construct a corresponding risk correlation matrix based on the preset spatial dependencies of each preset hidden danger monitoring area and the risk similarity. Based on the risk correlation matrix and spectral clustering algorithm, construct an undirected weighted graph and calculate the corresponding Laplace matrix. Obtain the eigenvector matrix from the eigenvalue decomposition of the Laplace matrix to generate multiple clusters, which serve as the sets of initial hidden danger joint monitoring areas. Based on the first average risk similarity between all pairs of areas within each initial hidden danger joint monitoring area set, the risk similarity variance, the second average risk similarity between any recombined subset of the initial hidden danger joint monitoring area set, and preset threshold groups, merge and split the initial hidden danger joint monitoring area sets. The merged and split results serve as the sets of each hidden danger joint monitoring area.

[0069] In other words, this application can obtain the risk similarity between two first-stage hidden risk sequences through similarity calculation, wherein the risk similarity calculation formula is as follows:

[0070]

[0071] Where S(x,y) represents the sequence R corresponding to any two preset hazard monitoring areas x and y. x ={r x1 ,r x2 ,…,r xn} and R y ={r y1 ,r y2 ,…,r yn The risk similarity is defined as follows: x and y both range from 1 to the total number m of the preset hazard monitoring areas, where n is the total number of elements in the sequence. The closer S(x,y) is to 1, the more similar the risks are between the two areas. min(r xi r yi ) refers to r xi and r yi The minimum of the two, max(r) xi r yi ) refers to r xi and r yi The maximum of the two values. This similarity calculation highlights risk factors shared by two regions across all disaster types. For example, if both regions have high risks of landslides and earthquakes, these shared risks are amplified and reflected in the similarity score. When one region has an extremely high risk for a certain disaster type while the other has a low risk, the impact of this difference on the similarity score is minimized (because the numerator is minimized), which helps avoid distortion of the similarity score due to extreme values ​​for a single disaster type.

[0072] Subsequently, a risk correlation matrix is ​​constructed by combining the pre-defined spatial dependencies between the pre-defined hazard monitoring areas and the similarity of each risk. The pre-defined spatial dependencies can be the Euclidean distance d between the positional coordinates of two pre-defined hazard monitoring areas x and y. xy Element M in the risk correlation matrix xy Calculated using the following formula:

[0073]

[0074] Wherein, σ1 represents the scale parameter of spatial distance, used to control the range of influence of spatial proximity on similarity; a larger value indicates a slower decay of similarity due to distance. σ2 is the scale parameter of attribute difference, used to control the weight of the influence of attribute difference on similarity; a larger value indicates a lighter penalty for similarity due to attribute difference. Both can be set by the user according to the actual use case; this application does not impose specific limitations on them. M xy It is a symmetric matrix.

[0075] Through each M xyA risk correlation matrix M is constructed, and a set of joint monitoring areas for potential hazards is obtained using a spectral clustering algorithm. Specifically, an undirected weighted graph G = (V, E) is constructed, where V is the set of vertices representing the pre-defined hazard monitoring areas, E is the set of edges, and the weights of the edges are determined by the elements in the risk correlation matrix M, i.e., the weights w. xy =M xy Then, the Laplacian matrix L = DW of graph G is calculated, where D is the degree matrix, a diagonal matrix, and its diagonal elements are... W is the risk correlation matrix M. Performing eigenvalue decomposition on the Laplace matrix L yields eigenvalues ​​λ1≤λ2≤…≤λ m and the corresponding feature vectors v1, v2, ..., v m Select the eigenvectors corresponding to the k smallest non-zero eigenvalues ​​to construct an m×k eigenvector matrix V. k Where k can be set by the user according to the actual scenario, this application does not impose specific limitations on it. Subsequently, the server can output the feature vector matrix V k Each row in the dataset is considered a k-dimensional data point. Clustering algorithms such as K-means clustering are used to cluster these data points, resulting in k clusters. Each cluster corresponds to an initial set of joint monitoring areas for potential hazards. Furthermore, spectral clustering algorithms are used to map the risk data to a low-dimensional feature space, achieving accurate clustering of nonlinearly separable data.

[0076] Subsequently, the server will perform region merging and splitting operations on the initial set of joint monitoring areas for potential hazards to further optimize the set. This involves aggregating areas with highly similar risks into larger sets to improve monitoring efficiency, and splitting areas with significantly different risks into smaller sets to avoid misjudgments or omissions. After multiple iterations, the region division remains unchanged, resulting in the optimal set of monitoring areas.

[0077] Specifically, the first average risk similarity S avg，1 Calculation formula:

[0078]

[0079] Where |T| represents the number of preset hidden danger monitoring areas in the initial set of joint hidden danger monitoring areas T.

[0080] Formula for calculating the variance of risk similarity Var(T):

[0081]

[0082] Second average risk similarity S avg，2 The calculation formula is as follows:

[0083]

[0084] Where T1 and T2 represent all possible subsets of the initial joint monitoring area set T, assuming T1∪T2=T and T1∩T2=φ. If S avg，2 If the risk of T1 and T2 is greater than the first preset threshold Th1 in the preset threshold group, then T1 and T2 are considered to have highly similar risks and can be merged, with the joint monitoring area set being T1∪T2. However, if Var(T) is greater than the second preset threshold Th2 in the preset threshold group, then the risk difference within the corresponding set T is considered to be significant, requiring splitting. In this case, clustering algorithms such as K-means or hierarchical clustering can be used to split set T into multiple subsets, and the above merging and splitting operations are performed until all resulting sets satisfy S. avg，2 If ≤Th1 and Var(T)≤Th2, the iteration stops, and a set of joint monitoring areas for each hidden danger is obtained. The larger the first preset threshold, the more areas are likely to be merged. The first and second preset thresholds can be set by the user according to specific scenarios, and are not specifically limited here.

[0085] For example, there is a set of joint monitoring areas T with initial hidden dangers. a ={A,B,C} and T b ={D,E}, calculate T a subset T a1 ={A,B} and T a2 = The second average risk similarity S of {C} avg，2 =0.85>Th1, at this point merge T a1 and T a2 However, if we calculate T a Risk similarity variance Var(T) a If )>Th2, then T a The process involves splitting the data; after multiple iterations, the final set of joint monitoring areas for potential hazards is defined as {A,C}, {B}, and {D,E}.

[0086] Through the above technical solutions, based on the risk correlation matrix and spectral clustering algorithm, and combined with the dual constraints of risk similarity and variance, as well as the regional merging and splitting operations, the joint monitoring area of ​​hidden dangers is more accurately divided. This allows for a more scientific and reasonable set of joint monitoring areas of hidden dangers, thereby improving the accuracy and effectiveness of real-time monitoring and early warning of geological disaster hidden dangers.

[0087] S103, the server determines whether there is a hazard status shift area based on the second hazard monitoring data corresponding to the set of joint hazard monitoring areas, the risk sequence of each first hazard, and the preset hazard evolution identification model.

[0088] The second hazard monitoring data includes at least real-time first hazard monitoring data and pre-correlated external activity data. The hazard status offset area is obtained based on the synchronous hazard evolution related between at least two preset hazard monitoring areas.

[0089] Pre-correlated external activity data can be understood as external activity data that can affect the development of geological hazard risks set in advance for each different preset geological hazard type and / or preset hidden danger monitoring area. The associated external activities include at least one or more of the following: rainfall, earthquake, construction activities, mining activities, and agricultural activities.

[0090] In this embodiment of the application, the determination of whether there is a hazard state shift region based on the second hazard monitoring data corresponding to each hazard joint monitoring area set, each first hazard risk sequence, and a preset hazard evolution identification model specifically includes:

[0091] Using time as the row and monitoring data features as the column, a monitoring data feature matrix is ​​established corresponding to the second hidden danger monitoring data and the risk sequence of the first hidden danger. The pre-correlated external activity data in the second hidden danger monitoring data is collected by an expert system that labels and collects associated external activities for each preset geological disaster type. Associated external activities include at least one or more of the following: rainfall, earthquakes, construction activities, mining activities, and agricultural activities. Based on preset geological environment parameters and pre-stored historical disaster data, a hidden danger state evolution network is determined among the preset hidden danger monitoring areas. This network is a node network constructed based on the hidden danger disaster association strength. The association strength is calculated by weighting the preset geological environment parameters and pre-stored historical disaster data. The monitoring data feature matrix and the hidden danger state evolution network are input into a preset hidden danger evolution identification model to determine the hidden danger state shift probability for each preset hidden danger monitoring area, and the shift probability is compared with a preset shift probability threshold. And determine the preset hidden danger monitoring areas where the probability of hidden danger state deviation is greater than the preset deviation probability threshold, and where there is a hidden danger state deviation evolution, so as to determine the preset hidden danger monitoring areas with related synchronous hidden danger evolution as hidden danger state deviation areas based on each preset hidden danger monitoring area with hidden danger state deviation evolution and the hidden danger state evolution network.

[0092] In other words, this application can, for each set of joint monitoring areas for hidden dangers, fuse the corresponding second hidden danger monitoring data with the first hidden danger risk sequence of each preset hidden danger monitoring area within that set, thereby constructing a monitoring data feature matrix with behavioral time as the characteristic of the detection data. This application may also employ methods such as principal component analysis to reduce the dimensionality of the detection data feature matrix to remove data noise and redundant information; the specific choice is made by the user, and this application does not impose any specific limitations on this.

[0093] Subsequently, the server acquires preset geological environment parameters and historical disaster data pre-stored in a preset database for each preset hazard monitoring area, establishing a correlation between geological disasters in each preset hazard monitoring area, thereby constructing a hazard state evolution network. The server can process and analyze the preset geological environment parameters and historical disaster data to determine the degree of correlation between geological disasters and hazard occurrences. This degree of correlation is used as the edge weight connecting the nodes of each preset hazard monitoring area to establish the hazard state evolution network. Next, a preset hazard evolution identification model is used to process the obtained monitoring data feature matrix and hazard state evolution network. This preset hazard evolution identification model can be a graph neural network model trained using second hazard monitoring data, first hazard risk sequences, and known hazard state shifts over a past period. This model can output the hazard state shift probability corresponding to each preset hazard monitoring area. By comparing each hazard state shift probability with a preset shift probability threshold, preset hazard monitoring areas with hazard state shift probabilities greater than the preset threshold are identified as having hazard state shift evolution. This preset shift probability threshold can be set by the user according to the actual usage scenario and is not specifically limited here.

[0094] Simultaneously, the server will also use the aforementioned hazard state evolution network to query whether there are other preset hazard monitoring areas that have undergone the same hazard state shift evolution and are connected by edges, along the edges of preset hazard monitoring areas where hazard state shift evolution has occurred. If so, it indicates the existence of hazard state shift areas. In simpler terms, due to external activities, multiple monitoring areas in different locations may experience synchronous hazards. Previously, these might have been monitored individually without establishing effective and reliable correlations. This application, however, can analyze the existence of hazard state shift areas to enable joint monitoring of multiple related monitoring areas, avoiding the problem of data asynchrony and delayed monitoring and early warning caused by single-point monitoring.

[0095] In one embodiment of this application, the above-mentioned determination of the hazard status evolution network among each preset hazard monitoring area based on preset geological environmental parameters and pre-stored historical disaster data specifically includes:

[0096] Based on pre-set geological environmental parameters and historical disaster data, a set of three-factor association indicators is determined for each pair of pre-set hazard monitoring areas. These three-factor association indicators include spatial correlation, geological similarity, and historical disaster synchronization frequency. The hazard-disaster association strength between two pre-set hazard monitoring areas is determined based on the weighted sum of the pre-set weighted three-factor association indicators and the three-factor association indicators. Based on the hazard-disaster association strength and the pre-set edge existence conditions, each pre-set hazard monitoring area is used as a node to construct a hazard state evolution network. The pre-set edge existence condition is that the hazard-disaster association strength is greater than a pre-set association strength threshold.

[0097] In other words, this application first uses pre-set geological environmental parameters and historical disaster data to obtain a pairwise correlation index triplet between two pre-set hazard monitoring areas. This triplet includes spatial correlation, geological similarity, and historical disaster synchronization frequency. Spatial correlation can be calculated using the reciprocal of Euclidean distance; geological similarity can be calculated by calculating the cosine similarity of the pre-set geological environmental parameters between the two areas; and the historical disaster synchronization frequency can be obtained by the ratio of the statistical number of disasters occurring simultaneously or sequentially in the past to the total number of disasters occurring in both areas. Subsequently, the server uses a pre-set weighted triplet to weight the elements in the correlation index triplet and calculates the sum. The calculated sum is used as the hazard disaster correlation strength between the two areas. If the hazard disaster correlation strength between two pre-set hazard monitoring areas is greater than a pre-set correlation strength threshold, it indicates that the two areas are correlated, that is, there is an edge between the corresponding nodes of the two pre-set hazard monitoring areas. Through the above steps, the hazard disaster correlation strength between each pre-set hazard monitoring area is calculated sequentially, thereby constructing a hazard state evolution network.

[0098] The above scheme can construct a scientific and reasonable network for the evolution of potential hazards, providing strong support for geological disaster monitoring and early warning.

[0099] S104, when the server determines that there is a hazard status deviation area, it determines the range of the hazard status deviation area based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard status deviation area, so as to generate hazard evolution early warning information based on the range of the hazard status deviation area and send it to the user terminal.

[0100] In this embodiment, when the server determines that there are no areas of hazard status deviation, it determines the monitoring data change curves corresponding to each preset hazard monitoring area. Using a preset sliding window and a preset anomaly detection algorithm, the server iterates through the monitoring data change curves to check for anomalies. If anomalies are found, a corresponding independent early warning message is generated and sent to the user terminal, and the corresponding first hazard risk sequence is updated to further divide the set of joint monitoring areas for each hazard. Otherwise, the server continues to acquire and analyze hazard monitoring data according to preset monitoring time intervals to monitor and issue early warnings for geological disaster hazards.

[0101] In other words, when there is no area of ​​hazard status deviation, single-point monitoring continues for each preset hazard monitoring area. Using a preset sliding window and a preset anomaly monitoring algorithm, anomalies are monitored in the monitoring data change curves of each area. The preset anomaly monitoring algorithm determines whether the curve value within the preset sliding window range is greater than a first set value and / or whether the curve slope is greater than a second set value, thus determining whether an anomaly exists. If an anomaly exists, it indicates that the warning conditions have been met, and an independent warning message is generated for that preset hazard monitoring area. Simultaneously, the first hazard risk sequence is updated to keep the set of joint hazard monitoring areas up-to-date and prevent delays in joint monitoring and warning of various areas.

[0102] In this embodiment of the application, determining the range of the hazard state offset area based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard state offset area specifically includes:

[0103] Based on the spatial distribution characteristics of real-time primary hazard monitoring data, interpolation processing is performed on the real-time primary hazard monitoring data to determine the spatial distribution field of the hazard monitoring data. Based on the spatial distribution field of the hazard monitoring data and the preset risk level intervals, contour lines for different hazard risk levels are determined to obtain the hazard status offset area for each risk level. The hotspot areas obtained from spatial hotspot analysis of the real-time primary hazard monitoring data and the hazard status offset areas for each risk level are then merged to obtain the range of the hazard status offset area.

[0104] In other words, this application uses the spatial distribution characteristics of the first hazard monitoring data as a basis to perform spatial interpolation calculations to obtain a continuous spatial distribution field of hazard monitoring data. Simultaneously, based on pre-defined risk level regions, it identifies contour lines for different hazard risk levels within the spatial distribution field of the hazard monitoring data, thereby constructing hazard state shift regions for different risk levels to visually demonstrate the spatial distribution characteristics of the hazard monitoring data. Furthermore, it employs spatial hotspot analysis methods such as Getis-Ord Gi* to obtain hotspot regions in the real-time first hazard monitoring data. Hotspot regions can be understood as areas with high hazard risk values, representing key areas for hazard state shifts. This application merges the aforementioned hotspot regions and hazard state shift regions for each risk level. If a hotspot region overlaps with the area defined by contour lines, the overlapping portion and the extended area of ​​the hotspot region are included in the hazard state shift region, thereby accurately determining the range of the hazard state shift region based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard state shift region.

[0105] In one embodiment of this application, the above-mentioned generation of hazard evolution early warning information based on the hazard status offset area specifically includes:

[0106] Based on the risk level corresponding to the area of ​​the hazard status deviation, a warning information template from the preset warning information list is matched to add the preset geological disaster type and impact range of the synchronous hazard evolution to the warning information template, thereby obtaining the hazard evolution warning information.

[0107] In other words, this application sets different warning information templates for different risk levels, thereby enabling the sending of warning information on the evolution of potential hazards to user terminals in a more intuitive way. The correspondence between the warning information templates and risk levels can be set by the user and is not specifically limited here. The user terminal can be a user's mobile phone, computer, or other device, and this application does not specifically limit this.

[0108] This application, through the aforementioned scheme, rationally establishes a correlation between multiple pre-set hazard monitoring areas, forming a joint hazard monitoring area set. This set can be used to identify the synchronous hazard evolution between at least two pre-set hazard monitoring areas, reveal the risk transmission mechanism between regions, and avoid blind spots in local monitoring. Simultaneously, it generates the hazard status offset area range, which can be effectively used for real-time joint monitoring and early warning of multiple pre-set hazard monitoring areas, and capture cross-regional, multi-factor hazard evolution. This application realizes the entire process from multi-source data acquisition and model analysis to early warning information generation and transmission, breaking through the limitations of single-point timed monitoring, and enabling real-time monitoring of geological disaster hazards and timely early warning.

[0109] Figure 2 This application provides a schematic diagram of the structure of a real-time monitoring and early warning system for geological hazards based on an AI algorithm. The AI-based real-time monitoring and early warning system 200 is capable of executing the aforementioned AI-based real-time monitoring and early warning method for geological hazards. Figure 2 As shown, the AI-based real-time monitoring and early warning system for geological disaster hazards includes:

[0110] The acquisition module 201 is used to acquire first hazard monitoring data from multiple hazard monitoring equipment groups and construct a hazard monitoring dataset according to preset geological hazard types. The hazard monitoring equipment groups include at least a GNSS receiver, a fiber optic earth pressure gauge, a water level gauge, and a rain gauge; the preset geological hazard types include at least one or more of the following: landslide, collapse, debris flow, and crack. The first determination module 202 is used to determine the first hazard risk sequence corresponding to each preset hazard monitoring area based on each hazard monitoring dataset and a pre-trained geological hazard prediction model, and to divide multiple hazard joint monitoring area sets according to each first hazard risk sequence. The second determination module 203 is used to determine whether there is a hazard state deviation area based on the second hazard monitoring data corresponding to each hazard joint monitoring area set, each first hazard risk sequence, and a preset hazard evolution identification model. The second hazard monitoring data includes at least real-time first hazard monitoring data and pre-correlated external activity data. The third determination module 204, if yes, is used to determine the range of the hazard state deviation area based on the spatial distribution characteristics of the real-time first hazard monitoring data and the hazard state deviation area, and to generate hazard evolution early warning information based on the range of the hazard state deviation area and send it to the user terminal.

[0111] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0112] The systems and methods provided in this application are one-to-one correspondences. Therefore, the system also has similar beneficial technical effects as its corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system will not be repeated here.

[0113] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0114] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A real-time monitoring and early warning method for geological disaster hazards based on an AI algorithm, characterized in that, The method comprises: acquiring each first hidden danger monitoring data from a plurality of hidden danger monitoring device groups, and constructing a hidden danger monitoring data set according to a preset geological disaster type; wherein the hidden danger monitoring device group at least includes a GNSS receiver, a fiber bragg grating earth pressure gauge, a water level gauge and a rain gauge; the preset geological disaster type at least includes one or more of the following: landslide, collapse, debris flow, crack; based on each of the hidden danger monitoring data set and the pre-trained geological disaster prediction model, determine the first hidden danger risk sequence corresponding to each preset hidden danger monitoring area respectively, and divide a plurality of hidden danger joint monitoring area sets according to each of the first hidden danger risk sequence; specifically including: according to the preset geological disaster type of each of the hidden danger monitoring data set, determine the risk prediction sub-model in the geological disaster prediction model corresponding to each of the hidden danger monitoring data set respectively; wherein the geological disaster prediction model is a pre-trained hybrid expert model; input each of the hidden danger monitoring data set into the corresponding risk prediction sub-model to determine one or more hidden danger risk values corresponding to the same preset hidden danger monitoring area; the plurality of hidden danger risk values correspond to different preset geological disaster types; generate the first hidden danger risk sequence corresponding to the preset hidden danger monitoring area in descending order; specifically further including: calculating the risk similarity between each of the first hidden danger risk sequence; wherein the risk similarity is calculated based on the maximum hidden danger risk value and the minimum hidden danger risk value corresponding to each of the preset geological disaster type in the two first hidden danger risk sequences respectively; according to the preset spatial dependence relationship of each of the preset hidden danger monitoring area and each of the risk similarity, construct the corresponding risk correlation matrix; based on the risk correlation matrix and the spectral clustering algorithm, construct the undirected weighted graph and calculate the corresponding Laplace matrix, so as to obtain the eigenvector matrix according to the eigenvalue decomposition result of the Laplace matrix, so as to cluster to obtain a plurality of clustering clusters as each initial hidden danger joint monitoring area set; based on the first average risk similarity, the risk similarity variance between all region pairs in each of the initial hidden danger joint monitoring area set, the second average risk similarity between any recombination subset of the initial hidden danger joint monitoring area set and the preset threshold group, each of the initial hidden danger joint monitoring area set is combined and split to obtain the combined and split result as each of the hidden danger joint monitoring area set; based on the second hidden danger monitoring data corresponding to each of the hidden danger joint monitoring area set, each of the first hidden danger risk sequence and the preset hidden danger evolution identification model, determine whether there is a hidden danger state deviation area; wherein the second hidden danger monitoring data at least includes real-time first hidden danger monitoring data and pre-associated external activity data; if yes, according to the spatial distribution characteristics of the real-time first hidden danger monitoring data and the hidden danger state deviation area, determine the hidden danger state deviation area range, generate hidden danger evolution early warning information according to the hidden danger state deviation area range and send it to the user terminal.

2. The AI algorithm-based real-time geological disaster hazard monitoring and early warning method according to claim 1, characterized in that, Acquire each first hidden danger monitoring data from a plurality of hidden danger monitoring device groups, and construct a hidden danger monitoring data set according to a preset geological disaster type, specifically including: After receiving each first hidden danger monitoring data from a plurality of hidden danger monitoring device groups, extract the characteristic attributes of each first hidden danger monitoring data of each hidden danger monitoring device group corresponding to each preset hidden danger monitoring area respectively; wherein the characteristic attributes at least include displacement, stress, underground water level, meteorological parameters; According to the preset related attribute list corresponding to each preset geological disaster type, each characteristic attribute and each preset geological environment parameter are sequentially added to the regional monitoring data subset according to the preset hidden danger monitoring area, and each regional monitoring data subset corresponding to the same preset geological disaster type is combined into the hidden danger monitoring data set. 3.The AI algorithm-based real-time monitoring and early warning method for geological disaster hazards according to claim 2, characterized in that, Based on each second hidden danger monitoring data corresponding to each hidden danger joint monitoring area set, each first hidden danger risk sequence and a preset hidden danger evolution identification model, determine whether there is a hidden danger state deviation area, specifically including: Take time as row and monitoring data characteristics as column to establish a monitoring data characteristic matrix corresponding to the second hidden danger monitoring data and the first hidden danger risk sequence; wherein the pre-associated external activity data in the second hidden danger monitoring data is obtained by marking the associated external activity for each preset geological disaster type through an expert system and collecting; the associated external activity at least includes one or more of the following: rainfall, earthquake, construction activity, mining activity, agricultural activity; Determine the hidden danger state evolution network between each preset hidden danger monitoring area based on the preset geological environment parameter and the pre-stored historical disaster data; wherein the hidden danger state evolution network is a node network constructed according to the hidden danger disaster correlation strength; the hidden danger disaster correlation strength is obtained by weighted calculation based on the preset geological environment parameter and the pre-stored historical disaster data; Input the monitoring data characteristic matrix and the hidden danger state evolution network into the preset hidden danger evolution identification model, so that the preset hidden danger evolution identification model determines the hidden danger state deviation probability of each preset hidden danger monitoring area, and compares the hidden danger state deviation probability with a preset deviation probability threshold; And determine the preset hidden danger monitoring area whose hidden danger state deviation probability is greater than the preset deviation probability threshold, which has hidden danger state deviation evolution, so as to determine each preset hidden danger monitoring area with related synchronous hidden danger evolution as the hidden danger state deviation area according to each preset hidden danger monitoring area with the hidden danger state deviation evolution and the hidden danger state evolution network.

4. The AI algorithm-based real-time geological disaster hazard monitoring and early warning method according to claim 3, characterized in that, Determine the hidden danger state evolution network between each preset hidden danger monitoring area based on the preset geological environment parameter and the pre-stored historical disaster data, specifically including: According to each preset geological environment parameter and each historical disaster data, determine the correlation index triplets between each preset hidden danger monitoring area; the correlation index triplets include spatial correlation degree, geological similarity, historical disaster synchronization frequency; According to the weighted calculation and value of the preset weight triplets and the correlation index triplets, the hidden danger disaster correlation strength between the corresponding two preset hidden danger monitoring areas is determined; According to the hidden danger disaster correlation strength and a preset edge existence condition, the hidden danger state evolution network is constructed by taking each preset hidden danger monitoring area as a node, wherein the preset edge existence condition is that the hidden danger disaster correlation strength is greater than a preset correlation strength threshold.

5. The AI algorithm-based real-time geological disaster hazard monitoring and early warning method according to claim 1, characterized in that, The method further comprises: In a case where it is determined that the hidden danger state deviation region does not exist, the monitoring data change curve corresponding to each preset hidden danger monitoring area is determined; Through a preset sliding window and a preset anomaly detection algorithm, it is determined whether there is an abnormal point in the monitoring data change curve; If there is, the corresponding independent early warning information is generated and sent to the user terminal, and the corresponding first hidden danger risk sequence is updated to divide the hidden danger joint monitoring area set again. 6.The AI algorithm-based real-time monitoring and early warning method for geological disaster hazards according to claim 1, characterized in that, According to the spatial distribution characteristics of the real-time first hidden danger monitoring data and the hidden danger state deviation region, the hidden danger state deviation region range is determined, specifically comprising: According to the spatial distribution characteristics of the real-time first hidden danger monitoring data, the real-time first hidden danger monitoring data is subjected to interpolation processing to determine a hidden danger monitoring data spatial distribution field; According to the hidden danger monitoring data spatial distribution field and a preset risk level interval, the isograms of different hidden danger risk levels are determined to obtain the risk level hidden danger state deviation region; The hotspot region obtained by performing spatial hotspot analysis on the real-time first hidden danger monitoring data and the risk level hidden danger state deviation region are regionally fused to obtain the hidden danger state deviation region range.

7. The AI algorithm-based real-time geological disaster hazard monitoring and early warning method according to claim 6, characterized in that, The hidden danger evolution early warning information is generated according to the hidden danger state deviation region range, specifically comprising: According to the risk level corresponding to the hidden danger state deviation region range, the early warning information template in the preset early warning information list is matched to add the preset geological disaster type and the influence range of the synchronous hidden danger evolution in the early warning information template to obtain the hidden danger evolution early warning information.

8. A real-time monitoring and early warning system for geological disaster hazards based on an AI algorithm, characterized in that, The system can perform the geological disaster hidden danger real-time monitoring and early warning method based on an AI algorithm according to any one of claims 1-7; the system comprises: An acquisition module is configured to acquire each first hidden danger monitoring data from a plurality of hidden danger monitoring device groups and construct a hidden danger monitoring data set according to a preset geological disaster type; wherein the hidden danger monitoring device group at least includes a GNSS receiver, a fiber bragg grating earth pressure gauge, a water level gauge and a rain gauge; the preset geological disaster type at least includes one or more of the following: landslide, collapse, debris flow, crack; A first determination module is configured to determine a first hidden danger risk sequence corresponding to each preset hidden danger monitoring area based on each hidden danger monitoring data set and a pre-trained geological disaster prediction model, and divide a plurality of hidden danger joint monitoring area sets according to each first hidden danger risk sequence; The second determining module is configured to determine whether there is a hidden danger state deviation region based on second hidden danger monitoring data corresponding to each of the hidden danger joint monitoring region sets, each of the first hidden danger risk sequences, and a preset hidden danger evolution identification model, wherein the second hidden danger monitoring data at least includes real-time first hidden danger monitoring data and pre-associated external activity data. The third determining module is configured to, if yes, determine a hidden danger state deviation region range according to spatial distribution characteristics of the real-time first hidden danger monitoring data and the hidden danger state deviation region, so as to generate hidden danger evolution early warning information according to the hidden danger state deviation region range and send the information to a user terminal.

Images