Landslide early warning method, device, equipment and medium

By combining Gaussian mixture models and hidden Markov models, a landslide state evolution model was constructed using data collected by Beidou micro-motion devices. This solved the problems of low data accuracy and insufficient timeliness in landslide early warning, and enabled real-time dynamic monitoring and accurate early warning of landslide geological disasters.

CN121281235APending Publication Date: 2026-01-06ANHUI POLYTECHNIC UNIV MECHANICAL & ELECTRICAL COLLEGE
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202511387993.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-26
Publication Date
2026-01-06

AI Technical Summary

Technical Problem

The existing mass monitoring and prevention model is unable to reflect the displacement trend and dynamic evolution characteristics of landslide bodies in a timely and accurate manner in landslide early warning, and cannot predict the time of landslide occurrence in advance, resulting in insufficient accuracy and timeliness of early warning.

Method used

By combining Gaussian mixture model and hidden Markov model, and based on the three-dimensional displacement time series data collected by Beidou micro-motion device, a landslide state evolution model is constructed. By analyzing real-time data, the current displacement state and remaining time of the landslide body are determined, thereby realizing landslide early warning.

Benefits of technology

It enables real-time dynamic monitoring and precise early warning of landslide geological hazards, significantly improving the accuracy and timeliness of early warnings, and allowing for timely preventive measures to reduce losses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121281235A_ABST
    Figure CN121281235A_ABST
Patent Text Reader

Abstract

The invention discloses a landslide early warning method, device and equipment and a medium, and the method comprises the steps: obtaining historical landslide three-dimensional displacement time sequence data of a historical landslide event; constructing a landslide state evolution model based on a Gaussian mixture model, a hidden Markov model and historical landslide three-dimensional displacement time series data; determining a current displacement state of the target landslide body based on the real-time three-dimensional displacement time sequence data of the target landslide body and a landslide state evolution model; obtaining the remaining time of the target landslide body from the current displacement state to the landslide occurrence state; and carrying out landslide early warning on the target landslide mass based on the current displacement state and the remaining time. By analyzing the displacement time sequence data of the historical landslide event and optimizing the prediction performance of the HMM model by using the GMM model, a scientific and reliable geological disaster monitoring and early warning system is constructed, real-time dynamic monitoring and accurate early warning of the landslide geological disaster are realized, and the timeliness and accuracy of early warning of the landslide geological disaster are remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of landslide detection technology, and in particular to a landslide early warning method, device, equipment and medium. Background Technology

[0002] Currently, landslide early warning systems primarily employ a community-based monitoring model. However, this model suffers from low data accuracy and lacks an efficient information feedback mechanism, resulting in delayed data reporting. This makes it difficult to reflect the displacement trend and dynamic evolution characteristics of landslide bodies in a timely and accurate manner, hindering the ability to predict landslide occurrence times and compromising the accuracy and timeliness of landslide early warnings. Summary of the Invention

[0003] This invention provides a landslide early warning method, device, computer equipment, and medium to solve the technical problem that the monitoring mode of mass monitoring and prevention is difficult to reflect the displacement trend and dynamic evolution characteristics of landslide bodies in a timely and accurate manner, and cannot predict the occurrence time of landslides in advance.

[0004] Firstly, a landslide early warning method is provided, which includes:

[0005] Obtain historical three-dimensional displacement time-series data of historical landslide events;

[0006] A landslide state evolution model is constructed based on Gaussian mixture model, hidden Markov model and historical three-dimensional displacement time series data of landslides;

[0007] Based on the real-time three-dimensional displacement time series data of the target landslide body and the landslide state evolution model, the current displacement state of the target landslide body is determined;

[0008] Obtain the remaining time from the current displacement state of the target landslide body to the state of landslide occurrence;

[0009] Landslide early warning is provided for the target landslide based on the current displacement status and remaining time.

[0010] Secondly, a landslide early warning device is provided, the device comprising:

[0011] The first acquisition module is used to acquire historical three-dimensional displacement time series data of historical landslide events.

[0012] The building module is used to construct a landslide state evolution model based on Gaussian mixture model, hidden Markov model and historical three-dimensional displacement time series data of landslides;

[0013] The determination module is used to determine the current displacement state of the target landslide body based on the real-time three-dimensional displacement time series data and the landslide state evolution model.

[0014] The second acquisition module is used to acquire the remaining time from the current displacement state to the landslide occurrence state of the target landslide body;

[0015] The early warning module is used to provide early warning of landslides based on the current displacement status and remaining time.

[0016] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the landslide early warning method described above.

[0017] Fourthly, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the landslide early warning method described above.

[0018] The above-mentioned landslide early warning method, device, computer equipment and storage medium realize a scientific and reliable geological disaster monitoring and early warning system by analyzing the displacement time series data collected by the Beidou micro-motion device and optimizing the prediction performance of the HMM model using the GMM model. This system enables real-time dynamic monitoring and accurate early warning of landslide geological disasters, significantly improving the timeliness and accuracy of landslide geological disaster early warning. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of a landslide early warning method according to an embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of the basic structure of a GMM-HMM model according to an exemplary embodiment;

[0022] Figure 3 This is a logical schematic diagram of a landslide displacement state sequence according to an exemplary embodiment;

[0023] Figure 4 This is a schematic diagram of the landslide early warning device in one embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in the present invention are only for illustrative and descriptive purposes and are not intended to limit the scope of protection of the present invention.

[0025] Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Moreover, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0026] Furthermore, the embodiments described herein are merely some, not all, of the embodiments of the invention. The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0027] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of a feature subsequently declared, but does not exclude the addition of other features. It should also be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] The following is a detailed description of this case, in conjunction with the relevant accompanying drawings in the instruction manual.

[0029] Please see Figure 1 This embodiment provides a landslide early warning method, which specifically includes the following steps:

[0030] S10: Obtain historical three-dimensional displacement time series data of historical landslide events.

[0031] It is understood that the executing entity of this invention can be a landslide early warning device, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0032] In this step, the historical landslide three-dimensional displacement time series data refers to the time series information of displacement changes of the landslide body in the three-dimensional coordinate system (X, Y, and Z axes correspond to the horizontal longitudinal, horizontal transverse, and vertical directions, respectively) during historical landslide events.

[0033] In practical applications, the BeiDou micro-motion monitoring system acquires three-dimensional displacement time-series data of landslides throughout their entire lifecycle. my country's complex and diverse geological structures, significant geographical differences, and uneven spatial and temporal distribution of climate conditions result in a wide variety of geological disasters that are widespread and severe. Among these, landslides, due to their suddenness and destructive power, have attracted considerable attention. High-precision monitoring data based on the BeiDou micro-motion device plays an increasingly important technical support role in landslide geological disaster early warning and prevention, providing crucial scientific evidence for disaster risk assessment.

[0034] S20: Based on Gaussian mixture model, hidden Markov model and historical three-dimensional displacement time series data of landslides, construct a landslide state evolution model.

[0035] In this step, in landslide displacement analysis, the Gaussian Mixture Model (GMM) can effectively characterize the multimodal distribution features of landslide displacement time series data, and realize the automatic identification and modeling of different motion states (such as the steady period, creep period, acceleration period, etc.).

[0036] In landslide state evolution modeling, the hidden states of a Hidden Markov Model (HMM) can be defined as typical evolutionary stages (such as initial stability, slow deformation, accelerated deformation, and imminent landslide), while the observed states are displacement time-series data acquired by BeiDou monitoring. Training the HMM with historical displacement time-series data yields the state transition probability matrix and observation probability distribution parameters. This application combines a Geometric Matrix (GMM) and a Hidden Markov Model (HMM) and performs joint training using historical landslide three-dimensional displacement time-series data to construct a complete landslide evolution probability model. This fused model not only accurately describes the displacement characteristics of each evolutionary stage but also reveals the transition patterns between states, providing a probabilistic analytical framework for landslide early warning.

[0037] In one embodiment of this application, a specific model construction scheme is provided. In S20, a landslide state evolution model is constructed based on a Gaussian mixture model, a hidden Markov model, and historical three-dimensional displacement time series data of landslides. This specifically includes the following steps S21-S24:

[0038] S21: Based on historical three-dimensional displacement time series data of landslides, calculate the resultant displacement rate of each monitoring point at each monitoring time.

[0039] In this step, during landslide monitoring, several monitoring points are typically deployed at key locations within the landslide body. A BeiDou micro-motion monitoring device is used to collect three-dimensional coordinate data from each monitoring point at fixed time intervals (e.g., daily). This time-series collection of three-dimensional spatial coordinate data constitutes the landslide's three-dimensional displacement time-series data. For areas where landslides have occurred, the data processing for each monitoring point first calculates the displacement increments in the X, Y, and Z directions within two adjacent monitoring periods. The spatial resultant displacement for that period is obtained through three-dimensional vector synthesis. Then, the resultant displacement is divided by the corresponding time interval (e.g., daily) to calculate the resultant displacement rate of that monitoring point in the current monitoring period. This rate parameter effectively characterizes the overall movement intensity of the monitoring point within that period and is an important quantitative indicator for evaluating the landslide deformation development stage.

[0040] S22: Based on a preset rate threshold, a label is added to the resultant displacement rate at each monitoring time, wherein the label includes at least one of the following: stable state and landslide occurrence state.

[0041] In this step, based on extensive research of historical landslide data and experience, a critical value for the resultant displacement rate, i.e., a preset rate threshold, is determined to distinguish different landslide states. When the resultant displacement rate is below this threshold, the landslide is considered to be in a stable state; when the resultant displacement rate is above this threshold, the landslide is considered to be in a state of imminent landslide, and this point is determined as the landslide initiation point. All observation points after this value are considered landslide occurrence data. The resultant displacement rate calculated at each monitoring time is compared with the preset rate threshold, and a corresponding label is added to the resultant displacement rate at that monitoring time based on the comparison result. If the resultant displacement rate is below the threshold corresponding to the stable state, it is marked as "stable state"; if the resultant displacement rate is above the threshold corresponding to the landslide occurrence state, it is marked as "landslide occurrence state".

[0042] S23: Using a Gaussian mixture model, the resultant displacement rate at each monitoring moment is divided to determine the target displacement state corresponding to the resultant displacement rate.

[0043] In this step, the resultant displacement rate data at each monitoring time is input into a Gaussian mixture model. The model then divides these resultant displacement rates into different Gaussian distributions based on the probability distribution characteristics of the data. Each Gaussian distribution corresponds to a specific displacement state, thereby determining the target displacement state corresponding to the resultant displacement rate at each monitoring time.

[0044] In practical applications, the calculated resultant displacement rate is compared with a preset rate threshold. If the resultant displacement rate at a certain monitoring moment exceeds the preset rate threshold for the first time, this is determined to be the landslide initiation point. From the landslide initiation point, all subsequent observation data are marked as "landslide occurrence state data," indicating that the landslide has entered the active deformation stage. The continuous time from the landslide initiation point to the last valid data point is denoted as the total duration t. For example, if the limit is first exceeded at the 5th sampling point, out of a total of 10 sampling points, then t = t_0. 10 -t5. Using this total duration t as a baseline, the landslide activity process is divided into several time periods. For example, if it needs to be divided into 5 stages, then the duration of each period is t / 5. This division method not only facilitates the subsequent modeling and calculation of displacement state by the GMM algorithm, but also ensures that the amount of data within each state interval remains balanced. At the same time, the uniformity of data distribution also helps to reduce the computational complexity of the HMM algorithm and improve the efficiency of model training and inference. Furthermore, due to various reasons (such as monitoring equipment failure, data transmission problems, etc.), the number of landslide occurrence state data points may be insufficient in some states. Insufficient data points may affect the subsequent analysis and modeling results. Therefore, if the number of data points corresponding to the calculated time period t is insufficient, the data is supplemented by interpolation methods to ensure that the state interval has sufficient data to meet the needs of model training.

[0045] In one embodiment of this application, a specific target displacement state calculation scheme is provided. In S23, the resultant displacement rate at each monitoring time is divided using a Gaussian mixture model to determine the target displacement state corresponding to the resultant displacement rate at each monitoring time. This specifically includes the following steps S231-S236:

[0046] S231: Obtain multiple preset displacement states.

[0047] In this step, several typical motion stages are predefined based on the kinematic characteristics of landslides. These displacement states fully characterize the evolution of landslides from initial deformation to final failure, providing a theoretical framework for subsequent state identification and early warning modeling.

[0048] Optionally, multiple preset displacement states include: stable state (displacement rate approaches zero); initial creep state (displacement increases at a deceleration rate); secondary creep state (displacement rate remains constant); accelerated creep state (displacement rate continues to increase); landslide-prone state (displacement accelerates rapidly); and failure state (the landslide body becomes unstable as a whole).

[0049] S232: Construct a Gaussian mixture model. Based on the number of multiple preset displacement states, determine the number of Gaussian distributions in the Gaussian mixture model, and each Gaussian distribution represents a preset displacement state.

[0050] In this step, the number of preset displacement states determines the number of Gaussian distributions in the Gaussian mixture model. For example, if there are 6 preset displacement states, then the Gaussian mixture model will contain 6 Gaussian distributions, each corresponding to one preset displacement state. For each state, an initial value M (M=2) of the Gaussian mixture number is set. The probability distribution characteristics of the resultant displacement rate under different displacement states are characterized by the Gaussian distributions.

[0051] S233: Initialize the mean, covariance, and mixing coefficients for each Gaussian distribution.

[0052] In this step, for each Gaussian distribution in each state, its mean, covariance, and mixing coefficients are randomly initialized. For each Gaussian distribution, the mean represents the center position of the distribution, reflecting the average level of the resultant displacement rate under the corresponding preset displacement state. The covariance describes the dispersion of the Gaussian distribution and the correlation between variables, reflecting the fluctuation of the resultant displacement rate around the mean under the corresponding preset displacement state. The mixing coefficients represent the weight of each Gaussian distribution in the entire mixing model, reflecting the probability of each preset displacement state occurring in the overall data.

[0053] S234: Train the Gaussian mixture model using the expectation-maximization algorithm to determine the model's parameters.

[0054] In this step, the Expectation-Maximization (EM) algorithm is an iterative algorithm used for parameter estimation in probabilistic models containing latent variables. In a Gaussian mixture model, the latent variables are the Gaussian distributions to which each data point (i.e., the net displacement rate at each monitoring time) belongs. The EM algorithm continuously optimizes the model parameters by alternating between two steps. Specifically, the E-step (expectation step) calculates the posterior probability of each data point belonging to each Gaussian distribution based on the current model parameters. The M-step (maximization step) updates the mean, covariance, and mixing coefficients of each Gaussian distribution based on the posterior probabilities calculated in the E-step, maximizing the likelihood function of the data. By iterating through the E-step and M-step multiple times until the model parameters converge, the optimal model parameters are determined.

[0055] S235: Input the resultant displacement rate into the trained Gaussian mixture model and calculate the probability value of each monitoring time belonging to each Gaussian distribution.

[0056] In this step, after the Gaussian mixture model is trained, the resultant displacement rate data at each monitoring time is input into the model. Based on the model parameters and the probability density function of the Gaussian distribution, the probability value of the resultant displacement rate at each monitoring time belonging to each Gaussian distribution is calculated. This probability value represents the degree of matching between the resultant displacement rate at that monitoring time and each preset displacement state.

[0057] S236: The resultant displacement rate at each monitoring moment is assigned to the displacement state corresponding to the Gaussian distribution with the largest probability value, and is denoted as the target displacement state.

[0058] In this step, for each monitoring time, the probability values ​​of its belonging to various Gaussian distributions are compared, and the Gaussian distribution with the highest probability value is selected. Since each Gaussian distribution corresponds to a preset displacement state, the resultant displacement rate at that monitoring time is assigned to the displacement state corresponding to this Gaussian distribution and recorded as the target displacement state.

[0059] The above method enables the classification of the state of the combined displacement rate at each monitoring moment, providing a foundation for subsequent landslide state evolution analysis and early warning.

[0060] S24: The hidden Markov model is trained based on the Gaussian distribution parameters of the Gaussian mixture model to obtain the landslide state evolution model.

[0061] In this step, a Hidden Markov Model (HMM) is a statistical model used to describe the state changes of a dynamic system. It includes hidden states and observable states. In landslide research, hidden states can be viewed as the different displacement states of the landslide body (such as stable, creeping, accelerating, etc.), while observable states are the monitored resultant displacement rates. During the model training phase, the target displacement state sequence preprocessed by the Hidden Markov Model (GMM) is used as training samples. The Baum-Welch algorithm is used for parameter estimation to optimize the state transition probability matrix and observation probability distribution parameters, bringing the model likelihood function to its optimal state. The trained HMM model can accurately depict the temporal patterns of landslide state transitions. Based on this model, the Viterbi decoding algorithm can be used to diagnose the current state of the landslide and predict the probability of state evolution, providing a quantitative basis for landslide early warning decisions.

[0062] In one embodiment of this application, a specific model training scheme is provided. In S24, the Hidden Markov Model is trained based on the Gaussian distribution parameters of the Gaussian mixture model to obtain the landslide state evolution model. This specifically includes the following steps S241-S244:

[0063] S241: Construct a Hidden Markov Model.

[0064] S242: Initialize the model parameters of the Hidden Markov Model, which include the state probability distribution, the state transition probability matrix, and the observation state probability distribution matrix.

[0065] For steps S241-S242, the basic structure of the model is defined, and the model parameters of the Hidden Markov Model are initialized using random initialization. These model parameters include: a state probability distribution, representing the probability that the landslide body is in different states such as a stable state or a creeping state at the initial moment; a state transition probability matrix, reflecting the likelihood of the landslide body transitioning from one displacement state to another; and an observation state probability distribution matrix, representing the probability of detecting different resultant displacement rates when the landslide body is in a certain displacement state.

[0066] S243: Iteratively train the state probability distribution and state transition probability matrix using the expectation-maximization algorithm until the difference between two adjacent iterations meets the preset condition.

[0067] In this step, during iterative training, the E-step (expectation step) calculates the probability of the system being in each hidden state and the probability of state transition at each time step based on the current model parameters. The M-step (maximization step) updates the state probability distribution and state transition probability matrix based on the probabilities calculated in the E-step, maximizing the probability of the observed sequence occurring. The E-step and M-step are repeated until the difference between the state probability distribution and the state transition probability matrix in two adjacent iterations meets a preset condition (e.g., the difference is less than a preset threshold). At this point, the model parameters are considered to have converged, and the training process ends.

[0068] S244: Determine the probability distribution matrix of the observed state based on the Gaussian distribution parameters of the Gaussian mixture model.

[0069] In this step, since the observable state in the Hidden Markov Model (HMM) is the resultant displacement rate, and the Gaussian Mixture Model (GMM) can well describe the probability distribution of the resultant displacement rate, the Gaussian distribution parameters of the GMM can be used to determine the probability distribution matrix of the observed state. Specifically, for each hidden state, based on the parameters of the corresponding Gaussian distribution, the probability of observing each resultant displacement rate in that hidden state is calculated, thus obtaining the elements of the probability distribution matrix of the observed state. This completes the determination of all parameters of the HMM, enabling it to be used for subsequent landslide state prediction and analysis.

[0070] In practical applications, the GMM-HMM model is trained based on the Baum-Welch algorithm, and the model parameters are calculated using the following formula:

[0071] λ=(π,A,B,μ(D(S t )),σ(D(S t )),S final );

[0072] Where λ is the landslide time prediction model based on GMM-HMM; π is the initial state probability distribution; A is the state transition probability matrix; B is the observed state probability distribution matrix; μ(D(S t )) is the mean of the GMM function, representing the average duration of each state; σ(D(S) t )) represents the variance in the GMM function; S final The landslide occurrence state is determined by the resultant displacement rate.

[0073] S30: Determine the current displacement state of the target landslide body based on real-time three-dimensional displacement time series data and landslide state evolution model.

[0074] In this step, the target landslide body refers to a potentially unstable slope body currently within the monitoring and early warning range. By collecting real-time three-dimensional displacement time-series data of the landslide body, calculating the real-time resultant displacement rate, and inputting it into a pre-trained landslide state evolution model, the probabilistic identification of the current motion state of the landslide can be achieved using the Viterbi algorithm, including the discrimination of typical evolutionary stages such as steady state, slow change state, and accelerated change state.

[0075] In one embodiment of this application, a specific scheme for determining the current displacement state is provided. In S30, the current displacement state of the target landslide body is determined based on the real-time three-dimensional displacement time series data and the landslide state evolution model. This specifically includes the following steps S31-S34:

[0076] S31: Obtain real-time three-dimensional displacement time series data of the target landslide body at the current monitoring time.

[0077] In this step, the target landslide refers to the landslide that is being monitored, whose stability needs to be assessed, and whose landslide probability needs to be predicted. The current monitoring time refers to the current moment. Real-time displacement information of the target landslide in three-dimensional space is acquired. The real-time three-dimensional displacement time-series data reflects the change in the landslide's position relative to a certain initial location at the current moment.

[0078] S32: Based on real-time three-dimensional displacement time series data, determine the real-time resultant displacement rate at the current monitoring moment.

[0079] In this step, based on the acquired real-time three-dimensional displacement time-series data and the displacement time-series data from the previous monitoring moment, the displacement changes of the landslide body in the X, Y, and Z directions between these two moments are calculated. Then, using a vector synthesis method, the displacement changes in these three directions are combined into a total displacement change, which is then divided by the time interval between the two moments to obtain the real-time combined displacement rate at the current monitoring moment. This rate can comprehensively reflect the speed of movement of the landslide body at the current moment.

[0080] S33: Input the real-time combined displacement rate into the landslide state evolution model and output the displacement state sequence of the target landslide body.

[0081] In this step, the real-time combined displacement rate at the current monitoring moment is input into the trained landslide state evolution model. The model, based on its algorithm and parameters, outputs a displacement state sequence. This sequence represents the changes in the displacement state of the target landslide body from the past to the present and potentially into the future. For example, the sequence might be "steady state - creeping state - accelerated deformation state," etc.

[0082] S34: Determine the current displacement state of the target landslide body based on the displacement state sequence.

[0083] In this step, after obtaining the displacement state sequence, the displacement state corresponding to the current monitoring time is extracted and determined as the current displacement state of the target landslide. For example, if the displacement state sequence shows that the current state is "accelerated deformation state", then it can be determined that the target landslide is currently in an accelerated deformation stage, which may mean that the risk of landslide is increasing and further attention and corresponding preventive measures are needed.

[0084] By using the above methods, we can promptly grasp the actual situation of the target landslide body, providing a basis for landslide early warning and disaster prevention.

[0085] S40: Obtain the remaining time from the current displacement state to the landslide occurrence state of the target landslide body.

[0086] In this step, after identifying the current displacement state of the target landslide, the remaining time from the current state to the imminent landslide state is estimated, providing a key time-dimensional decision-making basis for disaster early warning.

[0087] In one embodiment of this application, a specific scheme for calculating the remaining time is provided. In S40, that is, obtaining the remaining time from the current displacement state to the landslide occurrence state of the target landslide body, the following steps S41-S42 are specifically included:

[0088] S41: Input the current displacement state and the preset state transition time matrix into the Dijkstra algorithm to generate the shortest path from the current state to the landslide occurrence state.

[0089] S42: Determine the remaining time based on the shortest path time interval.

[0090] For steps S41-S42, during the research and monitoring of landslides, a state transition time matrix is ​​pre-defined based on a large amount of historical data and the analysis of landslide evolution patterns. This matrix records the average time required for a landslide to transition from one displacement state to another. The elements in the matrix can represent the time required to transition from a stable state to a creeping state, and the time required to transition from a creeping state to an accelerated deformation state, etc.

[0091] Using the current displacement state of the target landslide as the starting node and the landslide occurrence state as the target node, and employing a pre-defined state transition time matrix as the weight information for the edges in the graph, the algorithm inputs these values ​​into Dijkstra's algorithm. The algorithm then identifies the path from the current displacement state to the landslide occurrence state among all possible state transition paths. This shortest path represents the sequence of state changes the landslide undergoes in the most probable scenario, from its current state to the landslide occurrence state.

[0092] Furthermore, after obtaining the shortest path from the current state to the landslide occurrence state, the time intervals corresponding to each state transition along this path are added together. The sum is the predicted time required for the landslide to develop from the current state to the final landslide, which is the remaining time. For example, the shortest path is "current state (accelerated deformation state) - landslide-landslide state". The transition time from the accelerated deformation state to the landslide-landslide state is 2 days, and the transition time from the landslide-landslide state to the landslide-landslide state is 1 day, so the remaining time is 3 days.

[0093] S50: Provide landslide warning for the target landslide based on the current displacement status and remaining time.

[0094] In this step, early warning thresholds are pre-set based on geological characteristics and historical data, including preset early warning states and preset safe time thresholds. Based on these early warning thresholds, the current displacement state of the target landslide body and the remaining time to reach the landslide occurrence state are dynamically assessed to dynamically evaluate the landslide risk, thereby achieving landslide early warning.

[0095] The above methods can effectively and promptly reduce the losses caused by landslides.

[0096] In practical applications, a tiered early warning and response mechanism can be established based on the remaining time assessment results of landslide evolution time prediction models. When the model predicts that the remaining time is within a critical threshold (e.g., 24 hours), the system automatically triggers a red alert, activates the emergency response plan, and organizes the emergency evacuation and resettlement of people in the danger zone. When the predicted remaining time is in the mid-term stage (e.g., 1-7 days), an orange alert is issued, and traffic control and engineering emergency measures are implemented. For situations where the predicted remaining time is relatively long (e.g., more than 7 days), a blue alert is issued, with a focus on strengthening monitoring frequency and patrol efforts. This tiered early warning system based on accurate time prediction can significantly improve the scientific nature and timeliness of landslide disaster emergency management, minimizing disaster losses.

[0097] In one embodiment of this application, a specific landslide early warning scheme is provided. In S50, that is, based on the current displacement state and the remaining time, a landslide early warning is given to the target landslide body, which specifically includes the following steps S51-S52:

[0098] S51: If the current displacement state of the target landslide is in a preset warning state and / or the remaining time is less than a preset safe time threshold, it is determined that the target landslide has a landslide risk.

[0099] S52: Based on the current displacement status and remaining time, generate landslide early warning information and send the landslide early warning information to the maintenance personnel's terminal.

[0100] For S51-S52, the preset warning state and preset safe time threshold are pre-set warning standards, usually determined comprehensively based on factors such as historical landslide data, geological conditions, surrounding environment, and the potential degree of harm, representing a safety limit. When the current displacement state is in the preset warning state and / or the remaining time is less than the preset safe time threshold, it means that the probability of a landslide is high and time is of the essence, indicating a landslide risk that requires immediate action. Once a landslide risk is confirmed, the system automatically generates and sends a landslide warning message containing key information. This information includes the specific location of the target landslide, the current displacement state, and the predicted remaining time, allowing maintenance personnel to fully understand the landslide situation. The system will send the generated landslide alarm information to their terminal devices, such as mobile phones, tablets, or dedicated monitoring and early warning equipment. In this way, maintenance personnel can obtain the warning information immediately and take appropriate measures, such as organizing personnel evacuation, conducting on-site investigations, and carrying out rescue operations, thereby reducing the potential losses caused by the landslide.

[0101] The above methods can effectively monitor and provide early warnings of landslide disasters in real time, ensuring the safety of people's lives and property and the stable operation of infrastructure.

[0102] As can be seen, the above scheme employs a GMM-optimized HMM algorithm to construct a landslide early warning model, which, through training with historical data, enables the prediction of the timing of landslide geological disasters, overcoming the limitations of traditional monitoring methods that can only collect data in real time. This not only achieves real-time dynamic monitoring and early warning of landslide geological disasters but also significantly improves the accuracy and timeliness of the early warning system.

[0103] In one embodiment, the landslide early warning method based on GMM-HMM provided in this application specifically includes the following steps:

[0104] Step 1: Collect three-dimensional displacement time-series data of the landslide throughout its entire lifecycle using a BeiDou micro-motion device, and perform data preprocessing. Based on the three-dimensional displacement time-series data of each monitoring point at each monitoring time (day), calculate the resultant displacement rate. The formula for calculating the resultant displacement rate is:

[0105]

[0106] Where S is the resultant displacement rate; t is the current monitoring time; t-1 is the previous monitoring time (here, "day" is the smallest unit); (x t ,y t ,z t (x) represents the three-dimensional displacement time series data at time t; t-1 ,y t-1 ,z t-1 ) represents the three-dimensional displacement time series data at time t-1.

[0107] Step 2: Reconstruct the landslide displacement time series data based on the Gaussian Mixture (GMM) algorithm to perform multi-displacement state division, and set the number of displacement states N to 6, and the initial value of the Gaussian mixture number M in each displacement state to 2.

[0108] Step 3: The Baum-welch algorithm is used to train the three-dimensional landslide displacement time series data that has been divided into states, and a landslide evolution state model is constructed using a mixture of Gaussian hidden Markov models (GMM-HMM).

[0109] Step 3.1: First, randomly initialize π, A, and B. Then train the parameters π and A, using the condition that the difference between two adjacent iterations meets certain requirements to obtain updated π and A. The parameter B is calculated from the landslide displacement time series using the Gaussian distribution parameter estimation formulas (2)-(3).

[0110] The formula for calculating a Gaussian mixture model (GMM) is as follows:

[0111]

[0112] The formula for calculating the weighting coefficient is:

[0113]

[0114] Among them, b j (O) represents the probability of observing data O in state j; M represents the number of Gaussian mixture distributions; C j,m The mixture weights of the m-th Gaussian distribution for the j-th state (all C j,m The sum of these is 1); ξ(O,μ) j,m ,σ j,m Given a Gaussian distribution function, calculate the probability density of the resultant displacement velocity O under this Gaussian distribution; μ j,m σ is the mean of the m-th Gaussian distribution in the j-th state; j,m Let be the variance of the m-th Gaussian distribution in the j-th state.

[0115] like Figure 2 The diagram shows the basic structure of the GMM-HMM model, used to describe the state transitions and observation probability distributions during landslide evolution. The state transition probability matrix A represents the hidden states in the landslide evolution, such as S1 (stable state), S2 (initial deformation), and S3 (accelerated deformation). In the observation state probability distribution B, each hidden state S... t This corresponds to a Gaussian Mixture Model (GMM), which describes the probability distribution of the observed displacement time series data under this condition. The GMM is composed of a linear combination of multiple Gaussian distributions (normal distributions) and can fit complex displacement time series data patterns.

[0116] Step 3.2: Use the Viterbi algorithm to perform state decoding on the real-time acquired three-dimensional landslide displacement time series data to obtain the time series O. t The corresponding displacement state sequence S t ;

[0117] Step 3.3: After training with the Baum-welch algorithm, the values ​​of Gaussian mixture M, π, and A are obtained.

[0118] Mean duration of state μ(D(S) t The calculation formula is:

[0119]

[0120] Wherein, D(S) t ) is state S t The duration of the "slow deformation" state (e.g., the average duration of the "slow deformation" state is 10 days); L is the duration of S in historical data. t Total number of occurrences; μ(D(S) t )) represents state S t The average duration.

[0121] State duration standard deviation σ(D(S))t The calculation formula is:

[0122]

[0123] Wherein, σ(D(S) t )) represents state S t The range of fluctuation in duration.

[0124] Final state S final for:

[0125] S final =S N (6)

[0126] Among them, S N This refers to the final pre-slide state (i.e., the state in which a landslide occurs) defined in the model.

[0127] Step 4: Predict the possible time of landslide disaster based on model prediction results.

[0128] Step 4.1: State decoding of real-time three-dimensional displacement time series data. The state of the real-time three-dimensional displacement time series data is first decoded using formula (12), that is, the current evolution stage of the landslide is determined.

[0129] (1) Initialize (t=1), calculate the maximum probability of each state at the initial time:

[0130] δ1(i)=π i b i (O1), i = 1, 2, ..., N;

[0131] The predecessor state is initialized to empty:

[0132] Ψ1(i)=0.

[0133] (2) Recursive calculation (t=2 to t=T)

[0134] By combining formula (7) to perform recursive calculations, the maximum probability of each state at each time point and its corresponding predecessor state are recorded.

[0135] For each time t and each state i, the calculation formula is:

[0136]

[0137] That is, select the path that maximizes the current probability from all possible predecessor states j.

[0138] At the same time, record the state i at time t, which is most likely to transition from state j at time t-1, and record the sequence number of the state with the largest transition.

[0139] The formula for calculating the optimal precursor state is as follows:

[0140] Ψ t (i)=argmax[δ t-1 (j)a j,i (8)

[0141] (3) Termination (t=T)

[0142] The maximum value and state number (i.e., optimal state and probability) of the final time T are obtained through formulas (9)-(10). The specific calculation formulas are as follows:

[0143]

[0144] (4) Backtracking path (t=T-1 to t=1)

[0145] The optimal path is found by backtracking using formula (11).

[0146] According to Ψ t Backtracking the optimal path:

[0147]

[0148] Finally, the complete optimal state sequence is obtained:

[0149] S end =(S t-l ,...,S t-2 ,S t-1 ,S T (12)

[0150] Where l is the truncation position of the last segment of the sequence; T is the current time, and S is the position of the last segment of the sequence. end The value with the largest number of occurrences is used as the current state S. current .

[0151] like Figure 3 The diagram shown is a logical schematic of the landslide displacement state sequence, where S t-l The landslide state at the previous moment (e.g., tl day is in a "slow deformation" state); S t-l+1 S represents the state at the current time (i.e., the state at time t); t-i (i < l) represents a subsequence of states of length l that traces back from time tl; l is the preset sliding window length (e.g., the state sequence of the most recent seven days). By extracting the most recent l states (e.g., the state changes of the most recent 7 days), it is determined whether the landslide is accelerating towards the imminent landslide state.

[0152] Step 4.2: Use Dijkstra's algorithm to calculate the remaining time (RUL, Remaining Useful Life) from the current state to the final landslide-prone state (i.e., the state in which the landslide occurs).

[0153] Specifically, the real-time three-dimensional displacement time series data is state decoded and RUL calculated according to calculation formulas (12) and (13). The state values ​​are substituted into formula (13), and the S is obtained using the Dijkstra algorithm. current Arrival at S final The shortest time can be used to calculate the time interval between the current state and the last landslide occurrence state, i.e., RUL.

[0154] The calculation formula is:

[0155]

[0156] Wherein, D(S) t ) is state S t The duration of the state; μ() is the mean function, representing the expected value of the state duration.

[0157] Calculate from S current To S final The sum of the durations of all intermediate states is the remaining time.

[0158] Finally, assess the current displacement status and remaining time of the landslide. If the displacement exceeds the safety threshold, activate the alarm mechanism; otherwise, continue monitoring.

[0159] The landslide early warning method based on GMM-HMM provided in this application analyzes the displacement time series data collected by the Beidou micro-motion device, optimizes the prediction performance of the HMM model using the GMM model, and constructs a scientific and reliable geological disaster monitoring and early warning system. This system enables real-time dynamic monitoring and accurate early warning of landslide geological disasters, significantly improving the timeliness and accuracy of landslide geological disaster early warning.

[0160] In one embodiment, a landslide early warning device is provided, which corresponds one-to-one with the landslide early warning methods described in the above embodiments. For example... Figure 4 As shown, the landslide early warning device 100 includes: a first acquisition module 101, a construction module 102, a determination module 103, a second acquisition module 104, and an early warning module 105. Detailed descriptions of each functional module are as follows:

[0161] The first acquisition module 101 is used to acquire historical three-dimensional displacement time series data of historical landslide events.

[0162] Module 102 is used to construct a landslide state evolution model based on Gaussian mixture model, hidden Markov model and historical three-dimensional displacement time series data of landslides;

[0163] The determination module 103 is used to determine the current displacement state of the target landslide body based on the real-time three-dimensional displacement time series data and the landslide state evolution model of the target landslide body;

[0164] The second acquisition module 104 is used to acquire the remaining time from the current displacement state to the landslide occurrence state of the target landslide body;

[0165] The early warning module 105 is used to provide early warning of landslides based on the current displacement status and remaining time.

[0166] In one embodiment, the construction module 102 is specifically used for:

[0167] Based on historical three-dimensional displacement time series data of landslides, the resultant displacement rate of each monitoring point at each monitoring time is calculated;

[0168] Based on a preset rate threshold, a label is added to the resultant displacement rate at each monitoring time, wherein the label includes at least one of the following: stable state and landslide occurrence state;

[0169] The resultant displacement rate at each monitoring moment is divided using a Gaussian mixture model to determine the target displacement state corresponding to the resultant displacement rate.

[0170] The Hidden Markov Model is trained using the Gaussian distribution parameters of the Gaussian mixture model to obtain a landslide state evolution model.

[0171] In one embodiment, the construction module 102 is further configured to:

[0172] Obtain multiple preset displacement states;

[0173] Construct a Gaussian mixture model, and determine the number of Gaussian distributions in the Gaussian mixture model based on the number of multiple preset displacement states, with each Gaussian distribution representing a preset displacement state;

[0174] Initialize the mean, covariance, and mixing coefficients for each Gaussian distribution;

[0175] The Gaussian mixture model is trained using the expectation-maximization algorithm to determine the model's parameters;

[0176] The combined displacement rate is input into the trained Gaussian mixture model to calculate the probability value of each monitoring time belonging to each Gaussian distribution;

[0177] The resultant displacement rate at each monitoring moment is assigned to the displacement state corresponding to the Gaussian distribution with the highest probability value, and is denoted as the target displacement state.

[0178] In one embodiment, the construction module 102 is further configured to:

[0179] Constructing a Hidden Markov Model;

[0180] Initialize the model parameters of the Hidden Markov Model, which include the state probability distribution, the state transition probability matrix, and the observation state probability distribution matrix;

[0181] The state probability distribution and state transition probability matrix are iteratively trained using the expectation-maximization algorithm until the difference between two adjacent iterations meets the preset condition.

[0182] The probability distribution matrix of the observed state is determined based on the Gaussian distribution parameters of the Gaussian mixture model.

[0183] In one embodiment, the determining module 103 is specifically used for:

[0184] Acquire real-time three-dimensional displacement time series data of the target landslide body at the current monitoring time;

[0185] Based on real-time three-dimensional displacement time-series data, the real-time resultant displacement rate at the current monitoring moment is determined;

[0186] Input the real-time combined displacement rate into the landslide state evolution model and output the displacement state sequence of the target landslide body;

[0187] Based on the displacement state sequence, the current displacement state of the target landslide body is determined.

[0188] In one embodiment, the second acquisition module 104 is specifically used for:

[0189] Input the current displacement state and the preset state transition time matrix into the Dijkstra algorithm to generate the shortest path from the current state to the landslide occurrence state;

[0190] The remaining time is determined based on the time interval of the shortest path.

[0191] In one embodiment, the early warning module 105 is specifically used for:

[0192] If the current displacement state of the target landslide is in a preset warning state and / or the remaining time is less than a preset safe time threshold, it is determined that the target landslide has a landslide risk.

[0193] Based on the current displacement status and remaining time, a landslide early warning message is generated and sent to the maintenance personnel's terminal.

[0194] This invention provides a landslide early warning device 100. By analyzing the displacement time series data collected by the Beidou micro-motion device, and using the GMM model to optimize the prediction performance of the HMM model, a scientific and reliable geological disaster monitoring and early warning system is constructed. This system realizes real-time dynamic monitoring and accurate early warning of landslide geological disasters, significantly improving the timeliness and accuracy of landslide geological disaster early warning.

[0195] Specific limitations regarding landslide early warning devices can be found in the limitations of landslide early warning methods described above, and will not be repeated here. Each module in the aforementioned landslide early warning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0196] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the landslide early warning method described above.

[0197] In one embodiment, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the aforementioned landslide early warning method.

[0198] It should be noted that the functions or steps that can be implemented by the readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0199] 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. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0200] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0201] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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 spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A landslide warning method characterized by, The method comprises the following steps: acquiring historical landslide three-dimensional displacement time series data of historical landslide events; constructing a landslide state evolution model based on a Gaussian mixture model, a hidden Markov model and the historical landslide three-dimensional displacement time series data; determining a current displacement state of a target landslide body based on real-time three-dimensional displacement time series data of the target landslide body and the landslide state evolution model; acquiring a remaining time for the target landslide body to reach a landslide occurrence state from the current displacement state; issuing a landslide warning for the target landslide body based on the current displacement state and the remaining time.

2. The method of claim 1, wherein, The step of constructing a landslide state evolution model based on a Gaussian mixture model, a hidden Markov model and the historical landslide three-dimensional displacement time series data comprises the following steps: calculating a combined displacement rate of each monitoring point at each monitoring time based on the historical landslide three-dimensional displacement time series data; adding a label to the combined displacement rate of each monitoring time based on a preset rate threshold, wherein the label comprises at least one of the following: a stable state and a landslide occurrence state; dividing the combined displacement rate of each monitoring time by the Gaussian mixture model to determine a target displacement state corresponding to the combined displacement rate; training the hidden Markov model based on Gaussian distribution parameters of the Gaussian mixture model to obtain the landslide state evolution model.

3. The method of claim 2, wherein, The step of dividing the combined displacement rate of each monitoring time by the Gaussian mixture model to determine a target displacement state corresponding to the combined displacement rate comprises the following steps: acquiring a plurality of preset displacement states; constructing a Gaussian mixture model, determining the number of Gaussian distributions in the Gaussian mixture model based on the number of the plurality of preset displacement states, and each Gaussian distribution representing a preset displacement state; initializing the mean, covariance and mixing coefficient of each Gaussian distribution; training the Gaussian mixture model by an expectation maximization algorithm to determine the parameters of the model; inputting the combined displacement rate into the trained Gaussian mixture model to calculate the probability value of each Gaussian distribution to which each monitoring time belongs; dividing the combined displacement rate of each monitoring time into a displacement state corresponding to the Gaussian distribution with the maximum probability value, and recording the displacement state as a target displacement state.

4. The method of claim 2, wherein, The step of training the hidden Markov model based on Gaussian distribution parameters of the Gaussian mixture model to obtain the landslide state evolution model comprises the following steps: constructing a hidden Markov model; initializing model parameters of the hidden Markov model, wherein the model parameters comprise a state probability distribution, a state transition probability matrix and an observation state probability distribution matrix; iteratively training the state probability distribution and the state transition probability matrix by an expectation maximization algorithm until the difference between adjacent two iterations meets a preset condition; determining the observation state probability distribution matrix based on the Gaussian distribution parameters of the Gaussian mixture model.

5. The method of claim 1, wherein, The step of determining a current displacement state of a target landslide body based on real-time three-dimensional displacement time series data of the target landslide body and the landslide state evolution model comprises the following steps: acquiring real-time three-dimensional displacement time series data of the target landslide body at a current monitoring time; determine a real-time combined displacement rate of the current monitoring moment based on the real-time three-dimensional displacement time series data; input the real-time combined displacement rate into the landslide state evolution model, and output a displacement state sequence of the target landslide body; determine a current displacement state of the target landslide body based on the displacement state sequence.

6. The method of claim 1, wherein, The step of acquiring the residual time of the target landslide body from the current displacement state to a landslide occurrence state specifically includes: input the current displacement state and a preset state transition time matrix into a Dijkstra algorithm to generate a shortest path from the current state to the landslide occurrence state; determine the residual time based on a time interval of the shortest path.

7. The method of claim 1, wherein, The step of performing landslide warning on the target landslide body based on the current displacement state and the residual time specifically includes: if the current displacement state of the target landslide body is in a preset warning state and / or the residual time is less than a preset safety time threshold, determine that the target landslide body has a landslide risk; generate a landslide warning prompt information based on the current displacement state and the residual time, and send the landslide warning prompt information to an operation and maintenance personnel terminal.

8. A landslide warning device, characterized by, The landslide warning method comprises: a first acquisition module configured to acquire historical landslide three-dimensional displacement time series data of historical landslide events; a construction module configured to construct a landslide state evolution model based on a Gaussian mixture model, a hidden Markov model, and the historical landslide three-dimensional displacement time series data; a determination module configured to determine a current displacement state of a target landslide body based on real-time three-dimensional displacement time series data of the target landslide body and the landslide state evolution model; a second acquisition module configured to acquire a residual time of the target landslide body from the current displacement state to a landslide occurrence state; a warning module configured to perform landslide warning on the target landslide body based on the current displacement state and the residual time.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor implements the steps of the landslide warning method according to any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium having a program or instructions stored thereon, the program or instructions being executed by a processor to implement the steps of the landslide warning method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Matrix type landslide monitoring system and monitoring method

    CN121855438A