Unstable slope monitoring and early warning method, system, equipment and medium

Abstract

The invention relates to an unstable slope monitoring and early warning method, system and device and a medium, and belongs to the technical field of geological disaster monitoring, and the monitoring and early warning method comprises the steps: collecting original monitoring data of a target slope region, carrying out the edge preprocessing, and generating a standardized data package; transmitting the standardized data packet to a cloud server for integrity verification and logic verification, and generating a verified data set; performing multi-modal data fusion on the verified data set, constructing a displacement-rainfall-microseismic time sequence incidence matrix, extracting displacement acceleration, rainfall cumulative intensity and microseismic event frequency features, and generating a fusion feature vector set; inputting the fusion feature vector set into a pre-trained risk prediction model for real-time risk prediction, and outputting a landslide risk index and a displacement trend prediction result; and determining a corresponding early warning level according to the landslide risk index, and triggering an automatic response strategy corresponding to the early warning level. According to the invention, the real-time performance and accuracy of landslide early warning can be improved.

Description

Technical Field

[0001] This application relates to the field of geological disaster monitoring technology, and in particular to a method, system, equipment and medium for monitoring and early warning of unstable slopes. Background Technology

[0002] In the construction of railways, highways, and towns in mountainous areas prone to geological disasters, dynamic monitoring and early warning of unstable slopes (such as ancient landslides and potential landslide areas) is a core challenge to ensure project safety and the protection of people's lives and property. A landslide is a rapid sliding of soil or rock mass caused by gravity, typically occurring under the influence of natural factors such as rainfall and earthquakes, and can pose a significant threat to surrounding buildings, transportation facilities, and human lives. With the expansion of highways, railways, and urban construction in mountainous areas, the need for monitoring and early warning of unstable slopes is becoming increasingly urgent.

[0003] Currently, most common landslide monitoring systems rely on traditional statistical analysis methods or simple threshold models. These methods cannot effectively capture the complex temporal relationships and nonlinear characteristics of landslide events, especially in dynamically changing slope environments. They are unable to provide sufficient risk prediction capabilities, resulting in untimely warnings or inaccurate predictions when landslide disasters occur. Summary of the Invention

[0004] To improve the real-time performance and accuracy of landslide early warning, this application provides a method, system, equipment, and medium for monitoring and early warning of unstable slopes.

[0005] Firstly, this application provides a method for monitoring and early warning of unstable slopes, employing the following technical solution: An unstable slope monitoring and early warning method, the monitoring and early warning method comprising: Collect raw monitoring data of the target slope area; the raw monitoring data includes displacement data, rainfall data, and microseismic signal data; The raw monitoring data is preprocessed to generate standardized data packets; The standardized data packets are transmitted to the cloud server for integrity and logical verification to generate a verified dataset. Multimodal data fusion was performed on the verified dataset to construct a displacement-rainfall-microseismic time series correlation matrix, and features of displacement acceleration, cumulative rainfall intensity and microseismic event frequency were extracted to generate a fused feature vector set. The fused feature vector set is input into a pre-trained risk prediction model for real-time risk prediction, and the landslide risk index and displacement trend prediction results are output. The corresponding early warning level is determined based on the landslide risk index, and the corresponding automated response strategy is triggered based on the early warning level.

[0006] By adopting the above technical solutions, precise monitoring and early warning of unstable slopes are achieved through multi-layered technical means. Through edge computing and cloud-based collaborative processing, the system can monitor the dynamic changes of the slope area in real time. Based on multimodal data fusion and deep learning technology, the system can accurately assess landslide risks and predict future displacement trends in real time. Furthermore, the system has an efficient automated response mechanism that can quickly implement emergency measures, ensuring more efficient and timely disaster prevention and control.

[0007] Optionally, the step of performing edge preprocessing on the raw monitoring data to generate standardized data packets includes: Based on the extraction of UTC standard time from BeiDou satellite signals, the local timestamps of the original monitoring data are offset and compensated to generate a monitoring data stream with a unified time reference. Wavelet threshold denoising is performed on the monitoring data stream; The mean and standard deviation of the denoised monitoring data stream are calculated based on the sliding window statistical calculation, and abnormal data points exceeding the preset range are removed using the 3σ principle to generate an effective data set. The floating-point values ​​in the valid data set are converted into fixed-point integers, compressed and encoded using a static Huffman dictionary, and encapsulated into a standardized data packet including a timestamp, device identifier, and CRC32 checksum.

[0008] By adopting the above technical solutions, it is ensured that the monitoring data collected from multiple sensors can be transmitted to the back-end system in a standardized form after precise synchronization, noise reduction, anomaly removal and efficient encoding processing. This greatly improves data quality, reduces transmission burden, and enhances the system's response speed and accuracy in monitoring unstable slopes.

[0009] Optionally, the steps of performing multimodal data fusion on the verified dataset, constructing a displacement-rainfall-microseismic time series correlation matrix, extracting displacement acceleration, cumulative rainfall intensity, and microseismic event frequency features, and generating a fused feature vector set include: performing time alignment processing on the displacement data, rainfall data, and microseismic signal data in the verified dataset to obtain displacement data sequences, rainfall data sequences, and microseismic signal data sequences; The displacement data sequence is subjected to second-order difference processing to calculate instantaneous acceleration, thereby obtaining displacement acceleration characteristics; A sliding window cumulative calculation is performed on the rainfall data sequence to generate a cumulative rainfall intensity value; The cumulative rainfall intensity value is encoded according to a preset grading rule to obtain a rainfall intensity encoding vector; Event determination is performed on the microseismic signal data sequence, the number of microseismic pulses exceeding a set threshold within a preset time window is counted, and the number of pulses is normalized to obtain the frequency characteristics of microseismic events. The displacement acceleration features, rainfall intensity encoding vector, and microseismic event frequency features are combined in order of time window to construct a time-series correlation matrix; The time-series correlation matrix is ​​standardized, and the high-dimensional features are mapped to a three-dimensional space using a nonlinear dimensionality reduction algorithm to generate a fused feature vector set.

[0010] By adopting the above technical solution, time-series data from different sensors are effectively fused to generate an accurate and simplified fusion feature vector set, which significantly improves the quality and operability of the data and provides solid data support for subsequent landslide risk prediction and early warning.

[0011] Optionally, the system also includes a training step for the risk prediction model, the training step comprising: Based on the fused feature vector set and historical landslide event labels, a timestamped training dataset and a validation dataset are constructed; the training dataset is divided into an input sequence and an output label sequence according to a preset time window length; The input sequence is standardized to generate a standardized training tensor; Construct a neural network model comprising an LSTM network layer, a Dropout layer, and a fully connected layer; wherein the LSTM network layer is configured to return the output of a complete time step; The neural network model is trained based on the standardized training tensor and the output label sequence, and the model parameters are optimized to obtain the risk prediction model. The risk prediction model is validated and its parameters are corrected based on the validation dataset until the model's loss function meets the preset conditions, thus obtaining the trained risk prediction model.

[0012] By adopting the above technical solutions, a highly efficient and accurate risk prediction model is finally obtained. This model can accurately predict the landslide risk index and displacement trend based on real-time monitoring data, has strong time series learning capabilities, and can effectively avoid overfitting. It is suitable for complex landslide risk prediction scenarios.

[0013] Optionally, the LSTM network layer includes a first LSTM layer and a second LSTM layer. The activation function of the first LSTM layer is a hyperbolic tangent function, and the output dimension is consistent with the time step of the input sequence. The output of the second LSTM layer is the global feature vector of the input sequence in the time dimension. The fully connected layer is configured with two neurons, which correspond to the probability output of the landslide risk index and the regression prediction output of the displacement trend, respectively.

[0014] Optionally, after the steps of outputting the landslide risk index and displacement trend prediction results, the following may also be included: Based on the displacement trend prediction results, a continuous displacement heat map is generated using an interpolation algorithm; A dynamic trend animation is synthesized based on the time series data of the displacement trend prediction results, and timestamps and risk level labels are superimposed; the continuous displacement heat map and dynamic trend animation are loaded into a 3D GIS engine to build an interactive visualization interface.

[0015] By adopting the above technical solutions, a comprehensive and interactive landslide disaster monitoring system was constructed. This system not only reflects the spatial distribution of landslides but also dynamically displays their changing processes, helping decision-makers respond promptly and take necessary emergency measures. Simultaneously, the system supports user interaction with data, further enhancing operational flexibility and practicality.

[0016] Secondly, this application provides an unstable slope monitoring and early warning system, which adopts the following technical solution: An unstable slope monitoring and early warning system, comprising: The acquisition module is used to collect raw monitoring data of the target slope area; the raw monitoring data includes displacement data, rainfall data, and microseismic signal data. The preprocessing module is used to perform edge preprocessing on the raw monitoring data to generate standardized data packets; The verification module is used to transmit the standardized data packet to the cloud server for integrity verification and logical verification, and generate a verified dataset. The feature fusion module is used to perform multimodal data fusion on the verified dataset, construct a displacement-rainfall-microseismic time series correlation matrix, extract displacement acceleration, cumulative rainfall intensity and microseismic event frequency features, and generate a fused feature vector set; the risk prediction module is used to input the fused feature vector set into a pre-trained risk prediction model for real-time risk prediction, and output landslide risk index and displacement trend prediction results. The early warning module is used to determine the corresponding early warning level based on the landslide risk index and trigger the automatic response strategy corresponding to the early warning level.

[0017] Optionally, the monitoring and early warning system further includes: The heatmap generation module is used to generate a continuous displacement heatmap based on the displacement trend prediction results using an interpolation algorithm; the animation synthesis module is used to synthesize a dynamic trend animation based on the time series data of the displacement trend prediction results, and to overlay timestamps and risk level labels. The visualization interface building module is used to load the continuous displacement heat map and dynamic trend animation into the 3D GIS engine to build an interactive visualization interface.

[0018] Thirdly, this application provides a computer device, which adopts the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0019] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0020] In summary, this application includes at least one of the following beneficial technical effects: by comprehensively collecting multi-source data on displacement, rainfall, and microseismic events, ensuring data reliability through edge-end denoising and alignment and cloud-based verification, constructing a time-series correlation matrix using multi-modal fusion to extract key dynamic features, and combining a deep learning model to achieve accurate prediction of landslide risk index and displacement trend, and triggering graded emergency strategies based on risk levels, it significantly improves the real-time performance and accuracy of landslide early warning in complex geological environments, providing a full-chain solution from data perception and intelligent analysis to automated response for disaster prevention and control in mountainous railways, highways, and urban areas, effectively reducing casualties and economic losses, and enhancing the geological safety resilience of major projects. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the first process of an unstable slope monitoring and early warning method according to one embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the second process of an unstable slope monitoring and early warning method according to one embodiment of this application.

[0023] Figure 3 This is a schematic diagram of the third process of an unstable slope monitoring and early warning method according to one embodiment of this application.

[0024] Figure 4 This is a schematic diagram of the fourth process of an unstable slope monitoring and early warning method according to one embodiment of this application.

[0025] Figure 5 This is a schematic diagram of the fifth process of an unstable slope monitoring and early warning method according to one embodiment of this application. Detailed Implementation

[0026] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-5 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0027] This application discloses a method for monitoring and early warning of unstable slopes.

[0028] Reference Figure 1 An unstable slope monitoring and early warning method, comprising: Step S101: Collect raw monitoring data of the target slope area; The raw monitoring data includes displacement data, rainfall data, and microseismic signal data. In one embodiment of this application, to prevent potential unstable slopes from threatening surrounding buildings and the environment, multiple sensors can be deployed to comprehensively monitor various environmental changes in the target slope area. Specific sensors include GNSS displacement sensors, deep displacement gauges, rain gauges, and microseismic sensors, each undertaking different data acquisition tasks.

[0029] Specifically, GNSS displacement sensors are responsible for monitoring the overall displacement of the slope and can provide accurate displacement data; deep displacement gauges detect displacement in the deep underground layers and can reflect changes in the deep structure of the slope in real time, which is especially important for early warning of deep landslides; rain gauges are used to monitor precipitation in real time, and precipitation is an important factor that induces landslides, so changes in precipitation will directly affect the risk of landslides; microseismic sensors detect microseismic activity in the slope area and identify small earthquakes and stress changes underground, which is of great importance for judging potential triggering factors for landslides.

[0030] Understandably, by using these sensors in combination, the system can comprehensively and accurately capture various dynamic information about the slope area from different perspectives. The monitoring data from each sensor will serve as the basis for subsequent system analysis and risk assessment. The core of this step is to acquire real-time, comprehensive raw monitoring data through the integrated deployment of a sensor network, providing rich raw data support for subsequent analysis and early warning.

[0031] Step S102: Perform edge preprocessing on the raw monitoring data to generate standardized data packets; Edge preprocessing of raw monitoring data is a crucial step in ensuring data quality and usability. This process begins with noise removal from sensor data from different sources. Wavelet threshold filtering is a common method for denoising signals using wavelet transform. Wavelet transform effectively separates noise components from signals, eliminating irrelevant noise and thus ensuring data accuracy.

[0032] Secondly, to ensure data time synchronization and temporal consistency, UTC (Coordinated Universal Time) timestamps can be used to align data from different sensors. Because different sensors may have different response speeds and time accuracy, there may be deviations in the temporal comparison of data. Timestamp alignment can ensure that all sensor data are compared and fused at the same point in time, avoiding data corruption and inconsistency.

[0033] Furthermore, the 3σ principle can be used to remove outliers from the data. The 3σ principle, based on statistical methods, is used to detect and remove data points that deviate from the normal distribution. Outliers typically reflect anomalies or errors in the data collection process; removing these outliers helps improve the accuracy and reliability of the data.

[0034] In addition, the data needs to be compressed and encapsulated. Huffman coding can be used for data compression, which can effectively reduce the data volume and alleviate the pressure on transmission and storage. Huffman coding is a commonly used data compression method that can represent frequently occurring data with the minimum code length, thereby improving data compression efficiency. The data is finally encapsulated in JSON format, a lightweight data exchange format that facilitates data storage and transmission.

[0035] Understandably, edge preprocessing improves data quality and consistency through denoising, temporal alignment, outlier removal, and compression, ensuring that subsequent data processing can proceed on a high-quality basis. Furthermore, using JSON format for encapsulation facilitates data transmission and storage, enabling convenient data exchange and processing between different systems. This process not only optimizes data quality but also improves the overall efficiency of the system.

[0036] Step S103: Transmit the standardized data packet to the cloud server for integrity and logical verification to generate a verified dataset; After edge preprocessing, the data is transmitted to the cloud server via multiple communication protocols. Multi-protocol redundancy (such as BeiDou satellite communication and LoRa+4G hybrid networking) ensures reliable data transmission. BeiDou satellite communication serves as the primary channel, guaranteeing real-time transmission over a wide area, while the LoRa and 4G hybrid networking provides supplementary support in areas with weak or no BeiDou signal, ensuring continuity and stability of transmission.

[0037] Furthermore, the transmitted data undergoes integrity and logical verification on the cloud server. Cloud integrity verification employs SHA-256 hash value comparison technology. The SHA-256 algorithm is a widely used encryption algorithm for data integrity verification. By generating a hash value from the data content, even a tiny change will result in a change in the hash value, thus accurately detecting data tampering or corruption during transmission. Logical verification is used to determine the reasonableness and validity of the data. For example, data collected by a displacement sensor must be within a preset range. If the data exceeds this range, it may indicate a sensor malfunction or data error, requiring further investigation or rejection. Through such logical verification, the accuracy of the data transmitted to the cloud in terms of validity and reasonableness is ensured.

[0038] Step S104: Perform multimodal data fusion on the validated dataset, construct the displacement-rainfall-microseismic time series correlation matrix, extract displacement acceleration, cumulative rainfall intensity and microseismic event frequency features, and generate a fused feature vector set; By combining output data from different sensors (such as displacement, rainfall, and microseismic data), a comprehensive description of the slope's dynamic changes can be achieved. To reveal the potential relationships between the data, a displacement-rainfall-microseismic time-series correlation matrix is ​​first constructed, aligning and fusing the data from each sensor according to their time series. This time-series correlation matrix can demonstrate the correlation and interaction between the data from different sensors; for example, whether the displacement acceleration increases with increasing precipitation, or whether microseismic events occur more frequently.

[0039] Specifically, key features are extracted from the fused data, such as displacement acceleration, cumulative rainfall intensity, and the frequency of microseismic events. These features represent the critical dynamic characteristics of the slope environment and help determine slope stability. For example, a large displacement acceleration may indicate that the slope is in a phase of accelerated sliding, while a high frequency of microseismic events may indicate that the slope is under high tectonic stress.

[0040] Step S105: Input the fused feature vector set into the pre-trained risk prediction model for real-time risk prediction, and output the landslide risk index and displacement trend prediction results. In this embodiment, a pre-trained risk prediction model can be used to perform real-time analysis and prediction of the fused feature vector. LSTM (Long Short-Term Memory) is a commonly used time-series data processing model in deep learning, capable of effectively capturing long-term dependencies in time-series data. In this step, the LSTM network uses the fused feature vector as input to predict the landslide risk index and simultaneously predict future displacement trends. The LSTM network can identify the slope's movement trend based on the time-series patterns in the input data and learn the patterns of landslide occurrence based on historical data. When the landslide risk index reaches a certain preset threshold, the system will trigger an early warning, reminding relevant departments to take timely emergency measures.

[0041] Step S106: Determine the corresponding early warning level based on the landslide risk index, and trigger the automatic response strategy corresponding to the early warning level.

[0042] The system automatically determines the warning level based on the landslide risk index output by the LSTM model. The warning level is set according to the risk index value and is typically divided into multiple levels, such as green (low risk), yellow (medium risk), and red (high risk). When the landslide risk index reaches a certain preset threshold, the system triggers corresponding automated response strategies. For example, under a yellow warning level, the system may initiate an inspection task to conduct on-site checks; under a red warning level, it may trigger audible and visual alarms and activate the traffic control system.

[0043] In the above embodiments, multi-layered technical means are used to achieve accurate monitoring and early warning of unstable slopes. Through edge computing and cloud-based collaborative processing, the dynamic changes of the slope area can be monitored in real time. Based on multimodal data fusion and deep learning technology, the system can accurately assess landslide risks and predict future displacement trends in real time. Furthermore, the system also has an efficient automated response mechanism that can quickly take emergency measures to ensure that disaster prevention and control work is more efficient and timely.

[0044] Reference Figure 2 As one implementation of step S102, the step of performing edge preprocessing on the raw monitoring data to generate standardized data packets includes: Step S201: Extract UTC standard time based on BeiDou satellite signal, perform offset compensation on the local timestamp of the original monitoring data, and generate a monitoring data stream with unified time reference; Specifically, in multi-sensor systems, the timestamps of each sensor are typically recorded based on local time. This can lead to inconsistencies in the time points of data from different sensors, and even time discrepancies. Therefore, to ensure the temporal consistency and synchronization of the data, the data from all sensors must be aligned to a unified time reference.

[0045] In this embodiment, UTC time information is obtained from the BeiDou satellite and compared with the local timestamp of each sensor to calculate the time offset. Then, the data from all sensors are calibrated according to this time offset to generate a multi-sensor data matrix with a unified time reference. This matrix contains synchronized data from different sensors, ensuring that all data are compared and fused on the same timeline.

[0046] Step S202: Perform wavelet threshold denoising on the monitoring data stream; The monitoring data may contain noise components, especially in complex environments or under conditions of significant signal interference. Sensor-collected data typically contains varying degrees of noise. If this noise is not removed, it will affect the accuracy and reliability of subsequent data analysis.

[0047] In some embodiments, wavelet thresholding denoising technology can be employed. This technology is a denoising method based on wavelet transform, which can effectively remove noise without distorting the signal. Wavelet transform removes noise by decomposing the signal into multiple frequency bands and processing the coefficients of different frequency bands. Specifically, the db4 wavelet basis function is used to decompose the signal into 5 levels, and a dynamic threshold shrinking method is used to process the high-frequency coefficients. The purpose of wavelet decomposition is to decompose the signal into low-frequency and high-frequency components, and noise generally exists mainly in the high-frequency components. Therefore, by dynamically thresholding the high-frequency coefficients, most of the noise signal can be removed while retaining the effective information in the signal.

[0048] For example, GNSS displacement data is one of the key data for monitoring slope displacement. However, due to signal noise, sensor errors, and other reasons, GNSS data often contains significant noise. To extract effective displacement information from the data, a 5-level decomposition using the db4 wavelet basis function can be employed. The db4 wavelet basis function is a commonly used wavelet basis in signal processing, possessing excellent time-frequency localization characteristics and effectively capturing detailed features in the signal. By performing a 5-level wavelet decomposition on the GNSS displacement data, the data is divided into low-frequency components (representing the stationary part of the signal) and high-frequency components (representing the variable part of the signal). Noise generally exists in the high-frequency components; therefore, a dynamic threshold shrinkage method is used to process the high-frequency coefficients, retaining important signal components while suppressing noise. Dynamic threshold shrinkage calculates the threshold T = σ based on the standard deviation σ of the wavelet coefficients at each level. 2 *log(N), where N is the number of data points. High-frequency coefficients less than the threshold T will be set to zero to remove noise.

[0049] Step S203: Calculate the mean and standard deviation of the denoised monitoring data stream based on the sliding window statistical calculation, and use the 3σ principle to remove abnormal data points that exceed the preset range to generate an effective data set; The sliding window statistical method is a common time series data analysis technique. By calculating the mean and standard deviation of data within a certain time window, it can capture the trend and fluctuation range of data. In this step, the sliding window statistical method is used to calculate the mean (μ) and standard deviation (σ) of the denoised data stream, and then the 3σ principle is used to remove outlier data points. The 3σ principle is a commonly used method based on statistical theory. It assumes that the data follows a normal distribution, and data points exceeding three times the standard deviation are generally considered outliers.

[0050] For example, suppose the monitored displacement data fluctuates significantly within a certain time period. Using sliding window statistics, the mean of this data segment is calculated to be μ = 1.5 mm, and the standard deviation is σ = 0.2 mm. If three consecutive data points exceed the range [μ - 3σ, μ + 3σ] (i.e., 1.5 ± 0.6 mm), these data points will be considered outliers and removed.

[0051] Step S204: Convert the floating-point values ​​in the valid data set into fixed-point integers, compress and encode them using a static Huffman dictionary, and encapsulate them into a standardized data packet that includes a timestamp, device identifier, and CRC32 checksum.

[0052] To reduce the burden of data transmission and storage, data compression coding technology is employed. In this step, floating-point values ​​in the valid data set are first converted to fixed-point integers. This conversion helps reduce the precision requirements for data storage and significantly reduces the data volume. Fixed-point numbers occupy less storage space than floating-point numbers and also improve transmission efficiency.

[0053] Next, static Huffman coding is used to compress the data. Huffman coding is a compression method based on the frequency of data occurrence. It uses shorter codes to represent frequently occurring data and longer codes to represent sparse data, thus achieving compression. Huffman coding helps to further reduce data size and improve transmission efficiency.

[0054] Finally, this compressed data is encapsulated into standardized data packets, which include timestamps, device identifiers, and CRC32 checksums to ensure data integrity and reliability.

[0055] In the above implementation, the monitoring data collected from multiple sensors is ensured to be transmitted to the back-end system in a standardized form after precise synchronization, noise reduction, anomaly removal and efficient encoding processing. This greatly improves data quality, reduces transmission burden, and improves the system's response speed and accuracy in monitoring unstable slopes.

[0056] Reference Figure 3 As one implementation of step S104, the steps of performing multimodal data fusion on the validated dataset, constructing a displacement-rainfall-microseismic time series correlation matrix, extracting displacement acceleration, cumulative rainfall intensity, and microseismic event frequency features, and generating a fused feature vector set include: Step S301: Time alignment processing is performed on the displacement data, rainfall data and microseismic signal data in the verified dataset to obtain displacement data sequence, rainfall data sequence and microseismic signal data sequence respectively; Specifically, to ensure effective data fusion, the data are time-aligned and multiple time-series data sequences based on the same time base are generated. The displacement data sequence represents slope displacement changes, the rainfall data sequence represents precipitation changes over time, and the microseismic data sequence records the frequency and intensity of microseismic events. These data sequences will provide a foundation for subsequent feature extraction and multimodal fusion.

[0057] Step S302: Perform second-order difference processing on the displacement data sequence to calculate the instantaneous acceleration and obtain the displacement acceleration characteristics; Since displacement data itself cannot directly reflect the motion state or acceleration characteristics of a slope, the instantaneous acceleration of the displacement data can be calculated through second-order difference processing. The second-order difference method is a method for estimating acceleration by calculating the change between consecutive data points.

[0058] Specifically, for a displacement data sequence, the second-order difference formula is as follows: a t =x t+1 -2x t +x t-1 ; where a t Let x represent the instantaneous acceleration at time t. t This represents the displacement data at time t. This step effectively captures the rate of change of displacement, thereby revealing the acceleration characteristics of the slope.

[0059] Step S303: Perform sliding window cumulative calculation on the rainfall data sequence to generate cumulative rainfall intensity value; In order to better understand the impact of precipitation on slope stability, the sliding window method is used to calculate the cumulative rainfall intensity. The sliding window method can reflect the changes in precipitation intensity over a certain period of time by accumulating precipitation in each time window.

[0060] Specifically, the sliding window size can be set according to actual needs, for example, a 10-minute window, calculating the cumulative rainfall within 10 minutes every minute. This method allows for the acquisition of cumulative precipitation intensity for each time period, providing a more intuitive indicator of precipitation intensity for subsequent risk assessment.

[0061] Step S304: Encode the cumulative rainfall intensity value according to the preset classification rules to obtain the rainfall intensity encoding vector; wherein, in order to facilitate subsequent data processing and analysis, the cumulative rainfall intensity value needs to be encoded according to the preset classification rules. For example, different rainfall intensity thresholds are set to divide the cumulative rainfall intensity value into three levels: low, medium and high, each corresponding to a different encoding value, such as low as [1,0,0], medium as [0,1,0], and high as [0,0,1].

[0062] Understandably, this hierarchical coding method helps simplify the subsequent analysis process. By converting precipitation into vector form, it can be more easily combined and integrated with other data features (such as displacement acceleration, frequency of microseismic events, etc.) to form more representative data features.

[0063] Step S305: Perform event determination on the microseismic signal data sequence, count the number of microseismic pulses exceeding the set threshold within the preset time window, and normalize the number of pulses to obtain the frequency characteristics of microseismic events. Microseismic signals typically represent minor earthquakes or stress release events occurring underground and are important signals affecting slope stability. By analyzing microseismic signal data for event determination, the number of microseismic pulses (or microseismic events) within a preset time window can be counted. When the number of microseismic events exceeds a set threshold, it indicates a potential risk of instability in the slope area.

[0064] To ensure consistent processing of microseismic pulse counts, the pulse counts are normalized to allow for comparison of microseismic pulse frequencies across different time windows. Normalization typically uses the maximum and minimum values ​​from historical data, performed according to the following formula: Where, N event N represents the number of micro-vibration pulses within the current window. min and N max These represent the minimum and maximum number of pulses over the past 30 days, respectively.

[0065] Step S306: Combine the displacement acceleration features, rainfall intensity encoding vector, and microseismic event frequency features in the order of time windows to construct a time-series correlation matrix; By combining data features from different sensors (displacement acceleration, rainfall intensity, and microseismic event frequency) sequentially according to time windows, a time-series correlation matrix can be constructed. Each row in this matrix represents a data feature within a time window, while each column corresponds to a different feature (e.g., mean displacement acceleration, rainfall intensity, and microseismic event frequency). This correlation matrix reflects the temporal relationships between data from different sensors, facilitating subsequent analysis and model training.

[0066] Step S307: Perform standardization on the temporal correlation matrix and map the high-dimensional features to three-dimensional space using a nonlinear dimensionality reduction algorithm to generate a fused feature vector set.

[0067] To enable comparison of multimodal data at the same scale, the temporal correlation matrix must be standardized to ensure that the values ​​of each feature fall within the same range. Standardization can be achieved using Z-score standardization or Min-Max standardization, ensuring that each feature has a mean of 0 and a variance of 1, or mapping its values ​​to the range [0,1]. Next, a nonlinear dimensionality reduction algorithm (such as t-SNE) is used to map the high-dimensional features to a three-dimensional space. t-SNE (t-Distributed Stochastic Neighbor Embedding) is a nonlinear dimensionality reduction algorithm that preserves the local structure of high-dimensional data while mapping it to a low-dimensional space. The t-SNE algorithm is particularly suitable for situations where there are complex nonlinear relationships between data points, effectively compressing high-dimensional data into a set of three-dimensional feature vectors with visual significance.

[0068] In the above embodiments, time-series data from different sensors are effectively fused to generate an accurate and simplified fusion feature vector set, which significantly improves the quality and operability of the data and provides solid data support for subsequent landslide risk prediction and early warning.

[0069] Reference Figure 4 As one implementation of the risk prediction model, it also includes a training step for the risk prediction model, specifically including: Step S401: Based on the fused feature vector set and historical landslide event labels, construct a training dataset and a validation dataset with timestamps; This involves combining all monitoring data (such as displacement, rainfall, and microseismic data) with historical landslide event labels to construct training and validation datasets with timestamps. Each data record contains monitoring information at a specific time point and a label indicating whether a landslide occurred at that time. The introduction of timestamps ensures the temporal order of the data and maintains the chronological sequence of events.

[0070] Step S402 involves dividing the training dataset into input sequences and output label sequences according to a preset time window length. Specifically, in landslide risk prediction, the model needs to use historical data (e.g., monitoring data from the past 24 hours) to predict landslide risk over a future period (e.g., the next 6 hours). The preset time window length refers to dividing the data into fixed time periods, such as 24 hours. The input sequence for each time window is used to predict the output label sequence, which includes the landslide risk index and displacement trend for the next 6 hours.

[0071] Step S403: Standardize the input sequence to generate a standardized training tensor; The purpose of standardization is to eliminate the dimensional differences between different features, enabling the model to learn more effectively. In this step, each input feature is standardized using the Z-score standardization method, which calculates the mean and standard deviation of each feature, and then subtracts the mean from each feature value and divides it by the standard deviation to obtain the standardized feature value.

[0072] Step S404: Construct a neural network model containing an LSTM network layer, a Dropout layer, and a fully connected layer; wherein the LSTM network layer is configured to return the output of the complete time step; LSTM (Long Short-Term Memory) is a deep learning model for processing time-series data, capable of capturing dependencies over long periods. Dropout layers effectively prevent overfitting, while fully connected layers generate the final prediction output. LSTM layers are configured to return the output of a complete time step, meaning that LSTM outputs a hidden state at each time step to capture the evolution of the time series.

[0073] For example, the input to the LSTM layer is standardized training data (such as 24-hour displacement data), the network learns and outputs the state at each time step; during training, the Dropout layer randomly discards some neurons (e.g., the dropout rate is 0.2), which helps to improve the generalization ability of the model; finally, the final state output by the LSTM is mapped to the predicted values ​​of the landslide risk index and displacement trend through a fully connected layer.

[0074] As one implementation of a specific model architecture for a neural network, the LSTM network layer includes a first LSTM layer and a second LSTM layer. The activation function of the first LSTM layer is the hyperbolic tangent function, and the output dimension is consistent with the time step of the input sequence. The second LSTM layer is the global feature vector of the input sequence in the time dimension. The fully connected layer is configured with two neurons, which correspond to the probability output of the landslide risk index and the regression prediction output of the displacement trend, respectively.

[0075] Specifically, by activating the first LSTM layer with a hyperbolic tangent function and retaining the complete time step output, the local fluctuation characteristics of landslide monitoring time series data (such as displacement acceleration and rainfall intensity) are refined. The second LSTM layer compresses and generates a global feature vector to capture the long-term cumulative effect of geological evolution. Combined with the parallel output of risk probability and displacement prediction values ​​by two neurons in the fully connected layer, the model achieves multi-task collaborative optimization of landslide risk classification and early warning and displacement trend regression prediction while ensuring the model's lightweight nature. This significantly improves the real-time early warning accuracy and inference efficiency of edge computing nodes in complex geological environments, providing high-precision, low-latency intelligent decision support for geological disaster prevention and control.

[0076] Step S405: Train a neural network model based on the standardized training tensor and output label sequence, optimize the model parameters, and obtain a risk prediction model; The process involves training a neural network model using standardized input data and output label sequences. An optimization algorithm (such as the Adam optimizer) is then used to minimize the loss function, thereby optimizing the network parameters. Commonly used loss functions include mean squared error (MSE) or cross-entropy loss. During optimization, the network weights are updated using backpropagation to gradually reduce prediction error.

[0077] In some embodiments, mean squared error (MSE) can be used to measure the difference between the predicted and actual values ​​for landslide risk index prediction; mean squared error (MSE) can also be used to measure displacement trend prediction.

[0078] Step S406: Validate the risk prediction model and correct its parameters based on the validation dataset until the model's loss function meets the preset conditions, thus obtaining the trained risk prediction model.

[0079] Model validation is used to evaluate the model's performance on unseen data. Model performance is evaluated using a validation set. When the model's loss function (such as MSE) meets preset conditions (e.g., the error is less than a certain threshold), it indicates that the model training is complete and it can be deployed. If the loss function does not meet the criteria, parameters are adjusted and retraining is performed.

[0080] In the above implementation, an efficient and accurate risk prediction model is finally obtained through training. This model can accurately predict the landslide risk index and displacement trend based on real-time monitoring data, has strong time series learning ability, and can effectively avoid overfitting. It is suitable for complex landslide risk prediction scenarios.

[0081] Reference Figure 5 As a further implementation of the monitoring and early warning method, after step S105, which outputs the landslide risk index and displacement trend prediction results, the method further includes: Step S501: Based on the displacement trend prediction results, a continuous displacement heat map is generated using an interpolation algorithm; This process involves acquiring landslide displacement trend data, which may include displacement values ​​at different locations and is typically presented as discrete spatial data. Interpolation algorithms (such as inverse distance weighted interpolation) can then be used to generate continuous displacement heatmaps between these discrete data points. The core idea of ​​interpolation is to infer the values ​​of unknown points based on known discrete data points using a specific algorithm, thereby filling the space with these unknown displacement values. The weighting coefficients used in the interpolation process can control the strength of the influence between discrete data points.

[0082] For example, suppose there are 5 measurement points recording different displacement data in a landslide prediction area. Using an inverse distance weighted interpolation algorithm, the displacement data of other points can be calculated, thereby generating a displacement heatmap for the entire area.

[0083] Step S502: Synthesize a dynamic trend animation based on the time series data of the displacement trend prediction results, and overlay timestamps and risk level labels; The process involves generating a dynamic trend animation based on time-series data (e.g., hourly, daily, or other time-period data) from displacement trend prediction results. This is achieved by extracting displacement heatmap sequences at time steps (e.g., 5 minutes) and encoding them into a video format (e.g., H.264). The dynamic animation presents the displacement trend over time, facilitating monitoring of future developments.

[0084] Furthermore, when generating dynamic animations, overlaying timestamps and risk level labels allows for the real-time display of important information such as the level of risk and the direction of displacement through color changes, arrow labels, and other methods.

[0085] Step S503: Load the continuous displacement heat map and dynamic trend animation into the 3D GIS engine to build an interactive visualization interface.

[0086] The generated displacement heatmap and dynamic trend animation are loaded into a 3D GIS (Geographic Information System) engine to create an interactive visualization interface. The 3D GIS engine can display data in three-dimensional space, allowing users to observe the spatial distribution and changes of landslides from multiple perspectives (such as top view, side view, etc.).

[0087] Specifically, the visualization interface allows users to set layer rendering priorities, ensuring a clear display order for different data layers (such as displacement heatmaps, rainfall contour lines, and dynamic trend animations). Furthermore, it supports user interaction with the interface, such as viewing real-time displacement curves by clicking on device coordinates and controlling animation playback by sliding the timeline.

[0088] For example, in the interactive interface, users can click on a specific area to view the real-time displacement curve of that area, or drag the timeline to view the dynamic trend animation at different points in time.

[0089] The above implementation constructs a comprehensive and interactive landslide disaster monitoring system that not only reflects the spatial distribution of landslides but also dynamically displays their changing processes, helping decision-makers respond promptly and take necessary emergency measures. Simultaneously, the system supports user interaction with data, further enhancing operational flexibility and practicality.

[0090] This application also discloses an unstable slope monitoring and early warning system.

[0091] An unstable slope monitoring and early warning system, comprising: The acquisition module is used to collect raw monitoring data of the target slope area; the raw monitoring data includes displacement data, rainfall data, and microseismic signal data. The preprocessing module is used to perform edge preprocessing on the raw monitoring data and generate standardized data packets; The verification module is used to transmit standardized data packets to the cloud server for integrity and logical verification, and generate a verified dataset. The feature fusion module is used to perform multimodal data fusion on the validated dataset, construct the displacement-rainfall-microseismic time series correlation matrix, extract displacement acceleration, cumulative rainfall intensity and microseismic event frequency features, and generate a fused feature vector set; The risk prediction module is used to input the fused feature vector set into the pre-trained risk prediction model for real-time risk prediction and output the landslide risk index and displacement trend prediction results. The early warning module is used to determine the corresponding early warning level based on the landslide risk index and trigger the automatic response strategy corresponding to the early warning level.

[0092] As a further implementation of the monitoring and early warning system, it also includes: The heat map generation module is used to generate continuous displacement heat maps based on displacement trend prediction results using an interpolation algorithm. The animation synthesis module is used to synthesize dynamic trend animations based on time-series data of displacement trend prediction results, and to overlay timestamps and risk level labels. The visualization interface building module is used to load continuous displacement heatmaps and dynamic trend animations into the 3D GIS engine to build an interactive visualization interface.

[0093] An unstable slope monitoring and early warning system according to an embodiment of this application can implement any of the above-mentioned monitoring and early warning methods, and the specific working process of each module in the monitoring and early warning system can refer to the corresponding process in the above-mentioned method embodiments.

[0094] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0095] This application also discloses a computer device.

[0096] Computer equipment includes 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 monitoring and early warning methods described above.

[0097] This application also discloses a computer-readable storage medium.

[0098] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as any of the monitoring and early warning methods described above.

[0099] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0100] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0101] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for monitoring and early warning of unstable slopes, characterized in that, The monitoring and early warning methods include: Collect raw monitoring data of the target slope area; the raw monitoring data includes displacement data, rainfall data, and microseismic signal data; The raw monitoring data is preprocessed to generate standardized data packets; The standardized data packets are transmitted to the cloud server for integrity and logical verification to generate a verified dataset. Multimodal data fusion was performed on the verified dataset to construct a displacement-rainfall-microseismic time series correlation matrix, and features of displacement acceleration, cumulative rainfall intensity and microseismic event frequency were extracted to generate a fused feature vector set. The fused feature vector set is input into a pre-trained risk prediction model for real-time risk prediction, and the landslide risk index and displacement trend prediction results are output. The corresponding early warning level is determined based on the landslide risk index, and the corresponding automated response strategy is triggered based on the early warning level.

2. The method for monitoring and early warning of unstable slopes according to claim 1, characterized in that, The steps for performing edge preprocessing on the raw monitoring data to generate standardized data packets include: Based on the extraction of UTC standard time from BeiDou satellite signals, the local timestamps of the original monitoring data are offset and compensated to generate a monitoring data stream with a unified time reference. Wavelet threshold denoising is performed on the monitoring data stream; The mean and standard deviation of the denoised monitoring data stream are calculated based on the sliding window statistical calculation, and abnormal data points exceeding the preset range are removed using the 3σ principle to generate an effective data set. The floating-point values ​​in the valid data set are converted into fixed-point integers, compressed and encoded using a static Huffman dictionary, and encapsulated into a standardized data packet including a timestamp, device identifier, and CRC32 checksum.

3. The method for monitoring and early warning of unstable slopes according to claim 1, characterized in that, The steps of performing multimodal data fusion on the verified dataset, constructing a displacement-rainfall-microseismic time series correlation matrix, extracting displacement acceleration, cumulative rainfall intensity, and microseismic event frequency features, and generating a fused feature vector set include: The displacement data, rainfall data, and microseismic signal data in the verified dataset are time-aligned to obtain displacement data sequences, rainfall data sequences, and microseismic signal data sequences, respectively. The displacement data sequence is subjected to second-order difference processing to calculate instantaneous acceleration, thereby obtaining displacement acceleration characteristics; A sliding window cumulative calculation is performed on the rainfall data sequence to generate a cumulative rainfall intensity value; The cumulative rainfall intensity value is encoded according to a preset grading rule to obtain a rainfall intensity encoding vector; Event determination is performed on the microseismic signal data sequence, the number of microseismic pulses exceeding a set threshold within a preset time window is counted, and the number of pulses is normalized to obtain the frequency characteristics of microseismic events. The displacement acceleration features, rainfall intensity encoding vector, and microseismic event frequency features are combined in order of time window to construct a time-series correlation matrix; The time-series correlation matrix is ​​standardized, and the high-dimensional features are mapped to a three-dimensional space using a nonlinear dimensionality reduction algorithm to generate a fused feature vector set.

4. The method for monitoring and early warning of unstable slopes according to claim 1, characterized in that, It also includes a training step for the risk prediction model, the training step comprising: Based on the fused feature vector set and historical landslide event labels, a training dataset and a validation dataset with timestamps are constructed. The training dataset is divided into an input sequence and an output label sequence according to a preset time window length; The input sequence is standardized to generate a standardized training tensor; Construct a neural network model comprising an LSTM network layer, a Dropout layer, and a fully connected layer; wherein the LSTM network layer is configured to return the output of a complete time step; The neural network model is trained based on the standardized training tensor and the output label sequence, and the model parameters are optimized to obtain the risk prediction model. The risk prediction model is validated and its parameters are corrected based on the validation dataset until the model's loss function meets the preset conditions, thus obtaining the trained risk prediction model.

5. The method for monitoring and early warning of unstable slopes according to claim 4, characterized in that, The LSTM network layer includes a first LSTM layer and a second LSTM layer. The activation function of the first LSTM layer is the hyperbolic tangent function, and the output dimension is consistent with the time step of the input sequence. The output of the second LSTM layer is the global feature vector of the input sequence in the time dimension. The fully connected layer is configured with two neurons, which correspond to the probability output of the landslide risk index and the regression prediction output of the displacement trend, respectively.

6. A method for monitoring and early warning of unstable slopes according to any one of claims 1 to 5, characterized in that, Following the steps of outputting the landslide risk index and displacement trend prediction results, the following is also included: Based on the displacement trend prediction results, a continuous displacement heat map is generated using an interpolation algorithm; A dynamic trend animation is synthesized based on the time series data of the displacement trend prediction results, and timestamps and risk level labels are superimposed. The continuous displacement heatmap and dynamic trend animation are loaded into a 3D GIS engine to build an interactive visualization interface.

7. An unstable slope monitoring and early warning system, characterized in that, The monitoring and early warning system includes: The acquisition module is used to collect raw monitoring data of the target slope area; the raw monitoring data includes displacement data, rainfall data, and microseismic signal data. The preprocessing module is used to perform edge preprocessing on the raw monitoring data to generate standardized data packets; The verification module is used to transmit the standardized data packet to the cloud server for integrity verification and logical verification, and generate a verified dataset. The feature fusion module is used to perform multimodal data fusion on the verified dataset, construct a displacement-rainfall-microseismic time series correlation matrix, extract displacement acceleration, cumulative rainfall intensity and microseismic event frequency features, and generate a fused feature vector set. The risk prediction module is used to input the fused feature vector set into a pre-trained risk prediction model for real-time risk prediction and output the landslide risk index and displacement trend prediction results. The early warning module is used to determine the corresponding early warning level based on the landslide risk index and trigger the automatic response strategy corresponding to the early warning level.

8. The unstable slope monitoring and early warning system according to claim 7, characterized in that, The monitoring and early warning system also includes: The heat map generation module is used to generate a continuous displacement heat map based on the displacement trend prediction results using an interpolation algorithm. Animation synthesis module is used to synthesize dynamic trend animation based on the time series data of the displacement trend prediction results, and to overlay timestamps and risk level labels; The visualization interface building module is used to load the continuous displacement heat map and dynamic trend animation into the 3D GIS engine to build an interactive visualization interface.

9. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images