Slope displacement prediction method, device and equipment and storage medium

Abstract

The invention discloses a slope displacement prediction method, device and equipment and a storage medium, and the method comprises the steps: collecting displacement data and multi-source environment factor data of a to-be-monitored slope region, and carrying out the preprocessing of the displacement data and the multi-source environment factor data; performing time-frequency decoupling on displacement data by adopting a variational mode decomposition method to obtain a plurality of mode components, and merging the mode components with the multi-source environment factor data to obtain an input matrix; introducing a supervision loss function into the constructed initial prediction neural network model, and training based on an error feedback mechanism and the input matrix to obtain a time sequence prediction neural network model; and inputting monitoring data of a slope area to be monitored into the time sequence prediction neural network model, and generating a complete future displacement prediction sequence through inverse mode reconstruction. According to the method, the problems in the prior art are solved from data acquisition and processing, feature analysis, model training and prediction output, and the accuracy and reliability of slope displacement prediction are improved.

Description

Technical Field

[0001] This application relates to the field of displacement prediction technology, and more specifically, to a slope displacement prediction method, apparatus, equipment, and storage medium. Background Technology

[0002] In recent years, with the construction of mountain transportation infrastructure, resource extraction, and the increasing frequency of extreme weather events, the incidence of slope landslides has increased significantly, posing a huge threat to engineering safety and the lives and property of people. Environmental factors such as rainfall, pore water pressure, soil temperature and humidity, and changes in ground stress are widely considered to be the main triggering factors for slope deformation and even instability. Displacement monitoring methods based on GNSS (Global Navigation Satellite System) technology are commonly used in academia and engineering practice. More and more research is attempting to integrate multi-source environmental factors to predict GNSS displacement data in order to comprehensively reflect the dynamic response process of the slope system, thereby improving the ability to identify and warn of potential instability trends.

[0003] The following types of slope deformation prediction methods have been gradually developed in the existing technology: (1) Empirical regression model: based on historical data to fit the relationship between displacement and environmental factors, such as linear regression, exponential regression, etc.; (2) Statistical learning model: including ARIMA, support vector regression (SVR), random forest, etc., which make predictions by mining the correlation between data; (3) Deep learning model: such as LSTM, GRU, Transformer and other neural networks, which improve nonlinear modeling ability by learning time series features; (4) Physical model method: based on elastoplastic theory, stability analysis or hysteresis characteristics to build mathematical models; (5) Data fusion method: extract features through principal component analysis, wavelet decomposition, empirical mode decomposition, etc., and combine them with deep models.

[0004] GNSS technology acquires slope displacement data that contains multiple dynamic characteristics, including long-term trends, periodic disturbances, and short-term anomalies. It is susceptible to interference from factors such as multipath effects, atmospheric delay, and hardware errors, exhibiting high-frequency noise and low-frequency drift. This results in significant non-stationarity of the original sequence. Most depth models are sensitive to noise and lack effective data cleaning mechanisms, often leading to distorted or overfitted predictions, making it difficult to capture the dynamic driving effect of environmental factors on the slope displacement process. Summary of the Invention

[0005] To address at least one deficiency or improvement need in the existing technology, this invention provides a slope displacement prediction method, apparatus, device, and storage medium. This addresses the problem that slope displacement data acquired using GNSS technology contains various dynamic characteristics, is easily affected by multipath effects, atmospheric delay, hardware errors, and other factors, resulting in significant non-stationarity of the original sequence, a lack of effective data cleaning mechanisms, and difficulty in capturing the dynamic driving effect of environmental factors on the slope displacement process, leading to distorted or overfitted prediction results.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a slope displacement prediction method is provided, comprising: Displacement data and multi-source environmental factor data of the slope area to be monitored are collected, and the displacement data and multi-source environmental factor data are preprocessed. The variational mode decomposition method is used to decouple the displacement data from time to frequency to obtain multiple modal components, which are then merged with multi-source environmental factor data to obtain the input matrix; A physical supervision loss function is introduced into the constructed initial prediction neural network model, and a time-series prediction neural network model is obtained by training based on the error feedback mechanism and the input matrix. The monitoring data of the slope area to be monitored is input into the time-series prediction neural network model, and a complete future displacement prediction sequence is generated through inverse mode reconstruction.

[0007] In one possible implementation, the process includes collecting displacement data and multi-source environmental factor data of the slope area to be monitored, and preprocessing the displacement data and multi-source environmental factor data, further comprising: GNSS receivers and multi-source sensors are deployed in the slope area to be monitored to collect displacement data and multi-source environmental factor data of the slope area to be monitored. Multi-source environmental factor data are synchronized with timestamps and standardized in data format to construct multi-source monitoring data with a unified time axis. Preprocessing is performed on displacement data and multi-source monitoring data, including missing value handling, mutation detection, and data filtering.

[0008] In one possible implementation, a variational mode decomposition method is used to decouple the displacement data from time to frequency to obtain multiple modal components, which are then merged with multi-source environmental factor data to obtain an input matrix. This also includes: The variational mode decomposition parameters are set, and the objective function of the variational mode decomposition is solved iteratively by the alternating direction multiplier method. The displacement data is decoupled in time and frequency to obtain multiple modal components. Calculate the modal frequencies and variance contribution ratios, and classify the modal components based on the modal frequencies and variance contribution ratios; Modal components are merged with the synchronous sequence of multi-source environmental factor data to form a standardized input matrix.

[0009] In one possible implementation, variational mode decomposition parameters are set, and the objective function of the variational mode decomposition is solved iteratively using the alternating direction multiplier method. Multiple modal components are obtained by time-frequency decoupling of the displacement data. The implementation also includes: Based on the time-frequency characteristics of the displacement data, the number of decomposition modes, the penalty factor, and the convergence tolerance threshold of the variational mode decomposition are set. The Lagrange multiplier method is used to construct the optimization objective function of variational mode decomposition, and the solution is obtained iteratively by alternating direction multiplier method until the preset convergence condition is met. The modal components at convergence are reorganized based on the actual physical meaning and frequency structure of the modes.

[0010] In one possible implementation, a supervised loss function is introduced into the constructed initial predictive neural network model, and a time-series predictive neural network model is obtained by training based on an error feedback mechanism and an input matrix. This also includes: The velocity and acceleration information of displacement are extracted from the modal components, and the physical supervision loss function of the displacement evolution model and the main loss function of the initial prediction neural network model are constructed. The physical supervision loss function and the main loss function are weighted and fused to form a joint optimization objective function; An error feedback mechanism is introduced, and the initial prediction neural network model is iteratively trained using the input matrix and the joint optimization objective function to obtain the time-series prediction neural network model.

[0011] In one possible implementation, an error feedback mechanism is introduced, and a time-series prediction neural network model is obtained by iteratively training the initial prediction neural network model using the input matrix and a joint optimization objective function. This also includes: The predicted displacement dynamic curve is generated based on the prediction results of each round of training of the neural network model; The weights of the joint optimization objective function are adjusted based on the similarity between the prediction results and the predicted displacement dynamic curves in each round of training. When the maximum number of iterations or the preset prediction accuracy is reached, the iterative training stops and the temporal prediction neural network model is obtained.

[0012] In one possible implementation, monitoring data of the slope area to be monitored is input into a time-series predictive neural network model, and a complete future displacement prediction sequence is generated through inverse mode reconstruction. The method also includes: The monitoring data of the slope area to be monitored is input into the time-series predictive neural network model to predict future slope displacement. The predicted values ​​of each modal component are superimposed point by point using the inverse modal reconstruction method to generate a complete future displacement sequence.

[0013] According to a second aspect of the present invention, a slope displacement prediction device is also provided, comprising: The data acquisition and processing module is configured to acquire displacement data and multi-source environmental factor data of the slope area to be monitored, and to preprocess the displacement data and multi-source environmental factor data. The mode decomposition module is configured to use the variational mode decomposition method to decouple the displacement data from time and frequency to obtain multiple modal components, and then merge them with multi-source environmental factor data to obtain an input matrix. The model training module is configured to introduce a physical supervision loss function into the constructed initial predictive neural network model and train it based on the error feedback mechanism and the input matrix to obtain a temporal predictive neural network model. The displacement prediction module is configured to input monitoring data of the slope area to be monitored into a time-series prediction neural network model and generate a complete future displacement prediction sequence through inverse mode reconstruction.

[0014] According to a third aspect of the present invention, a slope displacement prediction device is also provided, comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of any of the above-described slope displacement prediction methods.

[0015] According to a fourth aspect of the present invention, a storage medium is also provided, which stores a computer program executable by a slope displacement prediction device, which, when run on the slope displacement prediction device, causes the slope displacement prediction device to perform the steps of any of the above-described slope displacement prediction methods.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: This invention provides a slope displacement prediction method that simultaneously collects displacement data and multi-source environmental factor data of the slope area to be monitored. Preprocessing the collected data removes noise and outliers, improving data quality and reliability, and addressing the lack of effective data cleaning mechanisms for raw sequences in existing technologies. By using variational mode decomposition to decouple the displacement data from time to frequency, different frequency components such as long-term trends, periodic disturbances, and noise can be accurately separated. Compared to traditional methods, this improves the processing effect of non-stationary signals and avoids slope displacement prediction distortion caused by high-frequency noise and low-frequency drift. Combining the modal components of the decoupled multi-source environmental factor data with the displacement data constructs a standardized multivariate input matrix, establishing a causal relationship between the two over time, enabling more accurate capture of the dynamic driving effect of environmental changes on displacement. When training the prediction neural network model, a physical supervision loss function based on a displacement evolution model is introduced for constraint. The strength of the physical constraint is dynamically adjusted through an error feedback mechanism to prevent prediction bias during landslide abrupt changes, improving the stability of long-term predictions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic flowchart of an embodiment of the slope displacement prediction method provided by the present invention; Figure 2 Provided by the present invention Figure 1 A schematic flowchart of an embodiment of step S101; Figure 3 Provided by the present invention Figure 1 A schematic flowchart of an embodiment of step S102; Figure 4 Provided by the present invention Figure 3 A schematic flowchart of an embodiment of step S301; Figure 5 Provided by the present invention Figure 1 A schematic flowchart of an embodiment of step S103; Figure 6 Provided by the present invention Figure 5 A schematic flowchart of an embodiment of step S503; Figure 7 A schematic diagram of a structure of an embodiment of the slope displacement prediction device provided by the present invention; Figure 8This is a schematic diagram of the slope displacement prediction device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0020] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] This invention provides a method, apparatus, device, and storage medium for predicting slope displacement, which will be described below.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the slope displacement prediction method provided by the present invention. In a specific embodiment of the present invention, a slope displacement prediction method is disclosed, comprising: S101. Collect displacement data and multi-source environmental factor data of the slope area to be monitored, and preprocess the displacement data and multi-source environmental factor data. S102. The time-frequency decoupling of the displacement data is performed by the variational mode decomposition method to obtain multiple modal components, which are then merged with the multi-source environmental factor data to obtain the input matrix. S103. Introduce a physical supervision loss function into the constructed initial prediction neural network model, and train it based on the error feedback mechanism and the input matrix to obtain a time-series prediction neural network model. S104. Input the monitoring data of the slope area to be monitored into the time-series prediction neural network model, and generate a complete future displacement prediction sequence through inverse mode reconstruction.

[0023] In the above embodiments, by deploying various specialized sensors, information affecting slope displacement can be acquired from different angles. Displacement data can accurately reflect the spatial changes of the slope, while multi-source environmental factor data, such as rainfall, pore water pressure, surface temperature and humidity, soil moisture content, and ground stress, are all important factors affecting slope stability. Collecting this data enables the subsequently trained model to more accurately capture the relationship between slope displacement and environmental factors, thereby improving the accuracy of predictions.

[0024] Preprocessing can include, but is not limited to, time alignment, missing value handling, mutation detection, and data filtering. A unified time standard and time alignment algorithm ensure consistency across various observational information over time, avoiding data analysis errors caused by time asynchrony. Missing value handling methods automatically select appropriate imputation methods based on the missing value, reducing the impact of missing data on model training. Mutation detection strategies can promptly identify outliers in the data, preventing abnormal data from interfering with model training and prediction results. Data filtering methods can remove high-frequency noise from the data, improving data quality and enabling the model to learn more realistic displacement change patterns.

[0025] Compared to traditional Empirical Mode Decomposition (EMD) methods, Variational Mode Decomposition (VMD) offers advantages in mathematical definition, reconstruction accuracy, and convergence. It is suitable for non-stationary and variable GNSS slope monitoring data, decomposing complex raw signals into physically meaningful components, thus improving the identifiability and structural independence of various geological deformation features in the time and frequency domains. By decomposing modes into different frequency components, the dynamic characteristics of displacement data, such as long-term trends, periodic disturbances, and short-term anomalies, can be clearly analyzed, providing more targeted input for subsequent modeling.

[0026] By merging modal components with environmental factor data, the internal variation characteristics and external influencing factors of displacement are comprehensively considered. Environmental factors such as rainfall and pore water pressure are significantly correlated with slope deformation. Using them together with the displacement modal components as an input matrix enables subsequent models to comprehensively understand the driving factors of slope displacement, thereby improving the model's predictive ability and interpretability for slope displacement.

[0027] As a preferred embodiment, the initial predictive neural network model is a Transformer model. The Transformer model's positional encoding, attention mechanism, and residual connections enable it to effectively handle long sequences and multivariate dynamic relationships. In slope displacement prediction, displacement data and environmental factor data are typically long sequences and interconnected. The Transformer model can capture the complex relationships between these data, improving the model's ability to model time series data.

[0028] A Velocity-Offset-Hysteresis (VOH) displacement evolution model is introduced as a weak physical supervision mechanism. This transforms the hysteretic characteristics of velocity and acceleration during slope deformation into quantifiable physical prior constraints, which are then embedded into the loss function of the neural network. This mechanism effectively guides the neural network to output prediction trajectories that better align with the logic of geophysical evolution. While maintaining numerical accuracy, it enhances the model's ability to perceive abrupt changes in trends and improves the physical reliability of the output results, solving the problems of poor interpretability and weak generalization ability in traditional pure data-driven models. Simultaneously, the system introduces an error feedback mechanism, dynamically adjusting the weights of the physical supervision terms based on the similarity between the model's predictions and the VOH structure. When the model predictions deviate from physical laws, the physical supervision intensity is automatically increased, causing the model to revert to a prediction direction consistent with physical logic. When the model predictions approach the VOH structure, the physical supervision intensity is appropriately reduced to strengthen the data-driven capability. This effectively avoids the two extreme phenomena of "physical overfitting" and "physical failure," improving the model's predictive stability during periods of rapid change, such as the initial landslide stage and acceleration phase. At the data input end, the system extracts modal component sequences by looking back a fixed-length time window from the current moment. This allows the model to use the latest data for prediction, improving the timeliness and accuracy of forecasts. A unified matrix data structure and standardized processing ensure the consistency and standardization of the input data, facilitating model processing and analysis. The multiple modal components output by the model are recombined through an inverse modal reconstruction process to form a complete future displacement prediction sequence. This process not only preserves the structural characteristics of each frequency component but can also be used to analyze the overall trend of slope displacement and the interaction between responses at different frequencies, providing richer explanatory evidence for disaster early warning and mechanism analysis.

[0029] Compared with existing technologies, this embodiment provides a slope displacement prediction method that simultaneously collects displacement data and multi-source environmental factor data of the slope area to be monitored. Preprocessing the collected data removes noise and outliers, improving data quality and reliability, and addressing the lack of effective data cleaning mechanisms for raw sequences in existing technologies. By using variational mode decomposition to decouple the displacement data from time to frequency, different frequency components such as long-term trends, periodic disturbances, and noise can be accurately separated. Compared with traditional methods, this improves the processing effect of non-stationary signals and avoids slope displacement prediction distortion caused by high-frequency noise and low-frequency drift. Combining the modal components of the decoupled multi-source environmental factor data with the displacement data to construct a standardized multivariate input matrix establishes a causal relationship between the two over time, enabling more accurate capture of the dynamic driving effect of environmental changes on displacement. When training the prediction neural network model, a physical supervision loss function based on a displacement evolution model is introduced for constraint. The strength of the physical constraint is dynamically adjusted through an error feedback mechanism to prevent prediction bias during landslide abrupt changes, improving the stability of long-term predictions.

[0030] Please see Figure 2 , Figure 2 Provided by the present invention Figure 1 A flowchart illustrating one embodiment of step S101. In some embodiments of the present invention, the process of collecting displacement data and multi-source environmental factor data of the slope area to be monitored, and preprocessing the displacement data and multi-source environmental factor data, further includes: S201. Deploy GNSS receivers, GNSS antennas, and multi-source sensors in the slope area to be monitored to collect displacement data and multi-source environmental factor data of the slope area to be monitored. S202. Multi-source environmental factor data are processed by time stamp synchronization and data format standardization to construct multi-source monitoring data with a unified time axis. S203. Perform preprocessing on displacement data and multi-source monitoring data, including missing value handling, mutation detection, and data filtering.

[0031] In the above embodiments, for displacement data, a Beidou GNSS receiver is used, which supports BDS B1 / B2 / B3 tri-frequency, and can perform high-precision three-dimensional coordinate observation. The equipment supports static measurement and continuous RTK positioning, and the sampling interval is adjustable from 1 to 30 seconds. It is connected to the near-field reference station through differential method, and the CORS system is used to realize centimeter-level three-dimensional displacement calculation, and output X, Y, and H displacement sequences with timestamps.

[0032] Simultaneously, to facilitate simultaneous observation of various factors, multiple environmental sensors are deployed near GNSS stations. For example, rainfall sensors utilize rain gauges with a tipping bucket mechanical structure to accurately collect minute-level rainfall intensity data, which is then connected to an edge gateway via an RS485 interface; pore water pressure sensors employ vibrating wire pore pressure gauges, buried at different depths within the slope, acquiring changes in pore pressure through frequency conversion sampling; surface temperature and humidity sensors use a combined sensor integrating temperature, relative humidity, wind speed, and wind direction, supporting LoRa wireless transmission; soil moisture monitoring employs sensors based on the time-domain reflectometry (TDR) principle, inserted into the soil at different depths within the slope to obtain time-series data on soil moisture content; and stress sensors deploy stress gauges, installed at typical fault planes or potential slip zones, acquiring stress data through a high-precision analog signal converter. All sensor signals are accessed through an edge computing gateway and synchronized in real-time to the slope monitoring data platform.

[0033] The raw data is timestamped according to the Uniform Time Standard (UTC). The platform's built-in time alignment algorithm achieves data format standardization and time sequence synchronization, constructing a multi-source monitoring database based on a "unified time axis". After data alignment, missing values ​​are handled. The system has two built-in methods for filling missing values: if the missing time period is less than 3 times the sampling period, local linear interpolation or nearest neighbor interpolation is used; if it exceeds 3 times the period, historical average values ​​are used for backfilling. Users can manually adjust the default threshold.

[0034] For mutation detection, a joint identification strategy based on the sliding standard deviation threshold method and box plot method is introduced. When a sudden change occurs in the monitored sequence within a short period of time, the system detects abrupt changes in the slope and abnormal amplitude points of the data sequence, achieving automatic identification and removal of abnormal points, and notifying manual confirmation when necessary. For data filtering, multiple filtering strategies are built-in. The recommended method is a combination of wavelet threshold denoising and Kalman filtering. For GNSS displacement sequences, standard Kalman filtering is used to smooth the trend term. The initial value of the observation noise covariance matrix of the Kalman filter model can be estimated based on the standard deviation of historical samples. The state transition matrix assumes that the displacement is a linear uniformly accelerated process. The processed data is stored in CSV and database formats, managed by an index of "observation point number-sensor type-time series" for subsequent modal decomposition and modeling training.

[0035] Please see Figure 3 , Figure 3 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S102. In some embodiments of the present invention, variational mode decomposition is used to decouple displacement data in time and frequency to obtain multiple modal components, which are then merged with multi-source environmental factor data to obtain an input matrix. The method further includes: S301. Set the variational mode decomposition parameters, iteratively solve the objective function of variational mode decomposition using the alternating direction multiplier method, and decouple the displacement data from time to frequency to obtain multiple modal components. S302. Calculate the modal frequency and variance contribution ratio, and classify the modal components based on the modal frequency and variance contribution ratio; S303. Merge the modal components with the synchronization sequence of multi-source environmental factor data to form a standardized input matrix.

[0036] In the above embodiments, the optimal number of decomposition modes K is determined based on the time-frequency characteristics of the GNSS displacement signal. By analyzing the spectral distribution characteristics of the signal, K is typically set to 3-5 modal components to fully capture different frequency components such as long-term trends, periodic disturbances, and noise. A penalty factor α (usually set to 2000) is defined to control the bandwidth constraint of the modes, achieving a balance between signal reconstruction accuracy and computational complexity. A convergence tolerance threshold ε (e.g., 10^-7) is set to determine whether the iterative process has converged, ensuring that the termination condition of the algorithm is clear.

[0037] The objective function of variational mode decomposition is: ; Among them, convolution kernel This represents the Hilbert transform, used to extract the analytical form of modal signals. These are the center frequencies of each mode. For modal functions, The displacement-time series is input into the VMD module for iterative calculation, and the modal function is updated in each iteration. and center frequency Simultaneously, the Lagrange multipliers are adjusted until convergence. The algorithm terminates when the change in the mode function is less than a set threshold ϵ for two consecutive iterations.

[0038] The system classifies modes based on modal frequency and variance contribution ratio, and calculates the dominant frequency, power spectral density, and energy ratio for each mode to facilitate the selection of appropriate input variables in subsequent modeling stages. Mode discrimination can be based on the following criteria: Modal energy ratio: ; Modal dominant frequency f: The peak frequency of each modality is calculated by FFT.

[0039] For modes with strong physical interpretability, their temporal sequence is retained as modeling input; while for high-frequency noise modes, they can be directly discarded or used as noise features for neural network reference. In addition, to enhance the model's ability to perceive trend changes, the system can also calculate the first derivative (velocity) and second derivative (acceleration) of the mode to form structural dynamic features.

[0040] In terms of multidimensional data fusion, the system merges the modal components of GNSS in three directions (X, Y, H) with the synchronization sequences of environmental factors to form a standardized modeling input matrix. For example, modes 1-3 serve as the trend and disturbance signals in each direction, and the merged input matrix is ​​as follows:

[0041] in , , , These represent standardized time series of rainfall, pore pressure, temperature, and moisture content, respectively.

[0042] This method employs variational mode decomposition to structurally decouple non-stationary GNSS displacement signals, effectively separating physical trends, disturbance causes, and system noise in the time and frequency domains. This provides a clear, controllable, and interpretable input channel for introducing a VOH (Vehicle Overriding) physical supervision structure and neural network modeling. Compared to traditional low-pass filtering or direct modeling, this method not only improves the relevance and accuracy of data modeling but also enhances the model's sensitivity to critical point abrupt changes or multi-scale response behavior.

[0043] Please see Figure 4 , Figure 4 Provided by the present invention Figure 3 A flowchart illustrating an embodiment of step S301 is provided. In some embodiments of the present invention, variational mode decomposition parameters are set, and the objective function of variational mode decomposition is solved iteratively using the alternating direction multiplier method. Multiple modal components are obtained by time-frequency decoupling of the displacement data. The method further includes: S401. Based on the time-frequency characteristics of the displacement data, set the number of decomposition modes, penalty factor, and convergence tolerance threshold for variational mode decomposition. S402. Construct the optimization objective function of variational mode decomposition using the Lagrange multiplier method, and solve it iteratively using the alternating direction multiplier method until the preset convergence condition is met. S403. Based on the actual physical meaning and frequency structure of the modes, the modal components at convergence are recombined.

[0044] In the above embodiments, initial parameters for VMD are set, including the number of decomposition modes K, the penalty factor α, and the tolerance threshold ϵ. For GNSS displacement signals, a value of 3 to 5 is recommended for K, which is sufficient to distinguish between long-term trends, medium-term disturbances, and high-frequency noise. In practical applications, it is recommended to set K=4, α=2000, and ϵ=10⁻⁷ to ensure decomposition quality and computational efficiency.

[0045] The iterative steps of the Alternating Direction Multiplier Method (ADMM) are as follows: Update mode Solve the system of linear equations using the Fourier domain.

[0046] Update center frequency Take the weighted average of the instantaneous frequencies of the modes.

[0047] Update the Lagrange multiplier λ: by Adjustment, where γ is the penalty parameter.

[0048] After VMD is completed, K modal components are output. , ,…, Each mode has a defined frequency structure. Mode recombination is performed based on the actual physical meaning: Modal Generally, these are low-frequency components, corresponding to the long-term deformation trend of the slope, such as slow deep displacement and release of ground stress. Modal Mid-frequency components are often associated with external disturbances such as periodic rainfall and surface loading. Modal High-frequency components are often system noise or sensor errors and can be filtered out later.

[0049] Please see Figure 5 , Figure 5 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S103 is provided. In some embodiments of the present invention, a supervised loss function is introduced into the constructed initial prediction neural network model, and a time-series prediction neural network model is obtained by training based on an error feedback mechanism and an input matrix. The method further includes: S501. Extract the velocity and acceleration information of the displacement from the modal components, and construct the physical supervision loss function of the displacement evolution model and the main loss function of the initial prediction neural network model; S502. The physical supervision loss function and the main loss function are weighted and fused to form a joint optimization objective function; S503. An error feedback mechanism is introduced, and the initial prediction neural network model is iteratively trained through the input matrix and the joint optimization objective function to obtain the time-series prediction neural network model.

[0050] In the above embodiments, before constructing the VOH supervision signal, velocity and acceleration information must first be extracted from the decomposed displacement mode sequence. Based on the trend principal mode (usually mode 1) after variational mode decomposition (VMD), the first and second derivatives in continuous time are constructed to obtain the velocity and acceleration, respectively. The VOH model does not model displacement in the form of explicit differential equations, but extracts the hysteresis structure through statistical analysis, that is, identifies the phase path of displacement velocity vt and acceleration at, and constructs a mathematical expression of its "potential law of displacement evolution". The system constructs a VOH supervision term to quantify the degree of deviation between the neural network output and the VOH structure. The VOH loss function term is defined as follows:

[0051] in: , Velocity and acceleration calculated from model predictions; The acceleration prior is derived from VOH; α and β are adjustment weights, empirically set between 0.3 and 0.7.

[0052] An improved Transformer is used as the initial predictive neural network model for time series modeling. The Transformer is suitable for handling long sequences and multivariate dynamic relationships, achieving information association across time steps through positional encoding, attention mechanisms, and residual connections. The network model structure includes: Input embedding layer: Maps the input feature matrix to a vector of uniform dimension and superimposes sine and cosine positional codes to enable the network to perceive temporal order; Multi-head self-attention mechanism: Extracts implicit dependencies between time points through multiple sets of parallel attention heads; Feedforward neural network layer: Uses two fully connected layers to enhance nonlinear expressive power; Predictive output layer: Maps the encoded result to a displacement prediction sequence for the next H steps.

[0053] The main loss function uses mean squared error (MSE): Then, the VOH loss term is weighted and fused with the neural network's main loss function (such as mean squared error MSE) to form a joint optimization objective function: , where λ controls the intensity of physical supervision.

[0054] To further improve the stability of the model in long-term sequence prediction, this invention introduces an error feedback mechanism after each training round, namely: 1. The value corresponding to the previous round of predictions , Similarity analysis was performed with the VOH curve; 2. If the deviation from the VOH structure exceeds a preset threshold continuously, the physical supervision term will be increased. The weights; 3. If the model prediction is already very close to the VOH structure, gradually reduce the supervision intensity to enhance the data learning ability.

[0055] This mechanism effectively prevents the model from falling into either "physical overfitting" or "physical failure," and helps maintain stable predictive ability during periods of drastic change, such as the initial stage and acceleration phase of a landslide.

[0056] Please see Figure 6 , Figure 6 Provided by the present invention Figure 5 A flowchart illustrating an embodiment of step S503 is shown. In some embodiments of the present invention, an error feedback mechanism is introduced, and a temporal prediction neural network model is obtained by iteratively training the initial prediction neural network model through an input matrix and a joint optimization objective function. The method further includes: S601. Generate a predicted displacement dynamic curve based on the prediction results of each round of training of the neural network model. S602. Adjust the weights of the joint optimization objective function based on the similarity between the prediction results and the predicted displacement dynamic curves of each round of training. S603. When the maximum number of iterations or the preset prediction accuracy is reached, stop the iterative training to obtain the time-series prediction neural network model.

[0057] In the above embodiment, after each round of training, the predicted output of the current model based on the input matrix (containing GNSS modal sequences, their derivative features, and environmental factors such as rainfall, pore pressure, temperature, and humidity) is processed. A dynamic curve of predicted displacement is plotted with the time series as the horizontal axis and the predicted displacement value as the vertical axis. For example, for the prediction results for the next few hours to days, the predicted displacement points at each time point are connected sequentially in chronological order to form a continuous curve.

[0058] To more accurately depict the trend of predicted displacement changes, smoothing techniques, such as the moving average method, can be used to smooth the dynamic curve of the predicted displacement. The moving average method calculates the average value of the predicted displacement within a certain time window, eliminating short-term fluctuations and making the curve smoother, thus better reflecting the long-term displacement change trend.

[0059] Commonly used similarity metrics include mean squared error (MSE) and correlation coefficient. MSE measures the average squared difference between the predicted displacement and the actual curve. The correlation coefficient measures the degree of linear correlation between two curves, ranging from -1 to 1. A value closer to 1 indicates a stronger positive correlation, a value closer to -1 indicates a stronger negative correlation, and a value close to 0 indicates no linear correlation. The specific value can be selected based on actual needs, and this invention does not impose further limitations on this.

[0060] Based on the similarity calculation results, the weights of the VOH loss function in the joint optimization objective function are dynamically adjusted. For example, if the similarity between the predicted result and the predicted displacement dynamic curve is low, it indicates that the model may not be following physical laws well or the data fit is not accurate enough. The weights can be adjusted based on the low similarity. If the deviation in terms of physical laws is large, i.e., the prediction of the displacement evolution model differs significantly from the actual predicted displacement curve, the weight of the VOH loss function can be increased to strengthen the constraint of physical laws. If the error in terms of data fitting is large, i.e., the prediction of the neural network model differs significantly from the actual predicted displacement curve, the weight of the VOH loss function can be decreased to strengthen data-driven training.

[0061] By adjusting weights based on similarity, the model can pay more attention to aspects of inaccurate predictions during training. When the model performs poorly in following physical laws, increasing the weight of physical constraints encourages the model to learn displacement evolution laws that better conform to physical reality. When the model's ability to fit data is insufficient, increasing the weight of the data-driven loss function improves the model's adaptability to data, allowing the model to optimize according to actual conditions at different training stages and improve the overall performance of the model.

[0062] The purpose of setting a maximum number of iterations is to prevent the model from overfitting or getting stuck in an infinite loop during training. Overfitting occurs when a model performs well on training data but poorly on new data; limiting the number of iterations can prevent this from happening to some extent.

[0063] Based on the needs of practical applications, a preset prediction accuracy index is established, such as the mean squared error (MSE) being less than a certain threshold or the correlation coefficient being greater than a certain threshold. For example, in slope displacement prediction, if the mean squared difference between the predicted and actual displacements is required to be less than 0.01, or the correlation coefficient between the predicted and actual displacements is required to be greater than 0.9, these two values ​​can be used as the preset prediction accuracy. Presetting the prediction accuracy ensures that the model stops training only after it reaches a certain level of prediction accuracy, guaranteeing that the model can meet the requirements of practical engineering applications.

[0064] During training, after each iteration, it is checked whether the maximum number of iterations or the preset prediction accuracy has been reached. If the maximum number of iterations has been reached, training stops regardless of whether the preset prediction accuracy has been achieved; if the preset prediction accuracy has been achieved, training stops even if the maximum number of iterations has not been reached. After stopping iteration, the neural network model obtained from the current training is the time-series prediction neural network model. After multiple rounds of iterative training and weight adjustment, this model has been able to effectively integrate data-driven and physical laws, possessing high prediction accuracy and reliability, and can be used for practical slope displacement prediction tasks.

[0065] In some embodiments of the present invention, the monitoring data of the slope area to be monitored is input into a time-series predictive neural network model, and a complete future displacement prediction sequence is generated through inverse mode reconstruction, further comprising: The monitoring data of the slope area to be monitored is input into the time-series predictive neural network model to predict future slope displacement. The predicted values ​​of each modal component are superimposed point by point using the inverse modal reconstruction method to generate a complete future displacement sequence.

[0066] In the above embodiments, after the neural network model training is completed, the slope displacement evolution trend in the future period is predicted based on the latest input monitoring data of multiple slope areas to be monitored (including GNSS modal sequences, their derivative characteristics, and environmental factors such as rainfall, pore pressure, temperature and humidity).

[0067] A fixed-length time window is traced back from the current moment to extract the preprocessed modal sequence and synchronization environment factors, which serve as the input sequence for the model. Each row of the input sequence contains multiple modalities and their derived features (velocity, acceleration) and environment factors, forming a matrix data structure with a unified format. This sequence is then standardized and input into the trained temporal prediction neural network model.

[0068] Based on the input data, the model infers and outputs predictions for the next few hours to days, which may include predictions for one or more modal components. If the model only outputs trend modes (dominant low-frequency components), these are used to construct the main displacement trend lines; if the model is designed with a multi-channel structure, it can also output mid-frequency disturbance modes simultaneously.

[0069] Since the original GNSS signal has been decomposed into multiple physically meaningful frequency components through mode decomposition during the preprocessing stage, "inverse mode reconstruction" is required at the prediction end to restore the complete displacement signal. The key to this process is to superimpose each mode point by point while maintaining its independence, reconstructing the complete predicted displacement value for each future moment. Because high-frequency modes are not the primary modeling target during the modeling stage, their fluctuation patterns are generally not gradually regressed during prediction. Instead, a "static completion" strategy is used for estimation: the average value, variance range, or periodic structure of high-frequency modes over a past period can be used for supplementation, thereby preserving certain disturbance characteristics without affecting trend judgment.

[0070] To better implement the slope displacement prediction method in this embodiment of the invention, based on the slope displacement prediction method, please refer to the corresponding documentation. Figure 7 , Figure 7 This is a schematic diagram of an embodiment of the slope displacement prediction device provided by the present invention. The embodiment of the present invention provides a slope displacement prediction device 700, comprising: The data acquisition and processing module 710 is configured to acquire displacement data and multi-source environmental factor data of the slope area to be monitored, and to preprocess the displacement data and multi-source environmental factor data. The mode decomposition module 720 is configured to use the variational mode decomposition method to decouple the displacement data from time and frequency to obtain multiple modal components, and then merge them with multi-source environmental factor data to obtain an input matrix. The model training module 730 is configured to introduce a physical supervision loss function into the constructed initial predictive neural network model and train a temporal predictive neural network model based on the error feedback mechanism and the input matrix. The displacement prediction module 740 is configured to input monitoring data of the slope area to be monitored into a time-series prediction neural network model and generate a complete future displacement prediction sequence through inverse mode reconstruction.

[0071] It should be noted that the device 700 provided in the above embodiments can implement the technical solutions described in the above method embodiments. The specific implementation principles of the above modules or units can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0072] Please see Figure 8 , Figure 8 This is a schematic diagram of the slope displacement prediction device provided in an embodiment of the present invention. Based on the above-described slope displacement prediction method, the present invention also provides a slope displacement prediction device, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The slope displacement prediction device 800 includes a processor 810, a memory 820, and a display 830. Figure 8 Only some components of the slope displacement prediction device are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0073] In some embodiments, the memory 820 may be an internal storage unit of the slope displacement prediction device 800, such as a hard disk or memory of the slope displacement prediction device 800. In other embodiments, the memory 820 may be an external storage device of the slope displacement prediction device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the slope displacement prediction device 800. Furthermore, the memory 820 may include both internal and external storage units of the slope displacement prediction device 800. The memory 820 is used to store application software and various types of data installed on the slope displacement prediction device 800, such as the program code for installing the slope displacement prediction device 800. The memory 820 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 820 stores a slope displacement prediction program 840, which can be executed by the processor 810 to implement the slope displacement prediction methods of the embodiments of this application.

[0074] In some embodiments, processor 810 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 820 or process data, such as executing slope displacement prediction methods.

[0075] In some embodiments, display 830 may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 830 is used to display information from the slope displacement prediction device 800 and to display a user interface for visualization. Components 810-830 of the slope displacement prediction device 800 communicate with each other via a system bus.

[0076] In one embodiment, when the processor 810 executes the slope displacement prediction program 840 in the memory 820, the steps in the slope displacement prediction method described above are implemented.

[0077] This embodiment also provides a computer-readable storage medium storing a slope displacement prediction program, which, when executed by a processor, performs the following steps: Displacement data and multi-source environmental factor data of the slope area to be monitored are collected, and the displacement data and multi-source environmental factor data are preprocessed. The variational mode decomposition method is used to decouple the displacement data from time to frequency to obtain multiple modal components, which are then merged with multi-source environmental factor data to obtain the input matrix; A physical supervision loss function is introduced into the constructed initial prediction neural network model, and a time-series prediction neural network model is obtained by training based on the error feedback mechanism and the input matrix. The monitoring data of the slope area to be monitored is input into the time-series prediction neural network model, and a complete future displacement prediction sequence is generated through inverse mode reconstruction.

[0078] In summary, the slope displacement prediction method provided by this invention simultaneously collects displacement data and multi-source environmental factor data of the slope area to be monitored. Preprocessing the collected data removes noise and outliers, improving data quality and reliability, and addressing the lack of effective data cleaning mechanisms for raw sequences in existing technologies. By using variational mode decomposition to decouple the displacement data from time to frequency, different frequency components such as long-term trends, periodic disturbances, and noise can be accurately separated. Compared to traditional methods, this improves the processing effect of non-stationary signals and avoids slope displacement prediction distortion caused by high-frequency noise and low-frequency drift. Combining the modal components of the decoupled multi-source environmental factor data with the displacement data to construct a standardized multivariate input matrix establishes a causal relationship between the two over time, enabling more accurate capture of the dynamic driving effect of environmental changes on displacement. When training the prediction neural network model, a physical supervision loss function based on a displacement evolution model is introduced for constraint. The strength of the physical constraint is dynamically adjusted through an error feedback mechanism to prevent prediction bias during landslide abrupt changes, improving the stability of long-term predictions.

[0079] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0080] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0081] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0083] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0084] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0085] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0086] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0087] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0088] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting slope displacement, characterized in that, include: Displacement data and multi-source environmental factor data of the slope area to be monitored are collected, and the displacement data and multi-source environmental factor data are preprocessed. The displacement data is decoupled from the time frequency using the variational mode decomposition method to obtain multiple modal components, which are then merged with the multi-source environmental factor data to obtain the input matrix. A physical supervision loss function is introduced into the constructed initial prediction neural network model, and a time-series prediction neural network model is obtained by training based on the error feedback mechanism and the input matrix. The monitoring data of the slope area to be monitored is input into the time-series prediction neural network model, and a complete future displacement prediction sequence is generated through inverse mode reconstruction.

2. The slope displacement prediction method as described in claim 1, characterized in that, The process of collecting displacement data and multi-source environmental factor data of the slope area to be monitored, and preprocessing the displacement data and multi-source environmental factor data, also includes: GNSS receivers and multi-source sensors are deployed in the slope area to be monitored to collect displacement data and multi-source environmental factor data of the slope area to be monitored. The multi-source environmental factor data are time-stamped and standardized in data format to construct multi-source monitoring data with a unified time axis. The displacement data and the multi-source monitoring data are preprocessed by handling missing values, detecting abrupt changes, and filtering data.

3. The slope displacement prediction method as described in claim 1, characterized in that, The step of using variational mode decomposition to decouple displacement data in time and frequency to obtain multiple modal components, and merging them with the multi-source environmental factor data to obtain an input matrix, further includes: Variational mode decomposition parameters are set, and the objective function of variational mode decomposition is solved iteratively by alternating direction multiplier method. The displacement data is then decoupled in time and frequency to obtain multiple modal components. Calculate the modal frequency and variance contribution ratio, and classify the modal components based on the modal frequency and variance contribution ratio; The modal components are merged with the synchronization sequence of the multi-source environmental factor data to form a standardized input matrix.

4. The slope displacement prediction method as described in claim 3, characterized in that, The process of setting variational mode decomposition parameters, iteratively solving the objective function of variational mode decomposition using the alternating direction multiplier method, and performing time-frequency decoupling on the displacement data to obtain multiple modal components further includes: Based on the time-frequency characteristics of the displacement data, the number of decomposition modes, the penalty factor, and the convergence tolerance threshold of the variational mode decomposition are set. The Lagrange multiplier method is used to construct the optimization objective function of variational mode decomposition, and the solution is obtained iteratively by alternating direction multiplier method until the preset convergence condition is met. The modal components at convergence are reorganized based on the actual physical meaning and frequency structure of the modes.

5. The slope displacement prediction method as described in claim 1, characterized in that, The step of introducing a supervised loss function into the constructed initial prediction neural network model and training it based on the error feedback mechanism and the input matrix to obtain a time-series prediction neural network model further includes: The velocity and acceleration information of the displacement are extracted from the modal components, and the physical supervision loss function of the displacement evolution model and the main loss function of the initial prediction neural network model are constructed. The physical supervision loss function and the main loss function are weighted and fused to form a joint optimization objective function; An error feedback mechanism is introduced, and the initial prediction neural network model is iteratively trained using the input matrix and the joint optimization objective function to obtain a time-series prediction neural network model.

6. The slope displacement prediction method as described in claim 5, characterized in that, The method of introducing an error feedback mechanism and iteratively training the initial prediction neural network model using the input matrix and the joint optimization objective function to obtain a time-series prediction neural network model further includes: The predicted displacement dynamic curve is generated based on the prediction results of each round of training of the neural network model; The weights of the joint optimization objective function are adjusted based on the similarity between the prediction results of each training round and the predicted displacement dynamic curve. When the maximum number of iterations or the preset prediction accuracy is reached, the iterative training stops and the temporal prediction neural network model is obtained.

7. The slope displacement prediction method as described in claim 1, characterized in that, The step of inputting monitoring data of the slope area to be monitored into the time-series prediction neural network model and generating a complete future displacement prediction sequence through inverse mode reconstruction also includes: The monitoring data of the slope area to be monitored is input into the time-series prediction neural network model to predict future slope displacement. The predicted values ​​of each modal component are superimposed point by point using the inverse modal reconstruction method to generate a complete future displacement sequence.

8. A slope displacement prediction device, characterized in that, include: The data acquisition and processing module is configured to acquire displacement data and multi-source environmental factor data of the slope area to be monitored, and to preprocess the displacement data and the multi-source environmental factor data. The mode decomposition module is configured to use the variational mode decomposition method to decouple the displacement data from time and frequency to obtain multiple modal components, and then merge them with the multi-source environmental factor data to obtain an input matrix. The model training module is configured to introduce a physical supervision loss function into the constructed initial predictive neural network model and train it based on the error feedback mechanism and the input matrix to obtain a temporal predictive neural network model. The displacement prediction module is configured to input monitoring data of the slope area to be monitored into the time-series prediction neural network model and generate a complete future displacement prediction sequence through inverse mode reconstruction.

9. A slope displacement prediction device, characterized in that, It includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the steps of the slope displacement prediction method according to any one of claims 1 to 7.

10. A storage medium, characterized in that, It stores a computer program that can be executed by the slope displacement prediction device. When the computer program is run on the slope displacement prediction device, it causes the slope displacement prediction device to perform the steps of the slope displacement prediction method according to any one of claims 1 to 7.

Images