Highway high slope landslide early warning method based on multi-source data fusion

Abstract

The invention discloses a highway high slope landslide early warning method based on multi-source data fusion. The method comprises the following steps: collecting high slope multi-source data; cleaning, synchronizing and standardizing the high slope multi-source data, and constructing a time-space association data set; a future displacement trend is predicted based on an improved ARIMA-GARCH dynamic time sequence model, a modified Mohr-Coulomb slope stability mechanical model is established in combination with a rock-soil mechanical strength criterion, and a potential sliding surface is determined; sHAP value analysis and RF screening key features are utilized to construct a landslide early warning model fused with multi-source data; a model is trained through historical data, parameters are optimized, risk grade dynamic evaluation is achieved in combination with real-time monitoring data, and graded early warning signals are triggered through a DBN; according to the invention, the problems of data isolation and insufficient model generalization ability of an existing early warning system are solved, and high-precision and real-time landslide active early warning is realized.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster early warning and slope monitoring technology, in particular to a highway high slope landslide early warning method based on multi-source data fusion, which is applicable to the fields of transportation, infrastructure construction, natural disaster monitoring, etc. Background Art

[0002] Landslides on high highway slopes are often influenced by a combination of factors, including geological structures, rainfall, earthquakes, and other natural factors. Due to the dynamic nature of these factors, accurate prediction of slope stability is difficult, making landslides both hidden and sudden. Consequently, high-slope landslides not only threaten driving safety but can also cause significant damage to residential areas, public facilities, and the environment along the highway. Slope stability is particularly vulnerable to disruption during extreme weather events (such as persistent heavy rain or drought) or earthquakes, leading to landslides.

[0003] Traditional landslide monitoring methods typically rely on single sensors or manual inspections. Sensors are typically used to monitor physical changes such as displacement and settlement, but this approach suffers from the low dimensionality of the monitoring data, making it difficult to fully reflect slope stability. Furthermore, while manual inspections can provide intuitive on-site observations, they rely on human judgment, are labor-intensive, and inefficient, making it difficult to detect potential landslide risks in a timely manner. Furthermore, traditional methods lack real-time performance, and data collection and processing often lag behind, resulting in an insufficiently fast early warning response. This makes it difficult to issue warnings before a landslide occurs, increasing the threat to people and property when a landslide occurs.

[0004] Currently, traditional early warning methods typically rely on geological surveys and manual inspections. Monitors visually inspect slopes for surface signs such as cracks and deformation, and infer slope stability based on weather conditions (such as rainfall and temperature fluctuations). However, a single monitoring method is insufficient to meet the monitoring needs of high-slope landslides on highways. Multi-source data fusion and intelligent analysis methods are needed to improve early warning accuracy and achieve more precise and timely landslide disaster prevention.

[0005] Existing slope deformation monitoring methods fail to integrate geotechnical parameters with protection status data, resulting in unclear physical meaning of the models. Slope displacement predictions fail to account for temporal and spatial heterogeneity, limiting generalization capabilities. Furthermore, existing early warning systems often rely on static thresholds and are unable to dynamically adapt to complex environmental changes. Summary of the Invention

[0006] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a highway high slope landslide early warning method based on multi-source data fusion, which can provide effective support for the prevention and mitigation of landslide disasters.

[0007] To achieve the above object, the present invention adopts a technical solution: a highway high slope landslide early warning method based on multi-source data fusion, comprising the following steps:

[0008] Step 1: Collect multi-source data of high slope;

[0009] Step 2: Clean, synchronize and standardize the multi-source data of the high slope to construct a spatiotemporal correlation dataset;

[0010] Step 3: Based on the improved ARIMA-GARCH dynamic time series model, the future displacement trend is predicted. Combined with the geotechnical strength criterion, a modified Mohr-Coulomb slope stability model is established to determine the potential sliding surface.

[0011] Step 4: Use SHAP value analysis and RF to screen key features and build a landslide early warning model that integrates multi-source data;

[0012] Step 5: Use historical data to train the model and optimize parameters, combine it with real-time monitoring data to achieve dynamic risk level assessment, and trigger graded warning signals through DBN.

[0013] As a further improvement of the present invention, in step 1, multi-source data of high slopes are collected through integrated air-space monitoring equipment, manual inspections and regular inspections. The multi-source data of high slopes include displacement settlement, deformation rate, rock and soil strength, structural surface characteristics and protection and reinforcement status data.

[0014] As a further improvement of the present invention, parameters of rock and soil strength and the inclination angle of the structural surface are collected through manual inspection. The parameters of rock and soil strength include cohesion and internal friction angle. They are measured through triaxial tests, and the formula is:

[0015] τ=c'+σ' n tanφ'#(1)

[0016] Where c' represents cohesion, σ' n represents the normal stress, and φ′ represents the internal friction angle.

[0017] As a further improvement of the present invention, it is characterized in that, in step 1, it also includes: installing displacement sensors, strain gauges, and inclinometers for long-term monitoring, and regularly recording and tracking the collected data.

[0018] As a further improvement of the present invention, the step 2 specifically includes the following steps:

[0019] Step 2.1: Check the integrity of the data, handle missing values, outliers, and erroneous data, and correct erroneous data or delete unreliable data;

[0020] Step 2.2: Synchronize data from different sources in time and space. Temporally align data from different time periods to ensure that all data are collected in the same time period. Spatially align data from different spatial scales to ensure that the data are spatially consistent.

[0021] Step 2.3: Convert and standardize the displacement data, settlement data, and geotechnical strength data; and analyze all data on the same scale.

[0022] Step 2.4: Construct a spatiotemporal correlation dataset to associate different types of monitoring data in time and space, providing a dynamically changing dataset for subsequent models.

[0023] As a further improvement of the present invention, in step 2.1, the isolation forest algorithm is used to identify abnormal data, and spatiotemporal kriging interpolation is used to fill in missing data caused by sensor failure or communication interruption.

[0024] As a further improvement of the present invention, the step 3 specifically includes the following steps:

[0025] Step 3.1, extracting time series displacement data from the constructed spatiotemporal correlation dataset;

[0026] Step 3.2: Determine the order (p, d, q) of the ARIMA model through the autocorrelation function and estimate the model parameters; the model expression is as follows:

[0027] Δd t =φ1Δd t-1 +…+φ p Δd t-p +∈ t +θ1∈ t-1 +…+θ q ∈ t-q #(2)

[0028] Where Δd t represents the displacement change after differentiation, φ1,…,φ p represents the autoregressive coefficients, θ1,…,θ q represents the moving average coefficient, ∈ t represents the residual term;

[0029] Step 3.3: Use the GARCH model to capture the volatility characteristics of the time series and estimate the model parameters. The model expression is:

[0030]

[0031] in, is the conditional variance, which indicates the intensity of fluctuation at time t, α0 indicates the basic fluctuation level, and α1 indicates the square of the residual error. The impact weight of β1 represents the historical fluctuation The weight of the persistent effect;

[0032] Step 3.4: Establish a modified Mohr-Coulomb slope stability mechanical model. Based on geological exploration data and laboratory test results, determine the mechanical parameters of the soil layer, including cohesion, internal friction angle, elastic modulus, and Poisson's ratio. A modified model that considers nonlinearity, strain softening, and time effects is introduced to adapt the model to the actual slope conditions.

[0033] Step 3.5: Adjust the parameters of the Mohr-Coulomb model through numerical simulation or optimization methods based on monitoring data;

[0034] Step 3.6: Calculate the safety factor under different conditions based on the modified Mohr-Coulomb model. The formula is:

[0035]

[0036] Where c′ represents the effective cohesion; φ′ represents the effective internal friction angle; σ n represents normal stress; u a ,u w represents pore gas pressure and pore water pressure; X represents the empirical coefficient.

[0037] As a further improvement of the present invention, step 4 specifically includes the following steps:

[0038] Step 4.1: Based on the constructed spatiotemporal correlation dataset, use RF to preliminarily screen the TOP-20 features;

[0039] Step 4.2: After training the model, use the SHAP value to analyze the importance of each feature to the model prediction structure, and select the key features that have the greatest impact on landslides. Construct a deep belief network (DBN) with a hidden layer structure of 256, 128, and 64. The loss function is weighted cross entropy, as shown in the following formula:

[0040]

[0041] in, Represents the SHAP value of the jth sample, and the quantitative feature i represents the contribution to the prediction result. i represents the reduction of Gini impurity of feature i in random forest, N, M represent the number of samples and the number of features, w c represents the category weight, y c represents the true label, represents the model prediction probability;

[0042] Step 4.3: Based on the selected important features, a landslide warning model is constructed using the RF algorithm.

[0043] As a further improvement of the present invention, the step 5 specifically includes the following steps:

[0044] Step 5.1: Train and validate the model. The dataset is divided into 70% training set, 15% validation set, and 15% test set.

[0045] Step 5.2: Input real-time data into the DBN model, output the landslide probability P, and trigger multi-level responses, including attention level, warning level, alert level, and alarm level.

[0046] The present invention is based on data fusion technology from multiple monitoring methods (such as sensors, remote sensing technology, meteorological data, seismic data, etc.), and can collect large amounts of multi-dimensional data in real time. This invention, for the first time, fuses a dynamic time series model (ARIMA-GARCH) with a geotechnical model (modified Mohr-Coulomb), and quantifies feature contributions through SHAP values, addressing the drawbacks of data isolation and static threshold dependency in existing technologies. Furthermore, the present invention collects slope displacement, stress, precipitation, and other data in real time, and processes and analyzes the data using big data analysis and artificial intelligence algorithms, enabling intelligent assessment and early warning of landslide risks. Based on multi-source data fusion, the system can provide accurate landslide warning information and promptly trigger emergency response mechanisms, effectively improving highway safety and emergency management capabilities.

[0047] The beneficial effects of the present invention are:

[0048] The present invention provides a highway high-slope landslide early warning model based on multi-source data fusion, which has the following effects compared with a single monitoring method:

[0049] 1. Traditional landslide monitoring typically relies on a single data source, potentially overlooking the multiple factors in complex environments. However, this invention, through multi-source data fusion, not only enables more comprehensive monitoring of slope conditions but also provides richer risk assessment information. By processing spatiotemporally correlated data and accurately integrating different data sources in time and space, the model captures more changing trends, improving the reliability and accuracy of the early warning system.

[0050] 2. The improved ARIMA-GARCH dynamic time series model is used to predict future displacement trends. At the same time, the modified Mohr-Coulomb slope stability mechanical model is combined to further evaluate the stability of the slope.

[0051] 3. Combining SHAP value analysis and the RF algorithm for feature screening further improves the model's interpretability and the scientific nature of feature selection. SHAP values provide an interpretable analysis of each feature's contribution to model predictions, helping to better understand which features are most important for landslide early warning. Furthermore, DBN is used to trigger graded warning signals, enabling automated landslide risk assessment and early warning based on real-time data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of an embodiment of the present invention;

[0053] Figure 2 Schematic diagram of the DBN model structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0055] Example

[0056] like Figure 1 As shown, a highway high slope landslide early warning method based on multi-source data fusion includes the following steps:

[0057] S1: Collect multi-source data on high slopes through integrated air and space monitoring equipment, manual inspections, and regular testing, including displacement and settlement, deformation rate, rock and soil strength, structural surface characteristics, and protection and reinforcement status data;

[0058] The specific process of step S1 includes the following steps:

[0059] S1.1: Obtain slope displacement, deformation rate, and image data through integrated aerospace equipment;

[0060] Specifically, in order to meet the needs of model learning and further improve data diversity, satellite remote sensing can be used to monitor slope deformation on a large scale, and drone aerial photography can be used to supplement local details, thus realizing the fusion of macro and micro data.

[0061] S1.2: Manual inspections are conducted to collect geotechnical strength parameters (cohesion, internal friction angle), structural surface inclination, and anchor tension data, and these are measured through triaxial tests. The formula is:

[0062] τ=c′+σ′ n tanφ′#(1)

[0063] In the above formula, c′ represents cohesion, σ′ n represents the normal stress, and φ′ represents the internal friction angle.

[0064] S1.3: Install displacement sensors, strain gauges, inclinometers, and other equipment for long-term monitoring, and regularly record and track the collected data. This provides standardized input for mechanical models, and the diverse data can also address traditional monitoring challenges.

[0065] S2: Clean, synchronize, and standardize the data to construct a spatiotemporal correlation dataset. Data from different sources (such as remote sensing and ground monitoring) are synchronized in time and space to ensure alignment within the same time period and spatial scale. Through data standardization, different data units are unified to provide a consistent data foundation for subsequent analysis. Finally, a spatiotemporal correlation dataset is constructed to correlate different types of monitoring data (displacement, settlement, deformation rate, etc.) in time and space, providing a dynamically changing dataset for subsequent models.

[0066] The specific process of step S2 includes the following steps:

[0067] S2.1: Check data integrity and handle missing values, outliers, and erroneous data. Erroneous data are corrected or unreliable data is deleted. During preprocessing, an innovative Isolation Forest (IF) algorithm (with 100 trees and an outlier threshold of 0.02) is used to identify anomalous data. Spatial-temporal kriging interpolation is used to fill in missing data caused by sensor failures or communication interruptions.

[0068] S2.2: Synchronize data from different sources (e.g., remote sensing data and ground monitoring data) in time and space. Temporally align data from different time periods to ensure that all data are collected within the same time period. Spatially align data from different spatial scales to ensure that the data are spatially consistent.

[0069] S2.3: Convert and standardize displacement, settlement, and geotechnical strength data. Analyze all data on the same scale. A sliding window approach (with a 24-hour window size) and spatial alignment ensure seamless integration of multi-source data, ensuring consistency for subsequent data input and improving data calculation speed. Standardization makes displacement, settlement, and geotechnical strength data comparable, enhancing data robustness and improving the model's generalization capabilities.

[0070] S3: Based on the improved ARIMA-GARCH dynamic time series model, the future displacement trend is predicted. In combination with the geotechnical strength criterion, a modified Mohr-Coulomb slope stability mechanical model is established to determine the potential sliding surface. The ARIMA model is used to capture the time series characteristics in the data, while the GARCH model is used to deal with volatility and predict future displacement trends. Using the modified Mohr-Coulomb model, combined with the geotechnical strength criterion, parameters such as shear strength and internal friction angle are analyzed to determine the potential sliding surface of the slope and evaluate the stability of the slope. Based on the improved ARIMA-GARCH dynamic time series model to predict future displacement trends, different patterns such as trends, periodicity, and nonlinear fluctuations can be identified in time series data. This is of great significance for the prediction of geological disasters such as landslides. In combination with the geotechnical strength criterion, a modified Mohr-Coulomb slope stability mechanical model is established, making the slope stability analysis more accurate and providing more reliable prediction results under complex geological conditions.

[0071] The specific process of step S3 includes the following steps:

[0072] S3.1: Extract time series displacement data from the spatiotemporal correlation dataset constructed in S2, with an interval of 1 hour and continuous recording for more than 30 days, and ensure the stationarity of the time series;

[0073] S3.2: Determine the order (p, d, q) of the ARIMA model using the autocorrelation function (ACF) or partial autocorrelation function (PACF), and estimate the parameters of the ARIMA model using the least squares method or maximum likelihood method. The model expression is as follows:

[0074] Δd t =φ1Δd t-1 +…+φ p Δd t-p +∈ t +θ1∈ t-1 +…+θ q ∈ t-q #(2)

[0075] In the above formula, Δd t represents the displacement change after differentiation, φ1,…,φ p Represents the autoregressive coefficient, reflecting the impact of historical displacement on the current value. θ1,…θ q Represents the moving average coefficient, reflecting the impact of historical residuals on the current value, ∈ tRepresents the residual term. The ARIMA model is used for its powerful time series forecasting capabilities, particularly for time-dependent data. By combining autoregressive (AR) and moving average (MA) components and smoothing the data through differencing (I), the ARIMA model effectively captures trends, seasonality, and cyclical variations in time series. This ensures that the model accurately fits historical data and enables precise forecasting. Furthermore, this method requires no external predictor variables, offers good generalizability, and offers low computational complexity. It uses maximum likelihood estimation to solve for parameters.

[0076] S3.3: Capture the volatility characteristics of the time series through the GARCH model. Estimate the model parameters and check the model residuals to ensure that they exhibit white noise characteristics (i.e., no autocorrelation and conform to the normal distribution). The model expression is:

[0077]

[0078] In the above formula, is the conditional variance, which indicates the intensity of fluctuation at time t, α0 indicates the basic fluctuation level, and α1 indicates the square of the residual error. The impact weight of β1 represents the historical fluctuation The GARCH model can effectively capture the volatility characteristics of time series. By modeling conditional variance, GARCH can capture the phenomenon of volatility clustering, reflect the change of volatility intensity over time, and help more accurately predict future uncertainties and risks.

[0079] S3.4: Develop a modified Mohr-Coulomb slope stability mechanical model. Based on geological exploration data and laboratory test results, determine the mechanical parameters of the soil layer, including cohesion, internal friction angle, elastic modulus, Poisson's ratio, etc., and introduce a modified model that considers factors such as nonlinearity, strain softening, and time effects to make the model adaptable to the actual slope conditions.

[0080] S3.5: Through numerical simulation (such as finite element analysis) or optimization methods based on monitoring data, the parameters of the Mohr-Coulomb model are adjusted to better fit the mechanical behavior of the actual slope. A modified Mohr-Coulomb slope stability model is used to comprehensively account for the complexity of the actual slope situation. By introducing correction factors such as nonlinearity, strain softening, and time effects, the model can more accurately reflect the actual mechanical behavior of the soil. Combined with geological exploration data and laboratory test results, it can provide more accurate predictions for slope stability analysis. This further improves the model's adaptability and predictive accuracy, ensuring reasonable slope stability assessment and risk warning.

[0081] S3.6: Calculate the safety factor under different conditions based on the modified Mohr-Coulomb model. The formula is as follows. Draw the sliding surface and perform sliding surface analysis using the slope stability analysis software FLAC:

[0082]

[0083] In the above expression, c′ represents the effective cohesion, which reflects the shear strength of the rock and soil; φ′ represents the effective internal friction angle, which describes the friction characteristics between particles; σ n represents normal stress; u a ,u w represents the pore gas pressure and pore water pressure; χ represents the empirical coefficient (taken as 0.8), which is used to correct the unsaturated soil effect.

[0084] As shown in formula (4), the introduction of χ(u a -u w ) better reflects the unsaturated state of actual slopes, providing more accurate slope stability analysis. By incorporating the effects of pore gas and pore water pressures, it accurately assesses safety factors under different conditions, ensuring the reliability and accuracy of slope risk assessments.

[0085] S4: Using SHAP value analysis and RF to screen key features, a landslide early warning model integrating multi-source data was constructed;

[0086] The specific process of step S4 includes the following steps:

[0087] S4.1: Based on the dataset constructed above, use RF to initially screen the top 20 features. The advantage of using RF to screen the top 20 features is that it automatically identifies the most important features for prediction, reduces redundant variables, and improves model interpretability and predictive accuracy. RF also has strong robustness to data noise and missing values, improving model stability.

[0088] S4.2: After training the model, use the SHAP value to analyze the importance of each feature to the model prediction structure, and select the key features that have the greatest impact on landslides, and construct Figure 2 The deep belief network (DBN) shown in the figure has a hidden layer structure of 256, 128, and 64, and the loss function is weighted cross entropy, as shown in the following formula:

[0089]

[0090] Based on the above formula, formula (5) calculates the reduction of the Gini impurity of feature i, reflecting its classification importance in the random forest; formula (6) combines the absolute value of the SHAP value and the Gini weight to quantify the comprehensive contribution of the feature to landslide prediction; formula (7) optimizes the DBN model through the weighted cross entropy loss function to improve the prediction weight of the minority class (such as the alarm level).

[0091] in Represents the SHAP value of the jth sample, and the quantitative feature i represents the contribution to the prediction result. i represents the reduction of Gini impurity of feature i in random forest, reflecting the importance of classification, N, M represent the number of samples and the number of features, w c represents the category weight, y c represents the true label, Represents the model predicted probability.

[0092] The proposed method uses SHAP values to analyze feature importance and combines this with DBN model training to comprehensively assess the contribution of each feature to landslide prediction, thereby accurately identifying key features. SHAP values provide a quantitative estimate of each feature's contribution to the prediction, ensuring the scientific and interpretable nature of feature selection. The deep learning of DBN and the weighted cross-entropy loss function help improve the model's classification accuracy and robustness. Combined with the reduction of Gini impurity, the method further optimizes feature selection and weighting, increasing the loss weight of minority classes (such as the alarm level) and preventing the model from favoring the majority class.

[0093] S4.3: Based on the selected important features, a landslide early warning model is constructed using the RF algorithm.

[0094] S5: Train the model and optimize the parameters through historical data, combine it with real-time monitoring data to realize dynamic risk level assessment, and trigger graded warning signals through DBN.

[0095] The specific process of step S5 includes the following steps:

[0096] S5.1: Train and validate the model. The dataset is divided into 70% training set, 15% validation set, and 15% test set.

[0097] S5.2: Input real-time data into the DBN model, output the landslide probability P, and trigger a multi-level response. The attention level (0.3≤P<0.5), warning level (0.5≤P<0.7), and alert level (0.7≤P<0.9) trigger on-site personnel evacuation and traffic control, and the alarm level (P≥0.9) activates the emergency rescue plan.

[0098] Model training and validation utilize a data partitioning strategy of 70% training, 15% validation, and 15% test sets, effectively avoiding overfitting and ensuring good model generalization. By inputting real-time data into the DBN model and outputting the landslide probability P, the model can monitor landslide risks in real time and trigger appropriate response mechanisms. Setting warning, alert, and alarm levels based on different probability intervals accurately reflects the likelihood of landslides, effectively guiding decision-makers to take timely preventive measures, improving the real-time and accuracy of the landslide early warning system and ensuring public safety.

[0099] The above-described embodiments merely represent specific implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A highway high slope landslide early warning method based on multi-source data fusion, characterized by: The following steps are involved: Step 1: Collect multi-source data of high slope; Step 2: Clean, synchronize and standardize the multi-source data of the high slope to construct a spatiotemporal correlation dataset; Step 3: Based on the improved ARIMA-GARCH dynamic time series model, the future displacement trend is predicted. Combined with the geotechnical strength criterion, a modified Mohr-Coulomb slope stability model is established to determine the potential sliding surface. Step 4: Use SHAP value analysis and RF to screen key features and build a landslide early warning model that integrates multi-source data; Step 5: Use historical data to train the model and optimize parameters, combine it with real-time monitoring data to achieve dynamic risk level assessment, and trigger graded warning signals through DBN.

2. The highway high slope landslide early warning method based on multi-source data fusion according to claim 1 is characterized in that: In step 1, multi-source data of high slopes are collected through integrated air-space monitoring equipment, manual inspections and regular inspections. The multi-source data of high slopes include displacement settlement, deformation rate, rock and soil strength, structural surface characteristics and protection and reinforcement status data.

3. The highway high slope landslide early warning method based on multi-source data fusion according to claim 2 is characterized in that: Manual inspections were conducted to collect parameters of geotechnical strength and the inclination of the structural surface. The parameters of geotechnical strength include cohesion and internal friction angle. They were also measured through triaxial tests. The formula is: τ=c′+σ′ n tanφ′#(1) Where c′ represents the cohesion, σ′ n represents the normal stress, and φ′ represents the internal friction angle.

4. The highway high slope landslide early warning method based on multi-source data fusion according to claim 1, 2 or 3, characterized in that: In step 1, it also includes: installing displacement sensors, strain gauges, and inclinometers for long-term monitoring, and regularly recording and tracking the collected data.

5. The highway high slope landslide early warning method based on multi-source data fusion according to claim 1 is characterized in that: The step 2 specifically includes the following steps: Step 2.1: Check the integrity of the data, handle missing values, outliers, and erroneous data, and correct erroneous data or delete unreliable data; Step 2.2: Synchronize data from different sources in time and space. Temporally align data from different time periods to ensure that all data are collected in the same time period. Spatially align data from different spatial scales to ensure that the data are spatially consistent. Step 2.3: Convert and standardize the displacement data, settlement data, and geotechnical strength data; and analyze all data on the same scale. Step 2.4: Construct a spatiotemporal correlation dataset to associate different types of monitoring data in time and space, providing a dynamically changing dataset for subsequent models.

6. The highway high slope landslide early warning method based on multi-source data fusion according to claim 5 is characterized in that: In step 2.1, the isolation forest algorithm is used to identify abnormal data, and spatiotemporal kriging interpolation is used to fill in missing data caused by sensor failure or communication interruption.

7. The highway high slope landslide early warning method based on multi-source data fusion according to claim 5 is characterized in that: The step 3 specifically includes the following steps: Step 3.1, extracting time series displacement data from the constructed spatiotemporal correlation dataset; Step 3.2: Determine the order (p, d, q) of the ARIMA model through the autocorrelation function and estimate the model parameters; the model expression is as follows: Δd t =φ1Δd t-1 +…+φ p Δd t-p +∈ t +θ1∈ t-1 +…+θ q ∈ t-q #(2) Where Δd t represents the displacement change after differentiation, φ1,…,φ p represents the autoregressive coefficients, θ1,…,θ q represents the moving average coefficient, ∈ t represents the residual term; Step 3.3: Use the GARCH model to capture the volatility characteristics of the time series and estimate the model parameters. The model expression is: in, is the conditional variance, which indicates the intensity of fluctuation at time t, α0 indicates the basic fluctuation level, and α1 indicates the square of the residual error. The impact weight of β1 represents the historical fluctuation The weight of the persistent effect; Step 3.4: Establish a modified Mohr-Coulomb slope stability mechanical model. Based on geological exploration data and laboratory test results, determine the mechanical parameters of the soil layer, including cohesion, internal friction angle, elastic modulus, and Poisson's ratio. A modified model that considers nonlinearity, strain softening, and time effects is introduced to adapt the model to the actual slope conditions. Step 3.5: Adjust the parameters of the Mohr-Coulomb model through numerical simulation or optimization methods based on monitoring data; Step 3.6: Calculate the safety factor under different conditions based on the modified Mohr-Coulomb model. The formula is: Where c′ represents the effective cohesion; φ′ represents the effective internal friction angle; σ n represents normal stress; u a ,u w represents the pore gas pressure and pore water pressure; χ represents the empirical coefficient.

8. The highway high slope landslide early warning method based on multi-source data fusion according to claim 7 is characterized in that: The step 4 specifically includes the following steps: Step 4.1: Based on the constructed spatiotemporal correlation dataset, use RF to preliminarily screen the top 20 features; Step 4.2: After training the model, use the SHAP value to analyze the importance of each feature to the model prediction structure, and select the key features that have the greatest impact on landslides. Construct a deep belief network (DBN) with a hidden layer structure of 256, 128, and 64. The loss function is weighted cross entropy, as shown in the following formula: in, Represents the SHAP value of the jth sample, and the quantitative feature i represents the contribution to the prediction result. i represents the reduction of Gini impurity of feature i in random forest, N, M represent the number of samples and the number of features, w c represents the category weight, y c represents the true label, represents the model prediction probability; Step 4.3: Based on the selected important features, a landslide warning model is constructed using the RF algorithm.

9. The highway high slope landslide early warning method based on multi-source data fusion according to claim 8 is characterized in that: The step 5 specifically includes the following steps: Step 5.1: Train and validate the model. The dataset is divided into 70% training set, 15% validation set, and 15% test set. Step 5.2: Input real-time data into the DBN model, output the landslide probability P, and trigger multi-level responses, including attention level, warning level, alert level, and alarm level.

Images