A digital modeling method for landslide surge disaster prediction

Abstract

The present application belongs to the technical field of surge disaster prediction, and particularly relates to a digital modeling method for landslide surge disaster prediction. The landslide surge test data is obtained by using a wave pool model test device, the sensitivity of the surge influence parameters to the surge characteristic parameters is analyzed, and the landslide surge disaster prediction model under the test condition is obtained. The actual surge influence parameters in the engineering case are substituted into the prediction model under the test condition to obtain the predicted surge characteristic parameters of the engineering case. The prediction accuracy of the prediction model is analyzed based on the predicted surge characteristic parameters and the actual surge characteristic parameters, the network parameters of the prediction model under the test condition are adjusted based on the model optimization strategy of the sensitivity of the surge influence parameters, the prediction model is continuously optimized, and the prediction model conforming to the actual engineering is obtained. The defect of the lack of engineering data quantity is solved, the prediction model is more consistent with the actual engineering, and the model prediction effect is improved.

Description

Technical Field

[0001] This invention belongs to the field of surge disaster prediction technology, specifically relating to a digital modeling method for landslide surge disaster prediction. Background Technology

[0002] Currently, the main methods for simulating and predicting landslide-induced surge disasters are numerical simulation methods based on physical processes and data-driven machine learning methods.

[0003] Numerical simulation methods based on physical processes discretize the landslide and water bodies into a finite number of elements and simulate the motion of the landslide and the fluctuation of the water by solving mechanical equations. For example, a two-dimensional or three-dimensional landslide-surge model can be established in relevant simulation software, taking into account the material properties, initial conditions and boundary conditions of the landslide, to simulate the sliding process of the landslide under gravity and its impact on the water, thereby obtaining the generation and propagation process of the surge. This method takes into account the influence of a variety of physical factors and can describe the complex mechanical behavior of the landslide and water bodies more accurately, but it has high computational costs and strict requirements for the initial conditions and parameter settings of the model.

[0004] Data-driven machine learning methods construct multi-layer neural networks and train them using a large amount of landslide surge data. They learn the mapping relationship between input features such as landslide volume, slope, and velocity and output results such as surge height and propagation speed. After training, they can predict surge disasters for new landslide situations. This method has strong nonlinear mapping capabilities and can handle complex input-output relationships. However, it is highly dependent on data and requires a large amount of experimental or engineering data for training. The physical meaning of the model is unclear and its interpretability is poor.

[0005] For data-driven machine learning methods, if only experimental data is used to train the model, there are usually some differences between the experimental device and the actual project. For example, for still water depth, landslide size, and landslide impact velocity, the actual project is usually larger than the experimental device. Therefore, the actual influencing parameters exceed the training range of the landslide surge disaster prediction model, which may lead to the model prediction results not conforming to the actual project. If only engineering data is used to train the model, but engineering data is often scarce, it may lead to poor prediction accuracy and poor prediction effect of the model. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a digital modeling method for predicting landslide surge disasters.

[0007] To achieve the above-mentioned objectives, the technical solution adopted in this invention is as follows: a digital modeling method for landslide surge disaster prediction, which utilizes a wave pool model test device to obtain landslide surge test data, analyzes the sensitivity of surge influence parameters to surge characteristic parameters, trains a landslide surge disaster prediction model using the landslide surge test data, and obtains a landslide surge disaster prediction model under test conditions; substitutes actual surge influence parameters from engineering cases into the landslide surge disaster prediction model under test conditions to obtain predicted surge characteristic parameters for engineering cases; analyzes the prediction accuracy of the landslide surge disaster prediction model based on the predicted surge characteristic parameters and actual surge characteristic parameters, adjusts the network parameters of the landslide surge disaster prediction model under test conditions based on a model optimization strategy based on the sensitivity of surge influence parameters, continuously optimizes the landslide surge disaster prediction model, and finally obtains a landslide surge disaster prediction model that conforms to actual engineering.

[0008] Preferably, the steps include:

[0009] S1. Determine the swell influence parameters and swell characteristic parameters, build a wave pool model test device, conduct experiments using the wave pool model test device, study the swell characteristic parameters under different swell influence parameters, and analyze the sensitivity of the swell influence parameters to the swell characteristics.

[0010] S2. Using the surge wave impact parameters as model input and the surge wave characteristic parameters as model output, a landslide surge wave disaster prediction model is constructed using a neural network. The simulated landslide surge wave test data is used as the training set to train the landslide surge wave disaster prediction model and obtain the landslide surge wave disaster prediction model under test conditions.

[0011] S3. Collect engineering cases of landslide surges, substitute the actual surge impact parameters of the engineering cases into the landslide surge disaster prediction model under test conditions, and obtain the predicted surge characteristic parameters of the engineering cases.

[0012] S4. Draw visualization charts based on the predicted surge characteristic parameters and the actual surge characteristic parameters, identify the anomalies predicted by the landslide surge disaster prediction model, analyze the surge influence parameters based on the anomalies, adjust the network parameters of the landslide surge disaster prediction model under the test conditions based on the model optimization strategy of surge influence parameter sensitivity, continuously optimize the landslide surge disaster prediction model, and finally obtain a landslide surge disaster prediction model that conforms to the actual engineering.

[0013] Preferably, the landslide surge disaster prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes 6 input neurons, the output layer includes 3 output neurons, and two hidden layers are set. The first hidden layer has 12 neurons, and the second hidden layer has 8 neurons. The surge influence parameters are used as model inputs, and the surge characteristic parameters are used as model outputs. The surge influence parameters include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The surge characteristic parameters include maximum standing wave height, propagation velocity, and maximum standing wave height propagation distance.

[0014] Preferably, step S1 includes the following process:

[0015] S11. Construct a wave pool model test device;

[0016] S12. Determine the parameters affecting the surge and the parameters characteristic of the surge. The parameters affecting the surge include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The parameters characteristic of the surge include maximum standing wave height, propagation velocity, and propagation distance at maximum standing wave height.

[0017] S13. Set three levels for each type of surge influence parameter: small, medium, and large. Design the combination of each level of surge influence parameter. Use the wave pool model test device to conduct simulation tests on these combinations and obtain the dataset P of test surge influence parameters - test surge characteristic parameters.

[0018] S14. Use the analysis of variance method to determine the contribution rate of the interaction between the surge influencing parameters to the changes in the surge characteristic parameter results;

[0019] S15. Construct a sensitivity matrix of surge influence parameters and surge characteristic parameters, and for each surge characteristic parameter, sort the six surge influence parameters in descending order of their contribution rate.

[0020] Preferably, in step S1, the sensitivity of surge influence parameters to surge characteristics is analyzed using a multi-factor combination method. Based on the full factorial design, an orthogonal design method is used to select appropriate and representative n combinations for experimentation. The selected n combinations ensure that each level of each surge influence parameter appears the same number of times in the combination and can balance the interaction between surge influence parameters.

[0021] Preferably, step S4 includes the following process:

[0022] S41. Draw a visualization chart based on the predicted surge characteristic parameters and the actual surge characteristic parameters to determine the anomalies predicted by the landslide surge disaster prediction model;

[0023] S42. Traverse all outliers and filter out extreme outliers based on whether the actual surge impact parameters exceed the model training range.

[0024] S43. Depending on whether the outlier is an extreme outlier, different optimization strategies are adopted to adjust the model network parameters, resulting in the adjusted landslide surge disaster prediction model.

[0025] S44. Re-input the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to generate new predicted surge characteristic parameters. Compare the error changes of the original anomaly points until the requirements are met. At the same time, adjust the sensitivity of the surge impact parameter and update the surge impact parameter sensitivity ranking and sensitivity matrix.

[0026] S45. Substitute the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to obtain new predicted surge characteristic parameters for the engineering case. Draw a visualization chart based on the new predicted surge characteristic parameters and the actual surge characteristic parameters. If anomalies still appear, repeat steps S42-S44.

[0027] Preferably, in step S43, if the anomaly is an extreme anomaly and the extreme surge influence parameter is a highly sensitive influence parameter corresponding to the surge characteristic parameter, then in the input layer, an extreme mapping neuron is added to the surge influence parameter to enhance the feature extraction of extreme values ​​through a piecewise function; and / or, the weight initialization method of the surge influence parameter in the hidden layer is adjusted.

[0028] Preferably, in step S43, if the outlier is not an extreme outlier, and if the model prediction deviation for the highly sensitive parameter of the outlier does not match the actual trend of parameter change, a nonlinear transformation layer for the highly sensitive parameter is added in the hidden layer to strengthen the nonlinear mapping; and / or, the input features of the highly sensitive parameter are standardized.

[0029] Preferably, in step S43, if the outlier is not an extreme outlier and is caused by a low-sensitivity influence parameter, then L1 regularization is applied to the weight corresponding to the low-sensitivity influence parameter to force the network to reduce its dependence on it; and / or, perturbation samples of the low-sensitivity influence parameter are added to the training data.

[0030] The present invention has the following beneficial effects:

[0031] Extensive landslide surge test data were obtained using a wave pool model experimental device. The sensitivity of various surge influence parameters to various surge characteristic parameters was analyzed using this data. The experimental surge influence parameters were used as model inputs, and the experimental surge characteristic influence parameters were used as model outputs. The landslide surge disaster prediction model was trained using the experimental data to obtain a landslide surge disaster prediction model under experimental conditions. A certain amount of engineering case data was acquired, and the actual surge influence parameters from these engineering cases were substituted into the landslide surge disaster prediction model under experimental conditions to obtain the predicted surge characteristic parameters for these engineering cases. The prediction accuracy of the landslide surge disaster prediction model was analyzed based on the predicted and actual surge characteristic parameters. A model optimization strategy based on the sensitivity of surge influence parameters was used to adjust the network parameters of the landslide surge disaster prediction model under experimental conditions, continuously optimizing the landslide surge disaster prediction model, and finally obtaining a landslide surge disaster prediction model that conforms to actual engineering projects. A landslide surge disaster prediction model was trained using landslide surge test data. Then, the prediction effect of the landslide surge disaster prediction model was verified and optimized using actual engineering data. This solved the problem of insufficient engineering data, made the landslide surge disaster prediction model more consistent with engineering reality, and improved the model's prediction effect. Attached Figure Description

[0032] Figure 1 This is a flowchart of the digital modeling method of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0034] It should be noted that the surge wave influence parameters studied in the digital modeling method for landslide surge wave disaster prediction of this invention include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, landslide incident angle, etc., wherein the landslide impact velocity is calculated from the landslide drop height and the time of fall into the water; the surge wave characteristic parameters studied include maximum standing wave height, propagation velocity, and propagation distance of maximum standing wave height (specifically, the distance between the location where the maximum standing wave height occurs and the landslide location), etc.

[0035] The invention of the digital modeling method disclosed in this invention is as follows: a large amount of landslide surge test data is obtained using a wave pool model test device; the sensitivity of each surge influence parameter to each surge characteristic parameter is analyzed using the landslide surge test data; the test surge influence parameters are used as model inputs, and the test surge characteristic influence parameters are used as model outputs; the landslide surge disaster prediction model is trained using the landslide surge test data to obtain the landslide surge disaster prediction model under test conditions.

[0036] A certain amount of engineering case data is obtained, and the actual surge impact parameters in the engineering cases are substituted into the landslide surge disaster prediction model under test conditions to obtain the predicted surge characteristic parameters of the engineering cases. Based on the predicted surge characteristic parameters and the actual surge characteristic parameters, the prediction accuracy of the landslide surge disaster prediction model is analyzed. Based on the model optimization strategy of surge impact parameter sensitivity, the network parameters of the landslide surge disaster prediction model under test conditions are adjusted, and the landslide surge disaster prediction model is continuously optimized to finally obtain a landslide surge disaster prediction model that conforms to actual engineering.

[0037] A landslide surge disaster prediction model was trained using landslide surge test data. Then, the prediction effect of the landslide surge disaster prediction model was verified and optimized using actual engineering data. This solved the problem of insufficient engineering data, made the landslide surge disaster prediction model more consistent with engineering reality, and improved the model's prediction effect.

[0038] like Figure 1 As shown, the present invention discloses the specific steps of a digital modeling method for predicting landslide surge disasters, as follows:

[0039] S1. Determine the surge influence parameters and surge characteristic parameters. The surge influence parameters include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The surge characteristic parameters include maximum standing wave height, propagation velocity, and propagation distance at maximum standing wave height. Construct a wave pool model test device and conduct experiments using the wave pool model test device. Construct a dataset P of experimental surge influence parameters and experimental surge characteristic parameters. Study the surge characteristic parameters under different surge influence parameters and analyze the sensitivity of the surge influence parameters to the surge characteristic parameters.

[0040] S2. Using surge wave impact parameters as model input and surge wave characteristic parameters as model output, a landslide surge wave disaster prediction model is constructed using a neural network. Simulated landslide surge wave test data is used as the training set to train the landslide surge wave disaster prediction model and obtain the landslide surge wave disaster prediction model under test conditions.

[0041] S3. Collect as many landslide surge case studies as possible to construct a dataset P containing actual surge impact parameters and actual surge characteristic parameters. zSubstituting the actual surge impact parameters of each engineering case into the landslide surge disaster prediction model under experimental conditions, the predicted surge characteristic parameters of that engineering case are obtained, and a dataset P of actual surge impact parameters and predicted surge characteristic parameters is constructed. c .

[0042] Specifically, P z =(p z1 ,p z2 …p zi …p zm ), p zi =(X zi ,Y zi ), i = 1, 2…m, where m is the number of engineering cases; X zi =(x z1 ,x z2 ,x z3 ,x z4 ,x z5 ,x z6 ), X zi For the i-th actual surge influence parameter group, x z1 x represents the actual still water depth. z2 x represents the actual landslide dip angle. z3 x represents the actual landslide size. z4 x represents the actual landslide density. z5 x represents the actual landslide impact velocity. z6 Y is the actual incident angle of the landslide; zi =(y z1 ,y z2 ,y z3 ), Y zi For the i-th predicted surge characteristic parameter set, y z1 To predict the maximum standing wave height, y z2 To predict the propagation speed, y z3 To predict the maximum vertical wave height and propagation distance.

[0043] Specifically, P c =(p c1 ,p c2 …p ci …p cm ), p ci =(X zi ,Y ci ), i = 1, 2…m, where m is the number of engineering cases; X zi =(x z1 ,x z2 ,x z3 ,x z4 ,x z5 ,x z6 ), X ziFor the i-th actual surge influence parameter group, x z1 x represents the actual still water depth. z2 x represents the actual landslide dip angle. z3 x represents the actual landslide size. z4 x represents the actual landslide density. z5 x represents the actual landslide impact velocity. z6 Y is the actual incident angle of the landslide; ci =(y c1 ,y c2 ,y c3 ), Y ci For the i-th predicted surge characteristic parameter set, y c1 To predict the maximum standing wave height, y c2 To predict the propagation speed, y c3 To predict the maximum vertical wave height and propagation distance.

[0044] S4. Based on the predicted and actual surge characteristic parameters, create visualization charts to identify outliers in the landslide surge disaster prediction model. Analyze the surge impact parameters based on these outliers, and adjust the network parameters of the landslide surge disaster prediction model under experimental conditions using a model optimization strategy based on the sensitivity of surge impact parameters. Continuously optimize the landslide surge disaster prediction model until the error between the predicted and actual surge characteristic parameters in the engineering project is lower than a set error threshold. Finally, obtain a landslide surge disaster prediction model that conforms to the actual engineering project. Outliers refer to situations where the error between the predicted and actual surge characteristic parameters is greater than or equal to a set error threshold, which can be set to 5-10%.

[0045] In a further implementation, step S1 includes the following steps:

[0046] S11. Constructing a wave pool model test device: Water from the Three Gorges Reservoir area is used as the water source for the wave pool model test device, and the surrounding slopes of the Three Gorges Reservoir area are used as the slopes. Multiple high-speed cameras are set up beside the pool to simultaneously capture the surging waves caused by the slope sliding down. Timecode or optical synchronization signals are added to the captured video to ensure that the data from multiple cameras are time-aligned. The high-speed cameras perform multi-view synchronous shooting at a sampling frequency of ≥1000 frames per second, which can accurately record the wave impact parameters and wave characteristic parameters. The shutter speed of the high-speed cameras is ≤1 / 10000s to avoid motion blur. The aperture and ISO are adjusted according to the light source intensity to ensure image clarity.

[0047] S12. Determine the parameters affecting the surge and the parameters characteristic of the surge. The parameters affecting the surge include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The parameters characteristic of the surge include maximum standing wave height, propagation velocity, and propagation distance at maximum standing wave height.

[0048] S13. For each surge influence parameter, three levels—small (low), medium, and large (high)—are set. Combinations between these levels are designed, and simulation experiments are conducted using a wave pool model test setup to obtain a dataset P of experimental surge influence parameters and experimental surge characteristic parameters. It should be noted that each level corresponds to a specific range of values, determined based on previous research experience.

[0049] Specifically, P = (p1, p2, ..., p) i …p n ), p i =(X i ,Y i ), i = 1, 2…n, where n is the number of test landslide surge data sets; X i =(x1,x2,x3,x4,x5,x6), X i For the i-th test surge influence parameter set, x1 is the test still water depth, x2 is the test landslide dip angle, x3 is the test landslide size, x4 is the test landslide density, x5 is the test landslide impact velocity, and x6 is the test landslide incident angle; Y i =(y1,y2,y3), Y i Let y1 be the test surge characteristic parameter set, y2 be the test propagation velocity, and y3 be the propagation distance of the test maximum standing wave height.

[0050] S14. Use analysis of variance to determine the contribution rate of surge influence parameters and their interactions to the changes in surge characteristic parameters. By calculating statistics such as variance ratio (F-value), determine whether each surge influence parameter and interaction has a significant impact on the changes in surge characteristic parameters. For example, if the F-value of a certain surge influence parameter is large and the corresponding P-value is less than the set significance level (such as 0.05), it indicates that the surge influence parameter has a significant impact on the surge characteristic parameters. Its contribution rate can be obtained by calculating the proportion of the sum of squares of the surge influence parameter to the total sum of squares.

[0051] S15. Based on the analysis results of step S14, construct the sensitivity matrix of surge influence parameters and surge characteristic parameters, and for each surge characteristic parameter, sort the 6 surge influence parameters in descending order of their contribution rate to obtain the sensitivity dataset R.

[0052] Specifically, R = (R1, R2, R3), where R1 is the set of surge influence parameters sensitivity for maximum standing wave height, R2 is the set of surge influence parameters sensitivity for propagation speed, and R3 is the set of surge influence parameters sensitivity for propagation distance from maximum standing wave height.

[0053] Based on their sensitivity, the surge influence parameters of each type of surge characteristic parameter are divided into high-sensitivity influence parameters and low-sensitivity influence parameters. The high and low sensitivity influence parameters can be specifically set as needed. For example, surge influence parameters with a sensitivity greater than 0.5 can be defined as high-sensitivity influence parameters, and those with a sensitivity less than or equal to 0.5 can be defined as low-sensitivity influence parameters. Alternatively, in this application, there are six surge influence parameters, and the first three in the R1, R2, and R3 sets are defined as high-sensitivity influence parameters, and the last three as low-sensitivity influence parameters. Other definition methods are also possible.

[0054] In a preferred embodiment, since the interactions among the surge influence parameters collectively affect the changes in surge characteristic parameters, a multi-factor combination method is used to analyze the sensitivity of the surge influence parameters to surge characteristics. Specifically, all combinations of all levels of all surge influence parameters are considered. With six surge influence parameters, each with three levels, the full factorial design has three possible combinations. 6 =729 combinations. Conducting experiments on all of these combinations would be an enormous undertaking. A preferred implementation method, based on a full factorial design, uses an orthogonal design method to select suitable and representative n combinations for experimentation. The selected n combinations ensure that each level of each surge parameter occurs the same number of times in the combination, and that the interactions between surge parameters are well balanced.

[0055] In a further implementation, in step S2, the experimental landslide surge disaster prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes 6 input neurons, the output layer includes 3 output neurons, and two hidden layers are set. The first hidden layer has 12 neurons, and the second hidden layer has 8 neurons.

[0056] In a further embodiment, step S4 includes the following steps:

[0057] S41. Based on the predicted and actual surge characteristic parameters, create visualization charts to identify anomalies predicted by the landslide surge disaster prediction model. Determine the highly sensitive influencing parameters of the surge characteristic parameters corresponding to these anomalies, and prioritize the analysis and adjustment of the model network parameters for these highly sensitive influencing parameters. Create different charts based on the different surge characteristic parameters, such as charts for maximum standing wave height, propagation velocity, and propagation distance at maximum standing wave height. Visualization charts can include line graphs and bar charts showing the actual and predicted values ​​for each surge characteristic parameter; with the actual value on the horizontal axis and the predicted value on the vertical axis, each point representing an engineering case; calculating the deviation value for each point and marking the magnitude of the deviation with a line segment perpendicular to the horizontal axis; other forms of charts are also possible.

[0058] S42. Traverse all outliers and filter out extreme outliers based on whether the actual surge impact parameters exceed the model training range. Extreme outliers refer to situations where, when comparing the actual surge impact parameters of the corresponding engineering case with the distribution of experimental surge impact parameters in the dataset P (experimental surge impact parameters - experimental surge characteristic parameters), the actual surge impact parameter is greater than the maximum value of the experimental surge impact parameters, or the actual surge impact parameter is less than the minimum value of the experimental surge impact parameters. In other words, determining whether an outlier is an extreme outlier is essentially determining whether the actual surge impact parameters exceed the training range of the experimental surge impact parameters used to train the landslide surge disaster prediction model. Surge impact parameters exceeding the training range are defined as extreme surge impact parameters.

[0059] S43. The strategy for adjusting model network parameters based on sensitivity is as follows:

[0060] If the outlier is an extreme outlier, and the extreme surge impact parameter is a highly sensitive impact parameter corresponding to the surge characteristic parameter, then in the input layer, an extreme mapping neuron is added to this surge impact parameter. This enhances feature extraction of extreme values ​​through a piecewise function, avoiding simple truncation or linear mapping of extreme values, and improving the model's nonlinear response to inputs outside its range. And / or, the initialization method of the surge impact parameter's weights in the hidden layer is adjusted, such as using Xavier initialization instead of the default initialization, to avoid gradient vanishing. The validation metric mainly focuses on the decrease in prediction error under extreme conditions; for example, if the relative error decreases from 50% to within 15%, the optimization is considered complete.

[0061] If the outlier is not an extreme outlier, and the model prediction bias for the highly sensitive parameter of the outlier does not match the actual trend of parameter change (e.g., the predicted value decreases when the actual parameter increases), then a nonlinear transformation layer for the highly sensitive parameter should be added to the hidden layer. For example, the ReLU activation function can be used first, followed by cross-interaction with other parameters to strengthen the nonlinear mapping; and / or, the input features of the highly sensitive parameter can be standardized, such as Z-score standardization, to avoid imbalance in weight updates due to excessively large numerical ranges.

[0062] If the outlier is not an extreme outlier and is caused by a low-sensitivity parameter, the cause of the outlier may be model overfitting. Applying L1 regularization to the weights corresponding to the low-sensitivity parameter can force the network to reduce its dependence on it; and / or, adding perturbation samples of the low-sensitivity parameter to the training data can increase the robustness of the model.

[0063] S44. Re-input the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to generate new predicted surge characteristic parameters. Compare the error changes of the original anomaly points until the requirements are met. At the same time, adjust the sensitivity of the surge impact parameter and update the surge impact parameter sensitivity ranking and sensitivity matrix.

[0064] S45. Substitute the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to obtain new predicted surge characteristic parameters for the engineering case. Draw a visualization chart based on the new predicted surge characteristic parameters and the actual surge characteristic parameters. If anomalies still appear, repeat steps S42-S44.

[0065] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, alterations, substitutions, or variations made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A digital modeling method for predicting landslide surge hazards, characterized in that: Landslide surge test data were obtained using a wave pool model experimental device. The sensitivity of surge influence parameters to surge characteristic parameters was analyzed. A landslide surge disaster prediction model was trained using the landslide surge test data to obtain a landslide surge disaster prediction model under experimental conditions. Actual surge influence parameters from engineering cases were substituted into the landslide surge disaster prediction model under experimental conditions to obtain predicted surge characteristic parameters for engineering cases. The prediction accuracy of the landslide surge disaster prediction model was analyzed based on the predicted surge characteristic parameters and actual surge characteristic parameters. The network parameters of the landslide surge disaster prediction model under experimental conditions were adjusted based on the model optimization strategy of surge influence parameter sensitivity. The landslide surge disaster prediction model was continuously optimized, and finally a landslide surge disaster prediction model that conforms to actual engineering was obtained. Includes the following steps: S1. Determine the swell influence parameters and swell characteristic parameters, build a wave pool model test device, conduct experiments using the wave pool model test device, study the swell characteristic parameters under different swell influence parameters, and analyze the sensitivity of the swell influence parameters to the swell characteristics. S2. Using the surge wave impact parameters as model input and the surge wave characteristic parameters as model output, a landslide surge wave disaster prediction model is constructed using a neural network. The simulated landslide surge wave test data is used as the training set to train the landslide surge wave disaster prediction model and obtain the landslide surge wave disaster prediction model under test conditions. S3. Collect engineering cases of landslide surges, substitute the actual surge impact parameters of the engineering cases into the landslide surge disaster prediction model under test conditions, and obtain the predicted surge characteristic parameters of the engineering cases. S4. Draw visualization charts based on the predicted surge characteristic parameters and the actual surge characteristic parameters, identify the anomalies predicted by the landslide surge disaster prediction model, analyze the surge influence parameters based on the anomalies, adjust the network parameters of the landslide surge disaster prediction model under the test conditions based on the model optimization strategy of surge influence parameter sensitivity, continuously optimize the landslide surge disaster prediction model, and finally obtain a landslide surge disaster prediction model that conforms to the actual engineering.

2. The digital modeling method for landslide surge disaster prediction according to claim 1, characterized in that: The landslide surge disaster prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes 6 input neurons, the output layer includes 3 output neurons, and two hidden layers are set. The first hidden layer has 12 neurons, and the second hidden layer has 8 neurons. The surge influence parameters are used as the model input, and the surge characteristic parameters are used as the model output. The surge influence parameters include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The surge characteristic parameters include maximum standing wave height, propagation velocity, and maximum standing wave height propagation distance.

3. The digital modeling method for landslide surge disaster prediction according to claim 2, characterized in that: Step S1 includes the following process: S11. Construct a wave pool model test device; S12. Determine the parameters affecting the surge and the parameters characteristic of the surge. The parameters affecting the surge include still water depth, landslide dip angle, landslide size, landslide density, landslide impact velocity, and landslide incident angle. The parameters characteristic of the surge include maximum standing wave height, propagation velocity, and propagation distance at maximum standing wave height. S13. Set three levels for each type of surge influence parameter: small, medium, and large. Design the combination of each level of surge influence parameter. Use the wave pool model test device to conduct simulation tests on these combinations and obtain the dataset P of test surge influence parameters - test surge characteristic parameters. S14. Use the analysis of variance method to determine the contribution rate of the interaction between the surge influencing parameters to the changes in the surge characteristic parameter results; S15. Construct a sensitivity matrix of surge influence parameters and surge characteristic parameters, and for each surge characteristic parameter, sort the six surge influence parameters in descending order of their contribution rate.

4. The digital modeling method for landslide surge disaster prediction according to claim 3, characterized in that: In step S1, the sensitivity of surge influence parameters to surge characteristics is analyzed using a multi-factor combination method. Based on the full factorial design, an orthogonal design method is used to select appropriate and representative n combinations for experimentation. The selected n combinations ensure that each level of each surge influence parameter appears the same number of times in the combination and can balance the interaction between surge influence parameters.

5. A digital modeling method for landslide surge disaster prediction according to claim 2, characterized in that: Step S4 includes the following process: S41. Draw a visualization chart based on the predicted surge characteristic parameters and the actual surge characteristic parameters to determine the anomalies predicted by the landslide surge disaster prediction model; S42. Traverse all outliers and filter out extreme outliers based on whether the actual surge impact parameters exceed the model training range. S43. Depending on whether the outlier is an extreme outlier, different optimization strategies are adopted to adjust the model network parameters, resulting in the adjusted landslide surge disaster prediction model. S44. Re-input the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to generate new predicted surge characteristic parameters. Compare the error changes of the original anomaly points until the requirements are met. At the same time, adjust the sensitivity of the surge impact parameter and update the surge impact parameter sensitivity ranking and sensitivity matrix. S45. Substitute the actual surge impact parameters from the engineering case into the adjusted landslide surge disaster prediction model to obtain new predicted surge characteristic parameters for the engineering case. Draw a visualization chart based on the new predicted surge characteristic parameters and the actual surge characteristic parameters. If anomalies still appear, repeat steps S42-S44.

6. The digital modeling method for landslide surge disaster prediction according to claim 5, characterized in that: In step S43, if the anomaly is an extreme anomaly and the extreme surge influence parameter is a highly sensitive influence parameter corresponding to the surge characteristic parameter, then in the input layer, an extreme mapping neuron is added to the surge influence parameter to enhance the feature extraction of extreme values ​​through a piecewise function. And / or, adjust the weight initialization method of the surge effect parameter in the hidden layer.

7. A digital modeling method for landslide surge disaster prediction according to claim 5, characterized in that: In step S43, if the outlier is not an extreme outlier, and if the model prediction deviation does not match the actual trend of the parameter's change, a nonlinear transformation layer for the high-sensitivity parameter is added in the hidden layer to strengthen the nonlinear mapping; and / or, the input features of the high-sensitivity parameter are standardized.

8. A digital modeling method for landslide surge disaster prediction according to claim 5, characterized in that: In step S43, if the outlier is not an extreme outlier and is caused by a low-sensitivity influence parameter, then L1 regularization is applied to the weight corresponding to the low-sensitivity influence parameter to force the network to reduce its dependence on it; and / or, perturbation samples of the low-sensitivity influence parameter are added to the training data.

Images