Method and system for predicting water content of volcanic ash soil slope

Abstract

The invention provides a volcanic ash soil side slope water content prediction method and system, and relates to the technical field of side slope water content prediction.The method comprises the steps that first information and second information are obtained, the first information comprises design data and historical monitoring data of a volcanic ash soil side slope, and the second information comprises the design data and the historical monitoring data of the volcanic ash soil side slope; the second information comprises real-time monitoring data of the volcanic ash soil slope; constructing a moisture migration rule model based on a Richard equation in a switching form, sending the first information to the moisture migration rule model for feature extraction, and classifying the extracted features to obtain classified first information; inputting the classified first information and the moisture migration rule model into a preset neural network model for training to obtain a trained moisture content prediction model; the second information is sent to the trained water content prediction model, and volcanic ash soil slope water content parameters are obtained. According to the method, the water content prediction precision and the operation speed are improved, and the performance loss and the time cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of slope water content prediction, and in particular, to a method and system for predicting the water content of volcanic ash soil slopes. Background Art

[0002] Due to its special geological and geomorphic features and climatic conditions, China is one of the countries in the world affected by geological disasters such as debris flows and landslides. A large number of studies and engineering practices have shown that water has a crucial impact on slope stability. Its disaster-causing mechanism is mainly reflected in that the change in water content causes an increase in soil weight, an increase in pore water pressure, and a decrease in soil shear strength, thereby inducing slope instability. How to scientifically and accurately predict the water content distribution of slopes has become an important topic in the research of geological disaster prevention and control. However, traditional rainfall infiltration description models are prone to numerical instability and calculation deviations under complex unsaturated-saturated conditions.

[0003] Traditional calculation methods usually rely on a large number of parameter inputs, including soil physical properties, rainfall intensity, and slope geometric characteristics, etc. The complex coupling relationship between these parameters significantly increases the manual operation cost and monitoring difficulty. In addition, the selection of parameters in traditional models relies more on manual experience, and the training process is slow, making it difficult to meet the real-time requirements of water content prediction under dynamic conditions, resulting in a lag in the response to slope instability risks and increasing the probability of disasters occurring.

[0004] There is an urgent need for a method and system for predicting the water content of volcanic ash soil slopes to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting the water content of volcanic ash soil slopes to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0006] In a first aspect, the present application provides a method for predicting the water content of volcanic ash soil slopes, including:

[0007] Obtain first information and second information, where the first information includes design data and historical monitoring data of the volcanic ash soil slope, and the second information includes real-time monitoring data of the volcanic ash soil slope;

[0008] Construct a water migration law model based on the switched form of the Richard equation, send the first information to the water migration law model for feature extraction, and classify the extracted features to obtain the classified first information;

[0009] Input the classified first information and the water migration law model into a preset neural network model for training to obtain a trained water content prediction model;

[0010] Send the second information to the trained water content prediction model to obtain the water content parameters of the volcanic ash soil slope.

[0011] In a second aspect, the present application also provides a water content prediction system for a volcanic ash soil slope, including:

[0012] An acquisition unit for acquiring first information and second information, where the first information includes design data and historical monitoring data of the volcanic ash soil slope, and the second information includes real-time monitoring data of the volcanic ash soil slope;

[0013] A classification unit for constructing a water migration law model based on the switched form of the Richard equation, sending the first information to the water migration law model for feature extraction, and classifying the extracted features to obtain the classified first information;

[0014] A training unit for inputting the classified first information and the water migration law model into a preset neural network model for training to obtain a trained water content prediction model;

[0015] A prediction unit for sending the second information to the trained water content prediction model to obtain the water content parameters of the volcanic ash soil slope.

[0016] The beneficial effects of the present invention are as follows:

[0017] In the improved water content prediction system for the volcanic ash soil slope of the present invention, by combining a physical model with a transfer learning neural network, the accuracy and operation speed of water content prediction are greatly improved, and the performance loss and time cost are reduced. The system can adapt to the complex geological conditions and dynamic environmental changes of volcanic ash soil, has strong fault tolerance and self-adaptive capabilities, and significantly improves the accuracy of monitoring and calculation. The present invention can also automatically adjust parameters according to user requirements, quickly generate prediction schemes that meet the requirements, reduce the labor cost of monitoring and calculation, and provide a reliable guarantee for slope stability analysis and disaster warning.

[0018] Other features and advantages of the present invention will be described in the subsequent specification, and part of them will become obvious from the specification, or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the structures specifically pointed out in the written specification, claims, and drawings. Description of the Drawings

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flow chart of the method for predicting the water content of volcanic ash soil slopes described in the embodiments of the present invention;

[0021] Figure 2 It is a schematic structural diagram of the system for predicting the water content of volcanic ash soil slopes described in the embodiments of the present invention.

[0022] In the figure: 701, acquisition unit; 702, classification unit; 703, training unit; 704, prediction unit. Specific embodiments

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but only represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0024] It should be noted that: similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0025] Embodiment 1:

[0026] This embodiment provides a method for predicting the water content of volcanic ash soil slopes.

[0027] See Figure 1 , the figure shows that this method includes steps S1, S2, S3, and S4.

[0028] Step S1: Obtain the first information and the second information. The first information includes the design data and historical monitoring data of the volcanic ash soil slope, and the second information includes the real-time monitoring data of the volcanic ash soil slope;

[0029] It can be understood that in this step, the design data of the volcanic ash soil slope includes basic information such as the geometric dimensions of the slope (such as slope, height, length), slope type (such as natural slope, artificially reinforced slope), engineering geological conditions of the slope (such as soil layer structure, geotechnical characteristics), rainfall infiltration parameters of the slope (such as infiltration coefficient, drainage conditions), as well as load requirements and slope construction costs. These data provide the basic parameters for constructing the physical model and are used to describe the dynamic moisture content change characteristics of the volcanic ash soil.

[0030] The historical monitoring data are the long-term monitoring data of the slope recorded by sensors, including the historical moisture content, pore water pressure, dynamic surface and groundwater levels, rainfall infiltration depth, and shear strength parameters of the slip zone soil (such as liquid limit, plastic limit, cohesion, and internal friction angle) at different monitoring points. These data show the moisture migration characteristics of the slope at different times and under different environmental conditions through time series and are used to verify and optimize the prediction accuracy of the physical model.

[0031] The real-time monitoring data, on the other hand, dynamically record the real-time change parameters during the operation stage of the slope, including environmental dynamic parameters such as the real-time moisture content distribution, pore water pressure field distribution, rainfall intensity, and soil temperature and humidity changes. These data will be used as the input for real-time model updates so that the prediction results can reflect the current true state of the slope.

[0032] By obtaining comprehensive and accurate design data, historical monitoring data, and real-time monitoring data, this step provides the necessary basis for the subsequent construction of the physical model based on the Richard equation, the optimization of the transfer learning neural network, and the prediction of the real-time moisture content. This method combining data-driven and physical laws can significantly improve the scientificity and reliability of model prediction and provide efficient support for slope disaster warning and management.

[0033] Step S2: Construct a moisture migration law model based on the switched form of the Richard equation, send the first information to the moisture migration law model for feature extraction, and classify the extracted features to obtain the classified first information;

[0034] It can be understood that in this step, Step S2 includes Step S21, Step S22, Step S23, and Step S24.

[0035] Step S21: Calculate the comprehensive saturation index based on a preset comprehensive saturation index calculation formula;

[0036] It can be understood that in order to achieve the dynamic switching between the saturated zone and the unsaturated zone in volcanic ash soil, the present invention proposes a calculation formula based on a comprehensive saturation index. The formula is as follows:

[0037]

[0038] Where P represents the comprehensive saturation index, and θ represents the current volumetric water content. θ s and θ r are the saturated water content and the residual water content respectively. represents the effective saturation, represents the pressure head normalization value, which describes the dynamic water pressure in the soil.

[0039] Step S22: Compare based on the comprehensive saturation index and the preset critical saturation threshold of the soil. If the comprehensive saturation index is greater than or equal to the preset critical saturation threshold of the soil, the Richard equation in the form of pressure head is used for formula construction to obtain the Richard equation in the form of pressure head;

[0040] It can be understood that according to the value of the effective saturation, the present invention sets the critical saturation threshold P crit of the soil, and divides the soil into a saturated zone and an unsaturated zone. When P ≥ P crit , the soil is in a saturated state, and the Richard equation in the form of pressure head is used for calculation.

[0041] Among them, the Richard equation in the form of pressure head is as follows:

[0042]

[0043] Where θ represents the volumetric water content. h represents the pressure head. D(θ) represents the water diffusivity. K(θ) represents the hydraulic conductivity. C(h) represents the soil water storage. t represents time. z represents the spatial position.

[0044] Step S23: If the comprehensive saturation index is less than the preset critical saturation threshold of the soil, the Richard equation in the form of water content is used for formula construction to obtain the Richard equation in the form of water content;

[0045] It can be understood that when P < P crit , the soil is in the unsaturated zone, and the Richard equation in the form of water content is used.

[0046] Among them, the Richard equation in the form of water content is as follows:

[0047]

[0048] Among them, θ represents the volumetric water content. h represents the pressure head. D(θ) represents the water diffusivity. K(θ) represents the hydraulic conductivity. C(h) represents the soil water storage. t represents time. z represents the spatial position.

[0049] Step S24: Use the constructed Richard equation in the form of pressure head and the Richard equation in the form of water content as the water migration law model.

[0050] It can be understood that the physical model based on the Richard equation provides high-quality physical prior information for the prediction of subsequent models by accurately describing the dynamic water migration law of volcanic ash soil, and maintains high computational accuracy and stability under complex geological conditions. This method combining physics and numerical solution lays a scientific foundation for the prediction of water content in volcanic ash soil slopes.

[0051] It can be understood that in this step, after step S24, there are also step S25, step S26, step S27, step S28, step S29, and step S210.

[0052] Step S25: Send the first information to the water migration law model for feature extraction to obtain the feature data corresponding to the first information.

[0053] It can be understood that the design data includes but is not limited to the geometric design parameters of the slope, such as slope gradient, slope height, soil layer thickness, soil type, slope surface material, load conditions, etc. The design data provides the basic structural information of the slope for the preliminary analysis of the model. Specifically, it includes slope gradient, slope height, soil layer thickness, and load conditions.

[0054] Historical monitoring data: It involves the historical slope stability monitoring data, mainly including the following contents.

[0055] Displacement monitoring data: The displacement data on the surface or inside of the slope collected by instruments.

[0056] Stress and strain data: The stress and strain data of the slope under different loading conditions.

[0057] Hydrometeorological data: It includes environmental factors such as rainfall, groundwater level, air temperature, etc., which are closely related to slope stability.

[0058] Deformation and failure mode data: The deformation or failure events that have occurred on the slope historically, and the corresponding slope response data.

[0059] Physical model feature data: The feature data obtained through physical experiments or numerical simulations according to the previously established equations under the switching mechanism of saturated soil and unsaturated soil, specifically including:

[0060] Mechanical properties of soil: such as friction angle, cohesion, elastic modulus, Poisson's ratio, etc.

[0061] Permeability data of soil: including the permeability coefficient of water flow through the soil.

[0062] Switching mechanism between saturated and unsaturated soils: Describing the different mechanical responses and hydrological behaviors of soils in saturated and unsaturated states through the established equations.

[0063] Distribution and contact characteristics of soil layers: The relative distribution of different soil layers and the properties of their contact surfaces. Inputting the above data into the water migration law model for feature extraction to obtain the feature data of the first information, which improves the classification accuracy and data utilization efficiency.

[0064] Step S26: Construct a hierarchical structure model from the first information and the feature data corresponding to the first information to obtain a hierarchical structure model.

[0065] It can be understood that the hierarchical structure model is defined as follows:

[0066] H = {L 1 , L 2 , …, L n}

[0067] Among them, H represents the hierarchical structure model, which is used to organize the hierarchical data structure of the volcanic ash soil slope. L i represents the data set of the i-th layer, indicating the relevant data information at this level, such as design parameters, physical model extraction features, etc. n represents the total number of layers of the hierarchical structure, reflecting the multi-level division depth of the data.

[0068] The data classification hierarchy is defined as follows:

[0069] L i = {(x j , y j ) | x j ∈ X, y j ∈ Y}

[0070] Among them, x j represents the feature vector of the slope (such as design data, physical characteristics, historical monitoring values). y j represents the classification label (such as slope category or moisture content level). X represents the feature space, which contains all possible data features. Y represents the label space, which contains all possible category labels.

[0071] It can be understood that through the establishment of the hierarchical structure model, the engineering team can analyze the slope data from the whole to the part. Using physical model features (such as pore water pressure field and moisture content distribution) to further refine the classification levels provides a logical framework for subsequent data clustering and redundancy removal.

[0072] Step S27: Classify the first information and the feature data corresponding to the first information based on the hierarchical structure model to obtain data information of at least two categories;

[0073] It can be understood that based on the hierarchical structure model, clustering analysis is performed on the design data, historical monitoring data, and physical model output features of the volcanic ash soil slope to obtain data information of at least three categories;

[0074] The clustering objective function is as follows:

[0075]

[0076] Among them, J represents the value of the clustering objective function, which is used to measure the optimization degree of the current clustering result. k represents the number of clustering categories, which is preset as the number of different slope characteristics. x j represents the feature vector of the j-th data point belonging to the i-th category. C i represents the data set of the i-th category, which contains all data points belonging to this category. μ i represents the centroid (center point) of the data set of the i-th category, indicating the feature center of this category. ∥x j -μ i ∥ 2 represents the square of the Euclidean distance between the j-th data point and the center point of the i-th category.

[0077] It can be understood that through the clustering analysis method, the complex data of the slope is systematically divided into multiple categories, reflecting the slope states under different working conditions, geological conditions, and hydraulic characteristics. The key parameters extracted by the physical model (such as water content distribution, rainfall infiltration depth, etc.) are used as the key indicators for clustering, which helps to more accurately identify the characteristics of different categories of slopes. This classification method not only improves the training efficiency of the model but also enhances the applicability of the prediction model to various slope conditions.

[0078] Step S28: Analyze the data information of the same category in pairs to determine whether one data is a subset of the other. If so, retain the subset data and delete the redundant data; if not, retain both groups of data to obtain the underlying data of each category;

[0079] It can be understood that the subset determination formula is as follows:

[0080]

[0081] Otherwise:

[0082] A∪B={a 1 ,a 2 ,…,b n}

[0083] Among them, A represents a certain subset in the data set. B represents another subset in the data set. Indicates whether set A is a subset of set B.

[0084] It can be understood that in this step, through refined redundancy removal processing, the redundancy and repetition in the data set are effectively reduced, while the most representative key data in the category is retained. The high-dimensional feature data generated by the physical model (such as pore water pressure gradient or moisture migration rate) plays an important role in the redundancy removal process, helping to screen out more effective training data. This method of streamlining and optimizing data significantly improves the computational efficiency and prediction accuracy of the neural network model.

[0085] Step S29: Repeatedly analyze the deleted redundant data in pairs, and divide the data through hierarchical optimization until all data cannot be further divided hierarchically, obtaining the hierarchical data information of each category;

[0086] It can be understood that the hierarchical optimization formula is as follows:

[0087]

[0088] Among them, represents the i-th hierarchical data set after optimization. \ represents the set difference operation, indicating removing the data points that can be merged from the current level. x j represents the redundant data points that can be merged.

[0089] It can be understood that in this step, through recursive optimization, the structure and hierarchical division of the data are further refined. Based on the dynamic features output by the physical model (such as the switching state between the unsaturated zone and the saturated zone), the logical relationships of each data are gradually adjusted to ensure that the data in the final hierarchical structure has high independence and representativeness. This method can improve the data utilization efficiency and provide a clear logical basis for the training of the transfer learning neural network.

[0090] Step S210: Perform a union operation on the hierarchical data information of all categories to obtain the data information of all categories.

[0091] It can be understood that the data classification formula is as follows:

[0092]

[0093] Among them, C k represents the complete data set of the k-th category. represents the i-th hierarchical data belonging to the k-th category. ∪ represents the union operation, indicating integrating the same-category data in different levels into a complete set.

[0094] It can be understood that through classification and structured organization, this step organizes data of different categories and levels into a clear hierarchical structure. This way of processing structured data helps to improve the visualization and manageability of data, and at the same time enhances the parsing ability of the transfer learning neural network for input data, providing efficient data support for the construction and deployment of the water content prediction model.

[0095] Step S3: Input the classified first information and the moisture migration law model into a preset neural network model for training to obtain a trained water content prediction model;

[0096] It can be understood that step S3 includes step S31, step S32, step S33, and step S34.

[0097] Step S31: Divide the first information and the feature data corresponding to the first information into a training set and a test set;

[0098] It can be understood that in this step, by reasonably dividing the training set and the test set, it is ensured that the training and testing processes of the model are representative, providing high-quality data support for subsequent optimization. The data set division formula is as follows:

[0099]

[0100] Among them, D represents the complete data set, which contains slope data information of all categories. D train represents the training set, which is used for model optimization. D test represents the test set, which is used for model verification.

[0101] Step S32: Input the training set into a preset neural network model for preliminary training, where the training set is preliminarily trained based on a preset neural network prediction formula and a loss function to obtain a preliminary prediction model;

[0102] It can be understood that in this step, through the preliminary fitting of the training set data, the basic prediction ability of the neural network is constructed, and initial parameters are provided for subsequent transfer learning optimization. Among them, the neural network prediction formula is as follows:

[0103]

[0104] Among them, represents the predicted value of the i-th sample. x i represents the feature vector of the i-th sample, including design data and physical model output features. f represents the neural network function. W represents the weight matrix of the neural network. B represents the bias vector of the neural network.

[0105] Preliminary training loss function:

[0106]

[0107] Among them, L data represents the data-driven loss function, which is used to measure the error between the predicted value and the true value. y i represents the true value of the i-th sample. N represents the total number of training samples.

[0108] Step S33: Optimize the parameters of the preliminary prediction model based on the preset transfer learning loss function and the preset physical consistency loss calculation formula to obtain a preliminarily optimized prediction model, where the parameters include weight parameters and bias parameters;

[0109] It can be understood that the weight parameters and bias parameters of the neural network are optimized through the loss function of transfer learning and the physical constraints of the physical model. In this step, the prediction accuracy and physical interpretability of the neural network are improved through the gradient descent method combined with physical consistency constraints. In this step, step S33 includes step S331, step S332, and step S333.

[0110] Step S331: Weight based on the preset data-driven loss calculation formula and the physical consistency loss formula to obtain a transfer learning loss function:

[0111] It can be understood that the transfer learning loss function is as follows:

[0112] L = α·L data + β·L physics

[0113] Among them, L represents the total loss function. L data represents the data-driven loss, which is used to fit the training set. L physics represents the physical consistency loss, which is used to constrain the model output to conform to physical laws. α and β represent weight coefficients, indicating the relative importance of data-driven and physical-driven. The weight coefficients α = 0.7 and β = 0.3 are set to highlight the priority of data fitting while maintaining the importance of physical constraints.

[0114] Among them, L data The calculation formula is as follows:

[0115]

[0116] Among them, y i represents the true value of the i-th sample. represents the predicted value of the neural network. N represents the number of samples in the training set.

[0117] Among them, the physical consistency loss L physics The calculation formula is as follows:

[0118]

[0119] Among them, θ represents the water content. K(θ) represents the permeability coefficient, which is a non-linear function dependent on the water content. h represents the pore water pressure. represents the time derivative of the water content. represents the spatial migration term of water. N represents the number of all sample points used in the calculation.

[0120] Step S332: Iteratively optimize the neural network parameters through the gradient descent method combined with the transfer learning loss function to obtain the iteratively optimized neural network model. Among them, the iterative optimization includes adjusting the weights and the bias parameters.

[0121] It can be understood that in this sub-step, the neural network parameters are iteratively optimized through the gradient descent method combined with the transfer learning loss function, and the weights and bias parameters are adjusted.

[0122] Among them, the calculation formula for the gradient of the loss function is as follows:

[0123]

[0124] Among them, represents the gradient of the loss function with respect to the weight parameter W. represents the gradient of the loss function with respect to the bias parameter b.

[0125] The calculation formula for parameter update is as follows:

[0126]

[0127] Among them, W represents the weight parameter matrix of the neural network. b represents the bias parameter vector of the neural network. η represents the learning rate. represents the gradient of the loss function L with respect to the weight W. represents the gradient of the loss function L with respect to the bias b. L represents the loss function, which measures the error between the prediction result and the true value.

[0128] The dynamic adjustment formula for η is as follows:

[0129] η t = η 0 ·e -k·t

[0130] Among them, η 0 represents the initial learning rate (such as 0.01). k represents the learning rate decay factor. t represents the current iteration number.

[0131] During the optimization process, a relatively small learning rate is initially adopted to focus on local optimization for refining model parameters. As the optimization progresses, the learning rate is gradually increased to enhance the global search ability, thereby avoiding the model falling into the dilemma of local optimal solutions. The stopping conditions for optimization include two aspects: one is reaching the maximum number of iterations T max (e.g., 1000 times); the other is that the loss function L converges to a preset threshold. Finally, what the optimization process outputs are the adjusted weight parameters to ensure that the neural network reaches the optimal state.

[0132] Step S333: Send the training set data to the iteratively optimized neural network model for retraining to obtain a preliminarily optimized prediction model.

[0133] It can be understood that the retraining loss function:

[0134] L final =α·L data +β·L physics +γ·L reg

[0135] where L final represents the final loss function value. α represents the weight coefficient of the data-driven loss. L data represents the data-driven loss, which is used to measure the error between the model prediction value and the true value. β represents the weight coefficient of the physical consistency loss. L physics represents the physical consistency loss. γ represents the weight coefficient of the regularization loss. L reg represents the regularization loss, which is used to constrain the magnitude of the model parameters to prevent overfitting.

[0136] L reg Calculation formula:

[0137] L reg =∥W∥ 2

[0138] where L reg represents the regularization loss, which is used to constrain the magnitude of the model parameters to prevent overfitting. W represents the weight matrix of the neural network.

[0139] It can be understood that in this step, the retraining process makes the model's prediction for complex slope conditions more robust and further optimizes its generalization ability.

[0140] Step S34: Send the test set to the preliminary prediction model for verification to obtain a trained water content prediction model.

[0141] It can be understood that the test set is input into the optimized neural network to verify the model performance. The verification metrics include the mean square error and the physical consistency residual.

[0142] Among them, the calculation formula of the mean square error MSE is as follows:

[0143]

[0144] Among them, M represents the total number of samples in the test set. y i represents the true value of the i-th sample. represents the predicted value of the i-th sample.

[0145] Among them, the calculation formula of the physical consistency residual PCR is as follows:

[0146]

[0147] Among them, M represents the total number of samples in the test set. represents the time change rate of the water content θ, describing the dynamic change of water with time. represents the divergence of the spatial change of water. represents the pore water pressure gradient. K(θ) represents the pore water pressure gradient. represents the residual of the Richard equation, reflecting whether the model output satisfies the physical law.

[0148] It can be understood that after this step, through the real-time data input, dynamic prediction and feedback adjustment mechanism, the optimized neural network model can adapt to the dynamic changes of the volcanic ash soil slope under complex working conditions. By combining the physical model constraints and transfer learning methods, the water content distribution and change trend of the volcanic ash soil slope are generated in real time, including key parameters such as the dynamic boundary position of the unsaturated zone and the saturated zone, and the rainfall infiltration depth. The feedback mechanism further ensures the accuracy and reliability of the prediction results. Even when the monitoring data changes significantly or the slope working conditions suddenly change, the model can still adjust the parameters in time through dynamic optimization, improve its robustness and adaptability, and provide a scientific basis for slope engineering safety management and disaster warning. Among them, after step S34, there are also step S35, step S36 and step S37.

[0149] Step S35: Normalize the first information and the second information to obtain the normalized first information and second information;

[0150] Among them, the input data needs to be normalized before entering the model, and the formula is as follows:

[0151]

[0152] Among them, x i represents the original input data. x' iRepresents the normalized data. max(x) represents the maximum value in the dataset. min(x) represents the minimum value in the dataset.

[0153] It can be understood that in this step, the design data, historical monitoring data, and real-time monitoring data of the volcanic ash soil slope are input into the optimized neural network model to ensure that the model can comprehensively consider the static characteristics of the slope and the dynamic environmental changes. At the same time, the normalization processing of the data improves the adaptability of the model to multi-source heterogeneous data and ensures the stability and accuracy of the prediction results.

[0154] Step S36: Input the first information after normalization processing into the trained water content prediction model for prediction to obtain the model prediction data.

[0155] Step S37: Calculate the deviation between the model prediction data and the second information after normalization processing, and optimize the trained water content prediction model based on the deviation to obtain the final water content prediction model.

[0156] It can be understood that the dynamic adjustment of the weight parameters α and β of the comprehensive loss function. α and β are dynamically adjusted according to the deviation δ between the real-time monitoring data and the model prediction value, and their adjustment formula is:

[0157] The model parameter update formula is:

[0158]

[0159] Among them, The real-time monitoring data y i And the predicted value The average deviation. γ represents the physical consistency loss, the physical consistency priority coefficient, which controls the lowest weight of the physical consistency constraint. The significance of the adjustment mechanism is that when the deviation δ is large (the model fitting effect is poor), α→1, β→0, and the optimization mainly depends on the data-driven loss L data .

[0160] When the deviation δ is small (the model fitting effect is good), α→0, β→1, and the optimization mainly depends on the physical consistency loss L physics .

[0161] According to the real-time deviation δ, dynamically adjust the learning rate η to adapt to different stages of optimization. The formula for adjusting the learning rate is as follows:

[0162]

[0163] Among them, η tDenote the current learning rate. λ represents the learning rate adjustment coefficient, which is used to control the attenuation rate of the learning rate. When the deviation δ is large, the learning rate is small, and the optimization process is more meticulous; when the deviation is small, the learning rate increases to accelerate convergence.

[0164] Moreover, in this step, based on dynamically adjusting the learning rate and weight parameters, the gradient descent method is used to optimize the neural network parameters (weights and biases), and the formula for adjusting the weight parameters is as follows:

[0165]

[0166] where w t denotes the current weight parameter. denotes that the gradient of the comprehensive loss function is decomposed into a weighted sum of the data-driven part and the physical consistency part.

[0167] Step S4: Send the second information to the trained water content prediction model of the volcanic ash soil slope to obtain the water content parameters of the volcanic ash soil slope.

[0168] It can be understood that this step outputs the water content prediction results of the volcanic ash soil slope, including the dynamic boundary position of the unsaturated zone and the saturated zone, the water content change trend, and the rainfall infiltration depth, etc., providing a scientific basis for the slope stability assessment. The prediction results can be used to guide the engineering team to implement targeted slope management and disaster prevention and control measures.

[0169] Embodiment 2:

[0170] As Figure 2 shown, this embodiment provides a water content prediction system for a volcanic ash soil slope. Refer to Figure 2 The system includes an acquisition unit 701, a classification unit 702, a training unit 703, and a prediction unit 704.

[0171] The acquisition unit 701 is configured to acquire the first information and the second information. The first information includes the design data and historical monitoring data of the volcanic ash soil slope, and the second information includes the real-time monitoring data of the volcanic ash soil slope;

[0172] The classification unit 702 is configured to construct a water migration law model based on the switched form of the Richard equation, send the first information to the water migration law model for feature extraction, and classify the extracted features to obtain the classified first information;

[0173] The training unit 703 is configured to input the classified first information and the water migration law model into a preset neural network model for training to obtain a trained water content prediction model;

[0174] A prediction unit 704 is configured to send the second information to the trained moisture content prediction model to obtain the moisture content parameter of the pozzolanic soil slope.

[0175] It should be noted that, regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0176] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0177] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for predicting the moisture content of volcanic ash soil slope, characterized in that: include: Acquiring first information and second information, wherein the first information includes design data and historical monitoring data of the volcanic ash soil slope, and the second information includes real-time monitoring data of the volcanic ash soil slope; Building a water migration law model based on the switched form of the Richard equation, sending the first information to the water migration law model for feature extraction, and classifying the extracted features to obtain the classified first information; Inputting the classified first information and the water migration law model into a preset neural network model for training to obtain a trained water content prediction model; The second information is sent to the trained moisture content prediction model to obtain the moisture content parameter of the volcanic ash soil slope.

2. The method for predicting the moisture content of volcanic ash soil slope according to claim 1, characterized in that ,Based on the switching form of Richard equation, a water migration law model is constructed, including: The comprehensive saturation index is calculated based on a preset comprehensive saturation index calculation formula; Based on the comparison between the comprehensive saturation index and the preset critical saturation threshold of the soil, if the comprehensive saturation index is greater than or equal to the preset critical saturation threshold of the soil, the Richard equation in the form of pressure head is used to construct a formula to obtain the Richard equation in the form of pressure head; If the comprehensive saturation index is less than the preset critical saturation threshold of the soil, the Richard equation in the form of moisture content is used to construct a formula to obtain the Richard equation in the form of moisture content; The constructed Richard equation in the form of pressure head and the Richard equation in the form of moisture content are used as models of water migration.

3. The method for predicting the moisture content of volcanic ash soil slope according to claim 1, characterized in that , and sending the first information to a preset classification model for classification processing, including: Sending the first information to a water migration law model for feature extraction to obtain feature data corresponding to the first information; Constructing a hierarchical structure model for the first information and feature data corresponding to the first information to obtain a hierarchical structure model; Classify the first information and feature data corresponding to the first information based on the hierarchical structure model to obtain at least two categories of data information; Analyze the data information of the same category in pairs to determine whether one of the data is a subset of the other data. If so, keep the subset data and delete the redundant data. If not, keep both sets of data to obtain the underlying data of each category. Repeatedly analyze the deleted redundant data in pairs and divide the data by hierarchical optimization until all data cannot be further divided into levels, and obtain hierarchical data information for each category; The hierarchical data information in all categories is combined to obtain the data information of all categories.

4. The method for predicting the moisture content of volcanic ash soil slope according to claim 3, characterized in that , inputting the classified first information and the water migration law model into a preset neural network model for training, including: Dividing the first information and feature data corresponding to the first information into a training set and a test set; Inputting the training set into a preset neural network model for preliminary training, wherein the training set is preliminarily trained based on a preset neural network prediction formula and loss function to obtain a preliminary prediction model; Optimizing the parameters of the preliminary prediction model based on a preset transfer learning loss function and a preset physical consistency loss calculation formula to obtain a preliminary optimized prediction model, wherein the parameters include weight parameters and bias parameters; The test set is sent to the preliminary prediction model for verification to obtain a trained moisture content prediction model.

5. The method for predicting the moisture content of volcanic ash soil slope according to claim 4, characterized in that ,Based on the preset transfer learning loss function and the preset physical consistency loss calculation formula, the parameters of the preliminary prediction model are optimized to obtain the preliminary optimized prediction model, including: Based on the preset data-driven loss calculation formula and physical consistency loss formula, we can get the transfer learning loss function by weighting: The neural network parameters are iteratively optimized by combining the gradient descent method with the transfer learning loss function to obtain an iteratively optimized neural network model, wherein the iterative optimization includes adjusting weights and adjusting bias parameters; The training set data is sent to the iteratively optimized neural network model for retraining to obtain a preliminary optimized prediction model.

6. A volcanic ash soil slope moisture content prediction system, characterized in that: include: an acquisition unit, configured to acquire first information and second information, wherein the first information includes design data and historical monitoring data of the volcanic ash soil slope, and the second information includes real-time monitoring data of the volcanic ash soil slope; A classification unit, configured to construct a water migration law model based on a switched form of the Richard equation, send the first information to the water migration law model for feature extraction, and classify the extracted features to obtain the classified first information; A training unit, used for inputting the classified first information and the water migration law model into a preset neural network model for training to obtain a trained water content prediction model; The prediction unit is used to send the second information to the trained moisture content prediction model to obtain the moisture content parameter of the volcanic ash soil slope.

7. The volcanic ash soil slope moisture content prediction system according to claim 6, characterized in that: The classification unit includes: A first construction subunit is used to calculate the comprehensive saturation index based on a preset comprehensive saturation index calculation formula; The second construction subunit is used to compare the comprehensive saturation index with the preset critical saturation threshold of the soil, and if the comprehensive saturation index is greater than or equal to the preset critical saturation threshold of the soil, the Richard equation in the form of pressure head is used to construct the formula to obtain the Richard equation in the form of pressure head; The third construction subunit is used to construct a formula using the Richard equation in the form of moisture content if the comprehensive saturation index is less than a preset critical saturation threshold of the soil, so as to obtain the Richard equation in the form of moisture content; The fourth construction subunit is used to use the constructed Richard equation in the form of pressure head and the Richard equation in the form of moisture content as a water migration law model.

8. The volcanic ash soil slope moisture content prediction system according to claim 6, characterized in that: The classification unit also includes: A first classification subunit is used to send the first information to a water migration law model for feature extraction to obtain feature data corresponding to the first information; A second classification subunit is used to construct a hierarchical structure model for the first information and the feature data corresponding to the first information to obtain a hierarchical structure model; A third classification subunit is used to classify the first information and the feature data corresponding to the first information based on the hierarchical structure model to obtain at least two categories of data information; The fourth classification subunit is used to analyze the data information of the same category in pairs to determine whether one of the data is a subset of the other data. If so, the subset data is retained and the redundant data is deleted. If not, both sets of data are retained to obtain the underlying data of each category. The fifth classification subunit is used to repeatedly analyze the deleted redundant data in pairs and divide the data by hierarchical optimization until all the data cannot be further divided into hierarchies, thereby obtaining hierarchical data information for each category; The sixth classification subunit is used to perform a union operation on the hierarchical data information in all categories to obtain data information of all categories.

9. The volcanic ash soil slope moisture content prediction system according to claim 8, characterized in that: The training unit comprises: A first training subunit, used to divide the first information and feature data corresponding to the first information into a training set and a test set; A second training subunit is used to input the training set into a preset neural network model for preliminary training, wherein the training set is preliminarily trained based on a preset neural network prediction formula and loss function to obtain a preliminary prediction model; A third training subunit is used to optimize the parameters of the preliminary prediction model based on a preset transfer learning loss function and a preset physical consistency loss calculation formula to obtain a preliminary optimized prediction model, wherein the parameters include weight parameters and bias parameters; The fourth training subunit is used to send the test set to the preliminary prediction model for verification to obtain a trained moisture content prediction model.

10. The volcanic ash soil slope moisture content prediction system according to claim 9, characterized in that: The third training subunit comprises: The fifth training subunit is used to perform weighting based on the preset data-driven loss calculation formula and the physical consistency loss formula to obtain the transfer learning loss function: The sixth training subunit is used to iteratively optimize the neural network parameters by combining the gradient descent method with the transfer learning loss function to obtain an iteratively optimized neural network model, wherein the iterative optimization includes adjusting weights and adjusting bias parameters; The seventh training subunit is used to send the training set data to the iteratively optimized neural network model for re-training to obtain a preliminarily optimized prediction model.

Images