A prediction method for the evolution process of seepage danger in levees during the ice flood period

Abstract

The present invention discloses a method for predicting the evolution process of seepage danger of levees during the ice flood period, which identifies the evolution mechanism of seepage danger of levees during the ice flood period over time; reveals the variation law of seepage danger of levees during the ice flood period and its key driving factors; establishes a numerical simulation model for the evolution process of seepage danger of levees during the ice flood period; verifies the simulation accuracy of the evolution process of levee danger during the ice flood period; constructs a prediction sample set for the evolution process of seepage danger of levees during the ice flood period; establishes an intelligent prediction model for the evolution process of seepage danger of levees during the ice flood period based on the dual drive of physics and data; and applies the prediction to the evolution process of seepage danger of levees during the ice flood period. The first step. Compared with the prior art, the present invention realizes the prediction of the evolution process of seepage danger of levees during the ice flood period based on the dual drive of physics and data. Since this method uses rich sample data, it has the characteristics of high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of flood prevention and disaster reduction prediction, and in particular to a method for predicting the evolution process of dike seepage hazard during ice flood season based on physics-data dual drive. Background Art

[0002] Prediction of the evolution process of seepage hazard of levees during ice flood season is an important technical means for the prevention of levee hazard disasters. The current prediction methods of seepage hazard of levees during ice flood season are mainly based on artificial levee inspection, numerical simulation and comprehensive evaluation. Among them, the numerical simulation method is still difficult to verify its reliability due to the restriction of prototype monitoring data of levee hazard, and the three-dimensional numerical model still has the problem of low calculation timeliness, which is difficult to be applied to the rapid and accurate prediction of the evolution process of seepage hazard of levees during ice flood season. Although the conclusions obtained by artificial levee inspection combined with field practice and expert experience are relatively reliable, they are still difficult to meet the emergency defense needs of levee hazard disasters in the new era due to strong subjectivity, lack of in-depth consideration of the evolution mechanism of levee seepage hazard and obvious time lag. Regarding the evaluation of levee hazard during ice flood season, a dynamic assessment method of flood inundation risk of ice flood levee breach has been proposed, but its results can only provide certain support for quantitative diagnosis of levee hazard sections and prediction of levee hazard during ice flood season due to the lack of integration of the evolution mechanism of levee seepage hazard. Its applicability and reliability of prediction results still need to be further verified. In summary, the current prediction methods are difficult to quickly and accurately reflect the evolution law of levee seepage hazard during the ice flood period, and further in-depth research is needed on the prediction method of levee seepage hazard evolution process considering physical mechanisms during the ice flood period. Deep learning is the most popular algorithm in the field of artificial intelligence data drive. Although the introduction of data-driven deep learning technology into the field of levee seepage hazard prediction can improve timeliness, it is still difficult to improve both prediction accuracy and efficiency due to the lack of consideration of the evolution mechanism of levee seepage hazard.

[0003] Related references are as follows:

[0004] [1] Dai Changlei, Li Yang, Chen Mo, et al. Simulation analysis of seepage of a dike in cold regions under ice flood background [J]. Hydropower Energy Science, 2018, 36(12): 79-82+38.

[0005] [2] Tian Fuchang, Yuan Ximin, He Lixin, et al. Research progress on risk assessment and prevention of ice flood disasters in river channels, levees and floodplains in cold regions[J]. Journal of Hydraulic Engineering, 2022, 53(05): 549-559+573;

[0006] [3] Tian Fuchang, Yuan Ximin, He Lixin, et al. Study on the evolution characteristics of the Yellow River in Inner Mongolia and its correlation with ice flood disasters based on fractal theory [J]. Journal of Hydraulic Engineering, 2022, 53(6): 674-685. Summary of the invention

[0007] In view of the defects of the prior art, the present invention aims to propose a method for predicting the evolution process of seepage hazard of embankments during ice flood season, which adopts a physics-data dual-driven method, utilizes a numerical simulation model of the evolution process of seepage hazard of embankments during ice flood season, and utilizes an attention mechanism to optimize the spatiotemporal graph convolutional neural network to establish an intelligent prediction model of the spatiotemporal evolution process of seepage hazard of embankments during ice flood season based on physics-data dual-driven method, so as to realize the prediction of the evolution process of seepage hazard of embankments during ice flood season.

[0008] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0009] A method for predicting the evolution process of seepage danger of embankments during ice flood season, the method specifically comprising the following steps:

[0010] Step 1: Standardization and compilation of basic data: Standardize and organize the embankment hazard data based on the prototype monitoring data of embankment seepage hazard during the flood season from different dangerous sections. The embankment hazard related data at least involves the displacement of the infiltration line and slope, river water level, air temperature and soil temperature indicators at different depths;

[0011] Step 2, identify the evolution mechanism of levee seepage hazard over time during the ice flood period: use the rescaled range analysis method R / S to calculate the Hurst index and fractal dimension of the levee hazard data during the ice flood period, analyze the change trend, volatility and long-range correlation of the infiltration line and slope displacement, river water level, air temperature and soil temperature at different depths during the ice flood period, river closure period and river opening period, so as to explore the nonlinear dynamic change characteristics and multi-scale self-similarity characteristics of various indicators of the levee hazard data, and explore the fractal dynamic mechanism of the evolution of levee seepage hazard; identify the spatiotemporal characteristics of levee seepage hazard and hydrological and meteorological indicators during the ice flood period through comparative analysis by time period and region;

[0012] Step 3: Reveal the changing law of levee seepage hazard during ice flood season and its key driving factors: establish the changing correlation between levee seepage line, slope displacement, river water level and soil temperature in different dangerous sections during ice flood season at a fine-grained time scale, and quantitatively analyze the impact of temperature change on soil freeze-thaw depth. Combined with the evolution mechanism of levee seepage hazard during ice flood season obtained in step 2, analyze the evolution mechanism over time, reveal the interactive difference characteristics of multi-indicator time series of levee seepage hazard during ice flood season and its dynamic changing law, and obtain the key driving factors of levee hazard evolution;

[0013] Step 4: Establish a numerical simulation model for the evolution of seepage hazards in levees during the ice flood season: adopt the zoning and stratification optimization method for the permeability coefficient of frozen soil levees, mesh the levee sections, set the calculation parameters based on the levee soil material and freeze-thaw conditions, and establish a fluid-solid coupling calculation model for the seepage stability of levees under different frozen soil conditions during the ice flood season based on the seepage simulation of groundwater, the foundation stress-strain simulation of the building foundation under the upper load, and the slope stability analysis;

[0014] Step 5, verify the simulation accuracy of the evolution process of levee danger during the ice flood period: Consider the key driving factors of the evolution of levee seepage danger during the ice flood period, take the river water level as the change condition, simulate the evolution process of levee seepage danger in typical dangerous sections, compare the calculation results of typical ice floods monitored by the prototype and the change process of levee seepage danger with the analysis conclusions of the prototype monitoring data in step 3, repeatedly debug the model parameters, inversely verify the calculation accuracy of the model, and obtain the simulation error range and its applicable conditions;

[0015] Step 6. Construct a prediction sample set for the evolution process of seepage hazard of levees during the ice flood season: Consider the influence of climate change on ice flood process and soil freeze-thaw intensity and the coupling effect of multiple key driving factors of seepage hazard of levees during the ice flood season, set various combination calculation schemes of flood change process with different fluctuation gradients and different frozen soil conditions, i.e., frozen soil depth and stratified permeability coefficient; simulate the dynamic evolution process of seepage field, displacement field and stress field of dangerous sections of levees under different flood processes and different frozen soil conditions during the ice flood season based on the numerical simulation model of the evolution process of seepage hazard of levees during the ice flood season; analyze the spatiotemporal (x, y, t) evolution characteristics of levee seepage, back-water side slope collapse, i.e., slope displacement hazard and safety factor of different schemes, and construct a prediction sample set {X (input condition), Y (output hazard value)} for the evolution process of seepage hazard of levees during the ice flood season based on prototype monitoring and numerical simulation;

[0016] Step 7. Establish an intelligent prediction model for the evolution of levee seepage hazard during the ice flood period based on physics and data dual drive: Use the levee seepage hazard evolution prediction sample set {X, Y} constructed based on the physical mechanism simulation results as data drive, use the spatiotemporal graph convolutional neural network to learn and train and deeply identify the levee seepage hazard evolution mechanism driven by multi-factor coupling, and use the attention mechanism to optimize the spatiotemporal graph convolutional neural network to establish an intelligent prediction model F(X) for the spatiotemporal evolution of levee seepage hazard during the ice flood period based on physics and data dual drive; According to the levee seepage hazard prediction results, analyze the prediction accuracy of the levee seepage hazard evolution process during the ice flood period, and obtain the model parameters with the best prediction accuracy; Different grid cells of the same levee profile are abstracted into a directed graph G = (q, V, A) for the prediction of the evolution of levee seepage hazard during the ice flood period, where q represents the levee seepage hazard index value corresponding to the grid cell, V represents the grid cell set, and A represents the adjacency matrix between grid cells;

[0017] The spatiotemporal graph convolutional neural network based on attention mechanism optimization is used to realize the automatic learning of the short-term variation characteristics and long-term variation laws of the seepage danger of the embankment profile. t is the model input vector at time t, including ice flood water level, embankment frozen soil depth, stratified permeability coefficient, etc. t The output vector of the model at time t includes the elevation of the dike seepage line, seepage rate, slope drop, displacement and safety factor;

[0018] Step 8. Prediction and application of the evolution process of levee seepage hazard during ice flood season: According to the input condition X format requirements of the prediction model sample set, prepare the key driving factor data corresponding to the levee seepage hazard during ice flood season, use it as the model input condition, run the model, and output the corresponding spatiotemporal dynamic change process of the levee hazard during ice flood season, thereby realizing the application of the prediction model.

[0019] The beneficial effects and advantages of the present invention are:

[0020] This method realizes the prediction of the evolution process of embankment seepage hazard during the ice flood season based on physical and data dual drive. Due to the use of rich sample data, it has the characteristics of high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the prediction process of the evolution of seepage danger of embankments during ice flood season of the present invention;

[0022] Figure 2 This is a diagram of the structure of a spatiotemporal graph convolutional neural network optimized based on the attention mechanism. DETAILED DESCRIPTION

[0023] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0024] Embodiment 1:

[0025] like Figure 1 As shown, the prediction method for the evolution process of seepage hazard of levees during ice flood season based on physics-data dual drive of the present invention is implemented as follows: The prediction method for the evolution process of seepage hazard of levees during ice flood season based on physics-data dual drive comprises the following steps in sequence:

[0026] Step 1: Standardization and compilation of basic data: Standardize and organize the embankment hazard data based on the prototype monitoring data of embankment seepage hazard during the flood season from key embankment sections in different embankment sections. The embankment hazard data include index data such as infiltration line and slope displacement, river water level, air temperature and soil temperature at different depths;

[0027] Step 2, identify the time mechanism of the evolution of seepage hazard of levees during ice flood period: use the rescaled range analysis method R / S to calculate the Hurst index and fractal dimension of the levee hazard data during ice flood period, clarify the physical meaning of fractal dimension, and analyze the change trend, volatility and long-range correlation of different indicators in different periods of ice flood period over time, so as to explore the nonlinear dynamic change characteristics and multi-scale self-similar characteristics of various indicators of the levee hazard data, where different periods refer to ice flow period, river closure period and river opening period; different indicators refer to the displacement of infiltration line and slope, river water level, air temperature and soil temperature at different depths, and the change trend refers to rising or falling. Hurst index and fractal dimension mainly reflect volatility and long-range correlation. Nonlinear dynamic change characteristics are dynamic change characteristics in which the two variables are not in direct proportion, that is, non-uniform and chaotic. Multi-scale self-similar characteristics are consistent characteristics presented by indicators at different scales, that is, strict similarity between parts and the whole. The self-similarity property can be characterized by the Hurst index and fractal dimension D: when H = 0.5, D = 1.5, c(t) = 0, it means that the time series belongs to an independent random Brownian motion; when H>0.5, D<1.5, c(t)>0, it means that the time series has a certain long-range correlation and persistence, the past and future change trends are consistent, and the change increments show a positive correlation; when H<0.5, D>1.5, c(t)<0, it means that the time series has anti-persistence, the past and future change trends are opposite, and the change increments show a negative correlation.

[0028] The fractal dynamic mechanism of the evolution of levee seepage hazard is explored, specifically by reflecting the dynamic mechanism through nonlinear dynamic change characteristics and multi-scale self-similarity characteristics, and also by characterizing and quantifying it through the Hurst index and fractal dimension D; through comparative analysis by time period and region, the spatiotemporal differentiation characteristics of levee seepage hazard and hydrological and meteorological indicators during the ice flood season are identified, where the spatiotemporal differentiation characteristics are different characteristics manifested in different spatial regions at different times, that is, the index value / characteristic value is affected by the difference in time and space, and changes with the change of time and space. In addition, "regional" refers to upstream to downstream, that is, low latitude to high latitude. Spatial differences are mainly determined by dividing long-distance levees into n small levee sections, studying the hazard change characteristics corresponding to the monitoring points of each levee section, and finding out the differences through the comparison of the characteristics of n levee sections or the comparison of the Hurst index and fractal dimension D; specifically, the time period refers to the ice flow period, river closure period and river opening period; the region refers to the upstream to downstream, that is, low latitude to high latitude;

[0029] The Hurst index and fractal dimension of the time series related to the seepage danger of the embankment during the ice flood period are as follows:

[0030] Assume that there exists a time series ξ(t), t=1,2,3,...,N; for any positive integer τ≥1, it is defined as follows:

[0031] Calculate the mean of the time series as shown below:

[0032]

[0033] Calculate the cumulative deviation of the time series as shown below:

[0034]

[0035] Calculate the range of the time series as shown below:

[0036] R(τ)=max X(t,τ)-min X(t,τ), 1≤t≤τ, τ=1,2,3,...N (3)

[0037] Calculate the standard deviation of a time series as shown below:

[0038]

[0039] Calculate the empirical scaling relationship as shown below:

[0040] R(τ) / S(τ)=R / S∝(τ / 2) H (5)

[0041] ln(R / S)=e+H lnτ (6)

[0042] D=2-H (7)

[0043] Where H is the Hurst index, D is the multi-time scale self-similar fractal dimension; e is a constant;

[0044] Step 3, reveal the changing law of levee seepage hazard during the ice flood period and its key driving factors: Considering the "leading-following" driving relationship between river water level, air temperature and soil temperature at different depths and levee hazard, the windowed time lag cross-correlation method WTLCC is used to establish the changing correlation between the levee infiltration line and slope displacement and river water level and soil temperature at different dangerous sections during the ice flood period at a fine-grained time scale, and quantitatively analyze the effect of temperature change on the freeze-thaw depth of soil. Combined with the time mechanism of the evolution of levee seepage hazard during the ice flood period obtained in step 2, the interactive difference characteristics of multiple indicators of levee seepage hazard during the ice flood period and its dynamic change law are revealed, and the key driving factors of the evolution of levee hazard are further clarified. The changing correlation relationship between each indicator is established at a fine-grained time scale, and after combined analysis with the evolution mechanism of step 2, the interactive difference characteristics of multiple indicators time series and their dynamic change law are obtained, and the influence of different indicators on the evolution of levee seepage is judged, and the factor with the largest influence is selected as the key driving factor;

[0045] Step 4. Establish a numerical simulation model for the evolution of seepage hazards in levees during the ice flood season: Considering the “bend effect” of river flow in river sections prone to ice flood disasters, combined with the impact of bridges and dam projects in local river sections, typical levee dangerous sections in different spatial positions are selected from low latitudes to high latitudes. In view of the differences in levee structure and detailed material properties, the frozen soil depth and permeability coefficient at different periods of the ice flood season, namely the ice flow period, river closure period and river opening period, are used to reflect the freezing and thawing conditions of the levee soil. The permeability coefficient of frozen soil levees is divided into zones and layers based on Geo-Studio. The levee section is meshed with a side length of 5-10 cm. The calculation parameters are set based on the levee soil materials and freezing and thawing conditions obtained through field measurements and surveys. SEEP / W, SIGMA / W and SLOPE / W are three modules that establish the fluid-solid coupling calculation model of levee seepage stability under different frozen soil conditions during the ice flood season. Among them, SEEP / W module is a module in Geo-studio, a numerical simulation software for geotechnical engineering and geotechnical environment based on unsaturated soil mechanics theory. It is a two-dimensional finite element numerical simulation analysis module. It has more advantages than other software in early modeling, seepage boundary setting and calculation, and later data processing. It can perform numerical simulation of saturated-unsaturated seepage field and complete two-dimensional steady-state and transient saturated-unsaturated seepage calculation. Its input conditions include stratum profile, soil-water characteristic curve of each stratum, unsaturated permeability function and boundary conditions. It is mainly used to simulate the seepage of groundwater in porous media (including soil and rock) and analyze various complex saturated and unsaturated seepage problems; this module adopts automatic meshing technology and unstructured meshing technology, and users can quickly and effectively mesh the study area. In the SEEP / W module, soil parameters such as soil-water characteristic curve and hydraulic conductivity curve can be customized or the software built-in typical soil parameters can be used. SIGMA / W module is Geo-Studio stress-deformation finite element analysis module. As a typical finite element analysis software, its unique constitutive model formula makes it have an absolute advantage in dealing with highly complex engineering problems. It can solve a series of engineering problems such as linear elastic-plastic problems, nonlinear elastic-plastic problems, and nonlinear problems. It can not only calculate the stability problems required in many projects, but also analyze stress and strain, and simulate the foundation stress and strain problems of building foundations under the action of upper loads in areas with large karst development; SLOPE / W module (Geo-Slope / W) is a professional analysis software for slope stability calculation and is a module in Geo-studio geotechnical engineering analysis software. It applies the limit equilibrium theory and comprehensively considers the influencing factors such as groundwater, pore dewatering pressure, soil-structure interaction, slope loading, earthquake, etc. It can establish a two-dimensional calculation model for complex soil layers and arbitrary sliding surface shapes, which is used to study the stability problems of geotechnical slopes or sites, calculate the safety factor of high slope stability, and analyze its stability.Geo-Slope / M can not only broaden the scope of its analysis through Geo-studio integrated solutions, but also overcome some limitations of pure limit equilibrium equations. Slope / W software is a commonly used software for calculating the safety factor of geotechnical slopes. It is mainly used for slope stability analysis. It can analyze geotechnical engineering problems such as simple or complex slip surface shape changes, pore water pressure conditions, soil properties, and different loading methods. The software uses limit equilibrium theory to model and analyze the pore water pressure distribution in slopes with different soil types, complex strata, and slip surface shapes. At the same time, it provides a variety of different types of soil models, and uses deterministic and random input parameter methods for analysis. It also allows users to do random stability analysis. The SLOPE / W module also uses finite element stress analysis to effectively calculate and analyze most slope stability problems. This module can also analyze almost all slope stability problems encountered by users in disciplines such as geological structure, civil engineering, and mining engineering.

[0046] Step 5, verify the simulation accuracy of the evolution process of levee danger during the ice flood period: Consider the key driving factors of the evolution of levee seepage danger during the ice flood period, take the river water level as the main change condition, simulate the evolution process of levee seepage danger in typical dangerous sections, compare the calculation results of typical ice floods monitored by the prototype and the change process of levee seepage danger with the analysis conclusions of the prototype monitoring data in step 3, repeatedly debug the model parameters, inversely verify the calculation accuracy of the model, and obtain the simulation error range and its applicable conditions;

[0047] Step 6. Construct a sample set for predicting the evolution process of seepage hazard of levees during the ice flood season: Consider the impact of climate change, especially extreme weather, extreme cold, extreme warmth, and sudden changes in cold and warm on the ice flood process and the freeze-thaw intensity of soil, consider the coupling effect of multiple key driving factors of the seepage hazard of levees during the ice flood season, set multiple combination calculation schemes of flood change processes with different fluctuation gradients and different frozen soil conditions, i.e., frozen soil depth and stratified permeability coefficient; simulate different flood processes during the ice flood season based on the numerical simulation model of the evolution process of seepage hazard of levees during the ice flood season , the dynamic evolution process of seepage field, displacement field and stress field in dangerous sections of embankment under different frozen soil conditions; analyzing the spatiotemporal (x, y, t) evolution characteristics of embankment seepage, backwater side slope landslide, i.e. slope displacement hazard and safety factor of different schemes, so as to construct a sample set {X (input conditions), Y (output hazard value)} for predicting the evolution process of embankment seepage hazard during ice flood period based on prototype monitoring and numerical simulation; the embankment seepage refers to the elevation of the infiltration line, the seepage rate and the seepage gradient; the backwater side slope landslide refers to the slope displacement.

[0048] Step 7. Establish an intelligent prediction model for the evolution of levee seepage hazard during the ice flood period based on physics-data dual drive: Take the levee seepage hazard evolution prediction sample set {X, Y} constructed based on the physical mechanism simulation results as data drive, use the spatiotemporal graph convolutional neural network to learn and train and deeply identify the levee seepage hazard evolution mechanism driven by multi-factor coupling, and use the attention mechanism to optimize the spatiotemporal graph convolutional neural network ASTGCN to enhance its ability to capture the spatiotemporal correlation of levee hazard, thereby establishing an intelligent prediction model F(X) for the spatiotemporal evolution of levee seepage hazard during the ice flood period based on physics-data dual drive; according to the levee seepage hazard prediction results, use the mean square error RMSE, mean absolute error MAE, accuracy, coefficient of determination R2 and resolvable variance VAR and other indicators to evaluate the reliability and robustness of the model, analyze the prediction accuracy of the evolution of levee seepage hazard during the ice flood period, and thus obtain the model parameters with the best prediction accuracy;

[0049] like Figure 1 As shown in Figure 1, the spatiotemporal graph convolutional neural network optimized based on the attention mechanism includes two parts: (1) the spatiotemporal graph convolutional neural network STGCN, which is composed of a combination model sequence of multiple graph convolutional networks GCN and gated recurrent unit networks GRU, where the graph convolutional network GCN is used to capture the spatial pattern of the evolution of the levee seepage hazard, and the gated recurrent unit network GRU is used to mine the neighboring time dependency, including the constructed update gate and reset gate, which is a variant of LSTM and can maintain high training efficiency under large sample data; (2) the attention mechanism optimization model, including three groups of spatial feature extraction components SAtt-temporal feature extraction components TAtt, which are used to capture the spatiotemporal correlation of indicator data between levee profile grids; and dynamically generated different connection weights {a t-n ,···,a t-1 ,a t}, where {h t-n ,···,h t-1 ,h t} represents the set of output states at time tn, t-n+1…t-1, t, {c t-n ,···,c t-1 ,c t} represents the set of candidate hidden states at time tn, t-n+1...t-1, t, corresponding to three sets of spatial feature extraction components SAtt-temporal feature extraction components TAtt. Different grid cells of the same levee profile are abstracted into a directed graph G = (q, V, A) for the prediction of the evolution process of levee seepage hazard during the ice flood period, where q represents the levee seepage hazard index value corresponding to the grid cell, V represents the grid cell set, and A represents the adjacency matrix between grid cells.

[0050] The spatial-temporal graph convolutional neural network (ASTGCN) based on the optimization of attention mechanism is used to realize the automatic learning of the short-term variation characteristics and long-term variation laws of the seepage danger of the embankment profile. t Y is the model input vector at time t, including ice flood water level, embankment frozen soil depth, stratified permeability coefficient, etc. t is the model output vector at time t, including the elevation of the dike infiltration line, seepage rate, slope, displacement and safety factor, etc. The spatiotemporal graph convolutional neural network model configuration is shown in the following formula:

[0051] u t =σ(W u ·[GC(A,X t ),h t-1 ]+b u ) (8)

[0052] r t =σ(W r ·[GC(A,X t ),h t-1 ]+b r ) (9)

[0053] c t =tanh(W c ·[GC(A,X t ),(r t ·h t-1 )]+b c ) (10)

[0054] h t =u t *h t-1 +(1-u t )*c t (11)

[0055] In the formula, u t is the update gate, r t is the reset gate, c t is the candidate hidden state at time t, σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, h t is the output state at time t, h t-1 is the output state at time t-1, W u , W r , W c are the weight matrices of the update gate, reset gate and candidate state, respectively, and b u 、b r 、b care the update gate, reset gate, and bias vector of the candidate state, respectively. GC(·) represents the spatiotemporal graph convolution operation. * represents the vector product. n represents n time steps. By inputting the X values ​​at time tn, t-n+1, ​​..., t, the Y value at the corresponding time t is predicted.

[0056] Step 8. Prediction and application of the evolution process of levee seepage hazard during ice flood season: According to the input condition X format requirements of the prediction model sample set, prepare the key driving factor data corresponding to the levee seepage hazard during ice flood season, use it as the model input condition, run the model, and output the corresponding spatiotemporal dynamic change process of the levee hazard during ice flood season, thereby realizing the application of the prediction model.

[0057] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A method for predicting the evolution of seepage hazards in levees during ice flood season, characterized in that: The method specifically comprises the following steps: Step 1: Standardization and compilation of basic data: Standardize and organize the embankment hazard data based on the prototype monitoring data of embankment seepage hazard during the flood season from different dangerous sections. The embankment hazard related data at least involves the displacement of the infiltration line and slope, river water level, air temperature and soil temperature indicators at different depths; Step 2, identify the evolution mechanism of levee seepage hazard over time during the ice flood period: use the rescaled range analysis method R / S to calculate the Hurst index and fractal dimension of the levee hazard data during the ice flood period, analyze the change trend, volatility and long-range correlation of the infiltration line and slope displacement, river water level, air temperature and soil temperature at different depths during the ice flood period, river closure period and river opening period, so as to explore the nonlinear dynamic change characteristics and multi-scale self-similarity characteristics of various indicators of the levee hazard data, and explore the fractal dynamic mechanism of the evolution of levee seepage hazard; identify the spatiotemporal characteristics of levee seepage hazard and hydrological and meteorological indicators during the ice flood period through comparative analysis by time period and region; Step 3: Reveal the changing law of levee seepage hazard during ice flood season and its key driving factors: establish the changing correlation between levee seepage line, slope displacement, river water level and soil temperature in different dangerous sections during ice flood season at a fine-grained time scale, and quantitatively analyze the impact of temperature change on soil freeze-thaw depth. Combined with the evolution mechanism of levee seepage hazard during ice flood season obtained in step 2, analyze the evolution mechanism over time, reveal the interactive difference characteristics of multi-indicator time series of levee seepage hazard during ice flood season and its dynamic changing law, and obtain the key driving factors of levee hazard evolution; Step 4: Establish a numerical simulation model for the evolution of seepage hazards in levees during the ice flood season: adopt the zoning and stratification optimization method for the permeability coefficient of frozen soil levees, mesh the levee sections, set the calculation parameters based on the levee soil material and freeze-thaw conditions, and establish a fluid-solid coupling calculation model for the seepage stability of levees under different frozen soil conditions during the ice flood season based on the seepage simulation of groundwater, the foundation stress-strain simulation of the building foundation under the upper load, and the slope stability analysis; Step 5, verify the simulation accuracy of the evolution process of levee danger during the ice flood period: Consider the key driving factors of the evolution of levee seepage danger during the ice flood period, take the river water level as the change condition, simulate the evolution process of levee seepage danger in typical dangerous sections, compare the calculation results of typical ice floods monitored by the prototype and the change process of levee seepage danger with the analysis conclusions of the prototype monitoring data in step 3, repeatedly debug the model parameters, inversely verify the calculation accuracy of the model, and obtain the simulation error range and its applicable conditions; Step 6. Construct a prediction sample set for the evolution process of seepage hazard of levees during the ice flood season: Consider the influence of climate change on ice flood process and soil freeze-thaw intensity and the coupling effect of multiple key driving factors of seepage hazard of levees during the ice flood season, set various combination calculation schemes of flood change process with different fluctuation gradients and different frozen soil conditions, i.e., frozen soil depth and stratified permeability coefficient; simulate the dynamic evolution process of seepage field, displacement field and stress field of dangerous sections of levees under different flood processes and different frozen soil conditions during the ice flood season based on the numerical simulation model of the evolution process of seepage hazard of levees during the ice flood season; analyze the spatiotemporal (x, y, t) evolution characteristics of levee seepage, back-water side slope collapse, i.e., slope displacement hazard and safety factor of different schemes, and construct a prediction sample set {X (input condition), Y (output hazard value)} for the evolution process of seepage hazard of levees during the ice flood season based on prototype monitoring and numerical simulation; Step 7. Establish an intelligent prediction model for the evolution of levee seepage hazard during the ice flood period based on physics and data dual drive: Use the levee seepage hazard evolution prediction sample set {X, Y} constructed based on the physical mechanism simulation results as data drive, use the spatiotemporal graph convolutional neural network to learn and train and deeply identify the levee seepage hazard evolution mechanism driven by multi-factor coupling, and use the attention mechanism to optimize the spatiotemporal graph convolutional neural network to establish an intelligent prediction model F(X) for the spatiotemporal evolution of levee seepage hazard during the ice flood period based on physics and data dual drive; According to the levee seepage hazard prediction results, analyze the prediction accuracy of the levee seepage hazard evolution process during the ice flood period, and obtain the model parameters with the best prediction accuracy; Different grid cells of the same levee profile are abstracted into a directed graph G = (q, V, A) for the prediction of the evolution of levee seepage hazard during the ice flood period, where q represents the levee seepage hazard index value corresponding to the grid cell, V represents the grid cell set, and A represents the adjacency matrix between grid cells; The spatiotemporal graph convolutional neural network based on attention mechanism optimization is used to realize the automatic learning of the short-term variation characteristics and long-term variation laws of the seepage danger of the embankment profile. t is the model input vector at time t, including ice flood water level, embankment frozen soil depth, stratified permeability coefficient, etc. t The output vector of the model at time t includes the elevation of the dike seepage line, seepage rate, slope drop, displacement and safety factor; Step 8. Prediction and application of the evolution process of levee seepage hazard during ice flood season: According to the input condition X format requirements of the prediction model sample set, prepare the key driving factor data corresponding to the levee seepage hazard during ice flood season, use it as the model input condition, run the model, and output the corresponding spatiotemporal dynamic change process of the levee hazard during ice flood season, thereby realizing the application of the prediction model.

2. The method for predicting the evolution of seepage danger of dikes during ice flood season according to claim 1 is characterized in that: In step 1, the rescaled range analysis method R / S is used to calculate the Hurst index and fractal dimension of the time series related to the seepage hazard of the embankment during the ice flood period. The specific formula is as follows: Calculate the mean of the time series as shown below: Calculate the cumulative deviation of the time series as shown below: Calculate the range of the time series as shown below: R(τ)=maxX(t,τ)-minX(t,τ),1≤t≤τ Calculate the standard deviation of a time series as shown below: Calculate the empirical scaling relationship as shown below: R(τ) / S(τ)=R / S∝(τ / 2) H ln(R / S)=e+Hlnτ D=2-H Where H is the Hurst index, D is the multi-time scale self-similar fractal dimension, e is a constant, ξ(t) is the time series, t = 1, 2, 3, ..., N, and τ is any positive integer τ ≥ 1.

3. The method for predicting the evolution of seepage danger of dikes during ice flood season according to claim 1 is characterized in that: The step 4 comprises: combining the influence of bridges and dam projects in local river sections, selecting typical dangerous sections of embankments in different spatial locations from low latitudes to high latitudes, and reflecting the freezing and thawing conditions of the embankment soil through the frozen soil depth and permeability coefficient in different periods of the ice flood season, i.e., ice flow period, river closure period and river opening period, in view of the differences in embankment structure and detailed material properties.

4. The method for predicting the evolution of seepage danger of dikes during ice flood season according to claim 1 is characterized in that: In step 6, the embankment seepage refers to the elevation of the infiltration line, the seepage rate and the seepage gradient.

5. The method for predicting the evolution of seepage danger of dikes during ice flood season according to claim 1 is characterized in that: The spatiotemporal graph convolutional neural network based on attention mechanism optimization includes two parts: (1) a spatiotemporal graph convolutional neural network STGCN, which is composed of a combination model sequence of multiple graph convolutional networks GCN and gated recurrent unit networks GRU, wherein the graph convolutional network GCN is used to capture the spatial pattern of the evolution of the levee seepage hazard, and the gated recurrent unit network GRU is used to mine the neighboring time dependency, including the constructed update gate and reset gate; (2) an attention mechanism optimization model, including three groups of spatial feature extraction components SAtt-temporal feature extraction components TAtt, which are used to capture the spatiotemporal correlation of indicator data between levee profile grids; and dynamically generated different connection weights {a t-n ,···,a t-1 ,a t }, where {h t-n ,···,h t-1 ,h t } represents the set of output states at time tn, t-n+1…t-1, t, {c t-n ,···,c t-1 ,c t } represent the set of candidate hidden states at times tn, t-n+1…t-1, and t respectively.

6. The method for predicting the evolution of seepage danger of dikes during ice flood season according to claim 1, characterized in that: The spatiotemporal graph convolutional neural network model configuration is shown in the following formula: u t =σ(W u ·[GC(A,X t ),h t-1 ]+b u ) r t =σ(W r ·[GC(A,X t ),h t-1 ]+b r ) c t =tanh(W c ·[GC(A,X t ),(r t ·h t-1 )]+b c ) h t =u t *h t-1 +(1-u t )*c t In the formula, u t is the update gate, r t is the reset gate, c t is the candidate hidden state at time t, σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, h t is the output state at time t, h t-1 is the output state at time t-1, W u , W r , W c are the weight matrices of the update gate, reset gate and candidate state, respectively, and b u 、b r 、b c are the update gate, reset gate, and bias vector of the candidate state, respectively. GC(·) represents the spatiotemporal graph convolution operation. * represents the vector product. n represents n time steps. The output Y at time tn, t-n+1, ​​..., t is predicted through the input X at time tn, t-n+1, ​​..., t.

Images