Riverbank collapse early warning method and device based on multi-source data fusion

Abstract

This invention provides a method and device for riverbank collapse early warning based on multi-source data fusion. It comprehensively considers various influencing factors such as water and sediment conditions and riverbank boundary conditions, enabling a relatively comprehensive identification of the collapse intensity in collapse-prone areas, and possessing high objectivity and reliability. The method includes: Step 1: Determining collapse-prone areas; Step 2: Calculating the collapse probability; Step 3: Calculating the collapse width; Step 4: Calculating the width of the outer bank beach; Step 5: Calculating the area of ​​riverside residential housing within the study area; Step 6: Using collapse probability, collapse width, outer bank beach width, and riverside residential housing area as early warning indicators, classifying the early warning limits for each indicator; Step 7: Fusion of indicator calculation results; Step 8: Assigning weights to the indicators; Step 9: Determining the comprehensive collapse early warning level; Step 10: Providing the collapse early warning level classification results for the main collapse-prone areas of the study river section, and drawing the early warning information on the river morphology map accordingly to obtain a riverbank collapse early warning information map.

Description

Technical Field

[0001] This invention belongs to the field of riverbank collapse monitoring and early warning technology, specifically relating to a riverbank collapse early warning method and device based on multi-source data fusion. Technical Background

[0002] Bank collapse is an important component of riverbed deformation in alluvial rivers. However, large-scale bank collapses not only affect the stability of river courses in local sections, but also threaten the safety of important water-related projects such as dikes, water intakes, and bridges, increasing flood control pressure.

[0003] The mechanism of bank collapse is highly complex, influenced by numerous factors, and is an interdisciplinary problem belonging to river dynamics and soil mechanics. Since 1980, many researchers both domestically and internationally have conducted bank collapse research, revealing the specific impacts of different influencing factors on bank collapse, proposing calculation methods for slope stability under different collapse modes, and simulation methods for bank collapse processes at different scales. However, existing research has not yet clarified the interaction mechanisms between various factors, making bank collapse prediction still very difficult. In addition, existing bank collapse early warning technologies are not yet fully mature, with inconsistent early warning indicators, diverse grading methods, and significant differences in grading settings, making it difficult to convert results obtained from different methods into one another, and all methods have a strong subjective empirical component. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and apparatus for early warning of riverbank collapse based on multi-source data fusion:

[0005] <Method>

[0006] The riverbank collapse early warning method based on multi-source data fusion provided by this invention is characterized by comprising the following steps:

[0007] Step 1: Identify areas prone to bank collapse;

[0008] Obtain monitoring data for the study section of the river, and preliminarily select areas prone to bank collapse based on the direction of the thalweg and historical bank collapse patterns;

[0009] Step 2: Calculate the probability of bank collapse P, including the following sub-steps:

[0010] Step 2-1: Based on the measured cross-sectional topography, mark whether a bank collapse occurred at a specific cross-section in a specific year. Mark the riverbank that collapsed as 1, and the other bank as 0.

[0011] Step 2-2: Select water and sediment factors and river boundary condition factors, mainly including the following 13 factors: annual average flow Q and sediment transport rate Q s Maximum daily average flow Q max With the maximum daily average sediment transport rate Q s,max Floodplain duration T fRiver drainage rate R d The ratio of the slope above water to the slope below water, S u and S l Relative position of the deep thalassium L t Whether or not revetment is required (P), number of soil layers (N), difference in elevation between the beach and channel (H), and thickness of the cohesive soil layer (H). c The data from these 13 factors can form a data set for a specific cross-section CS under a specific year, denoted as S. cs-year Data group S cs-year The data included includes:

[0012] S cs-year ={Q, Q s Q max Q s,max T f R d S u S l L t P, N, H, H c};

[0013] Steps 2-3: Construct a random forest model to calculate the bank collapse probability P under specific water and sediment conditions. The steps are as follows:

[0014] (1) Sampling: Integrate measured water and sediment data and channel boundary condition data from different cross sections and years to form a total sample set, consisting of n S samples. cs-year Data set; based on the Bootstrap sampling method, m S are randomly selected each time. cs-year The data sets constitute the sub-training sample set D i Each sub-training sample set D i It contains m*13 data points;

[0015] (2) Calculate Gini impurity factor-wise: for each sub-training sample set D i A decision tree can be constructed for each of the following: at each node of the decision tree, M' factors are randomly selected from the M factors affecting the collapse of the bank, and the Gini impurity (D) of the M' factors at that node is calculated respectively.

[0016] (3) Determine the node splitting factor and construct the decision tree: Select the influence factor corresponding to the minimum Gini impurity as the splitting factor of the node until the decision tree splitting depth reaches the preset depth, and the decision tree construction is completed.

[0017] (4) Calculate the probability of bank collapse: use the validation set T i The data is fed into the constructed decision tree, and the tree is segmented node by node to obtain classification results for collapsed and non-collapsed banks. The proportion of nodes judged as collapsed banks to the total depth is the sub-training sample set D. iCalculate the probability p of bank collapse at a certain cross-section CS in a certain year. Ti (c / v).

[0018] (5) Repeat random sampling r times to form r sub-training sample sets D1, D2, D3...D r Build r decision trees and use the validation set T i Calculate the corresponding bank collapse probability p t (c / v)(t=T1,T2……T r ), and take the average of them to get the final bank collapse probability P:

[0019]

[0020] In the formula, c represents the predicted collapse, v is the total sample value, and p(c / v) represents the probability of bank collapse occurring under specific water and sediment conditions;

[0021] Step 3: Simulate the bank collapse process and calculate the bank collapse width B;

[0022] Step 4: Calculate the width W of the outer beach area based on the water quality index;

[0023] Step 5: Calculate the residential housing area Ar of riverside residents within the study area;

[0024] Step 6: Using the probability of bank collapse P, the width of bank collapse Bd, the width of the outer beach W, and the area of ​​riverside residential housing Ar as early warning indicators, divide the early warning limits for each indicator;

[0025] Step 7: Based on the Dempster-Shafer evidence theory, integrate the index calculation results;

[0026] Step 8: Assign weights to the four indicators: probability of bank collapse, width of bank collapse, width of riverbank outside the dike, and area of ​​riverside housing. i , i = 1, 2, 3, 4. The weight values ​​are determined based on the actual bank collapse situation.

[0027] Step 9: Determine the overall level of the bank collapse early warning system;

[0028] Step 10: Provide the classification results of the bank collapse early warning levels for the main bank collapse-prone areas in the research river section, and draw the early warning information on the river morphology map accordingly to obtain the river bank collapse early warning information map.

[0029] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following features: In step 3, the flow elements and riverbed scouring and deposition amplitude of each section are first calculated by a one-dimensional hydrodynamic module; based on the calculated flow conditions, the scouring degree of the riverbank toe is calculated, and the seepage module is used to calculate the change process of groundwater level inside the riverbank soil of each section; finally, the parameters such as flow conditions, riverbed scouring and deposition amplitude, soil moisture content and pore water pressure are used as input parameters of the collapse bank module to calculate the stability of the riverbank soil, determine whether it will collapse and calculate the collapse bank width Bd.

[0030] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following features: In step 4, firstly, a high-quality remote sensing image dataset of the study area and the flood season period is downloaded from the geospatial data cloud; subsequently, the improved water body index MNDWI is calculated using green light and shortwave infrared band surface reflectance images.

[0031] MNDWI=(ρ GREEN -ρ SWIR ) / (ρ GREEN +ρ SWIR );

[0032] In the formula, ρ GREEN With ρ SWIR These are the surface reflectances for green light and shortwave infrared bands, respectively.

[0033] The MNDWI water index is used to identify the water body and riverbank in the image, and the riverbank coordinates are extracted. Then, the coordinates of the dike are imported into the map, and the riverbank position is compared with the dike position during the flood season. The lateral distance between the riverbank and the dike is calculated as the width W of the outer beach area.

[0034] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following features: In step 6, the collapse probability P, collapse width Bd, width of the outer bank beach W, and area of ​​riverside residential housing Ar are used as early warning indicators; each indicator is divided into four levels A to D according to its numerical value, where A to C correspond to warning levels I to III, and D corresponds to no warning; the natural discontinuity grading method is used to classify the warning limits of the three indicators: collapse probability P, width of the outer bank beach W, and area of ​​riverside residential housing Ar.

[0035] For the width of bank erosion, banks with an average annual erosion width greater than 80m are classified as Grade A; those between 50 and 80m are classified as Grade B; those between 20 and 50m are classified as Grade C; and those less than 20m are classified as Grade D.

[0036] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following features: In step 7, a level set U = {A, B, C, D} is defined, and the level P (P ∈ U) to which each indicator belongs is determined according to the calculated value of the indicator, and the probability allocation function m under that level is assigned. i (P) value is set to 1.0, and m for other levels i (P) is set to 0.0; and the m of each level of a certain indicator i The sum of (P) is 1, which means that the relation is satisfied:

[0037]

[0038] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following characteristics: In step 9, for the comprehensive level P, the probability assignment function m(P) is calculated using the following formula:

[0039]

[0040] The level corresponding to the maximum probability value m(P) is selected as the final comprehensive level for bank collapse early warning.

[0041] Preferably, the riverbank collapse early warning method based on multi-source data fusion provided by the present invention may also have the following features: In step 9, after obtaining the comprehensive level result of the bank collapse early warning, the rationality of the calculation result of the comprehensive level of the early warning is judged based on the actual bank collapse situation: if the deviation is large, the weight is adjusted, and the calculation and judgment are repeated until the calculation result of the comprehensive level of the early warning is reasonable.

[0042] <device>

[0043] Furthermore, the present invention also provides a riverbank collapse early warning device based on multi-source data fusion that automatically implements the above-mentioned <method>, characterized in that it includes:

[0044] The prone area identification department obtains monitoring data of the study section of the river and preliminarily identifies the prone areas of bank collapse based on the direction of the thalweg and historical bank collapse situation.

[0045] The bank collapse probability calculation department calculates the bank collapse probability P through the following steps:

[0046] Step 2-1: Based on the measured cross-sectional topography, mark whether a bank collapse occurred at a specific cross-section in a specific year. Mark the riverbank that collapsed as 1, and the other bank as 0.

[0047] Step 2-2: Select water and sediment factors and river boundary condition factors, mainly including the following 13 factors: annual average flow Q and sediment transport rate Q s Maximum daily average flow Q max With the maximum daily average sediment transport rate Qs,max Floodplain duration T f River drainage rate R d The ratio of the slope above water to the slope below water, S u and S l Relative position of the deep thalassium L t Whether or not revetment is required (P), number of soil layers (N), difference in elevation between the beach and channel (H), and thickness of the cohesive soil layer (H). c The data from these 13 factors can form a data set for a specific cross-section CS under a specific year, denoted as S. cs-year Data group S cs-year The data included includes:

[0048] S cs-year ={Q, Q s Q max Q s,max T f R d S u S l L t P, N, H, H c};

[0049] Steps 2-3: Construct a random forest model to calculate the bank collapse probability P under specific water and sediment conditions. The steps are as follows:

[0050] (1) Sampling: Integrate measured water and sediment data and channel boundary condition data from different cross sections and years to form a total sample set, consisting of n S samples. cs-year Data set; based on the Bootstrap sampling method, m S are randomly selected each time. cs-year The data sets constitute the sub-training sample set D i Each sub-training sample set D i It contains m*13 data points;

[0051] (2) Calculate Gini impurity factor-wise: for each sub-training sample set D i A decision tree can be constructed for each of the following: at each node of the decision tree, M' factors are randomly selected from the M factors affecting the collapse of the bank, and the Gini impurity (D) of the M' factors at that node is calculated respectively.

[0052] (3) Determine the node splitting factor and construct the decision tree: Select the influence factor corresponding to the minimum Gini impurity as the splitting factor of the node until the decision tree splitting depth reaches the preset depth, and the decision tree construction is completed.

[0053] (4) Calculate the probability of bank collapse: use the validation set T iThe data is sequentially fed into the constructed decision tree, and each node is segmented to obtain classification results for collapsed and non-collapsed banks. The proportion of nodes judged as collapsed banks to the total depth is the sub-training sample set D. i Calculate the probability of bank collapse at a certain cross-section CS in a certain year.

[0054] (5) Repeat random sampling r times to form r sub-training sample sets D1, D2, D3...D r Build r decision trees and use the validation set T i Calculate the corresponding bank collapse probability p t (c / v)(t=T1,T2……T r ), and take the average of them to get the final bank collapse probability P:

[0055]

[0056] In the formula, c represents the predicted collapse, v is the total sample value, and p(c / v) represents the probability of bank collapse occurring under specific water and sediment conditions;

[0057] The simulation calculation unit simulates the bank collapse process and calculates the bank collapse width B.

[0058] The beach width calculation department calculates the width W of the beach outside the dike based on the water quality index.

[0059] The Housing Area Calculation Department calculates the housing area (Ar) of riverside residents within the research area;

[0060] The warning limit division uses the probability of bank collapse P, the width of bank collapse Bd, the width of the beach outside the dike W, and the area of ​​riverside residential housing Ar as warning indicators to divide the warning limits for each indicator.

[0061] The fusion computing department calculates fusion index results based on Dempster-Shafer evidence theory.

[0062] The weighting section assigns weights to four indicators: bank collapse probability, bank collapse width, width of the outer dike beach, and area of ​​riverside housing. i , i = 1, 2, 3, 4;

[0063] The warning level determination department determines the comprehensive warning level for bank collapse.

[0064] The early warning department provides the classification results of the bank collapse early warning levels for the main bank collapse-prone areas in the research river section, and generates early warning information on the river morphology map based on this, thus obtaining the river bank collapse early warning information map.

[0065] The control unit is connected to the prone area determination unit, the bank collapse probability calculation unit, the simulation calculation unit, the beach width calculation unit, the housing area calculation unit, the warning limit division unit, the fusion calculation unit, the weight assignment unit, the early warning level determination unit, and the early warning unit, and controls their operation.

[0066] Preferably, the riverbank collapse early warning device based on multi-source data fusion provided by the present invention may also have the following features: an input display unit, which is communicatively connected to the prone area determination unit, the bank collapse probability calculation unit, the simulation calculation unit, the beach width calculation unit, the housing area calculation unit, the warning limit division unit, the fusion calculation unit, the weight assignment unit, the early warning level determination unit, the early warning unit, and the control unit, for allowing users to input operation commands, and displaying the data and files of the corresponding units in text, table, or graphic manner according to the operation commands.

[0067] Preferably, the riverbank collapse early warning device based on multi-source data fusion provided by the present invention may also have the following features: The early warning unit displays the early warning information in the corresponding area of ​​the river course map according to the riverbank collapse early warning level classification results on the river course early warning information map, distinguishes different early warning levels with different colors, and when the early warning level of an area reaches a preset threshold, it sends an emergency riverbank collapse early warning information to the mobile or fixed terminal of the user responsible for the safety of engineering buildings in that area within a predetermined time, and at the same time, it provides a key reminder in that area on the river course map.

[0068] The role and effect of invention

[0069] Based on a large amount of measured data such as water and sediment, topography, and remote sensing images, this invention selects water and sediment factors and river boundary condition factors to construct a random forest model, establishes the relationship between multiple factors and bank collapse phenomena, thereby determining the probability value of bank collapse in a bank collapse-prone area in a specific year, calculates the width W of the outer beach based on the water index, delineates early warning limits, and uses the Dempster-Shafer evidence theory to integrate data of two types of indicators, namely bank collapse intensity and bank collapse hazard degree, to identify and classify the river bank collapse early warning level, and obtain a river section bank collapse early warning information map. This invention comprehensively considers various influencing factors such as water and sediment conditions and riverbank boundary conditions, enabling it to identify the intensity of bank collapse in prone areas in a relatively comprehensive manner. Moreover, the data fusion and bank collapse early warning identification and classification results mainly rely on measured data and are less empirical, thus possessing high objectivity and reliability. Therefore, this invention can scientifically, effectively, and rationally identify and predict bank collapse conditions in different areas, which helps to ensure timely and effective riverbank collapse management and emergency protection. It provides scientific and reliable technical support for improving riverbank stability and enhancing the safety of important water-related projects such as dikes, water intakes, and bridges. Attached Figure Description

[0070] Figure 1This is a flowchart of a riverbank collapse early warning method based on multi-source data fusion, as described in an embodiment of the present invention.

[0071] Figure 2 This is a flowchart illustrating the calculation of bank collapse probability based on a random forest model, as described in an embodiment of the present invention.

[0072] Figure 3 This is a schematic diagram showing the calculation results of the width W of the outer beach body of the Jing 45R fixed section and the area A of the riverside residential housing in an embodiment of the present invention;

[0073] Figure 4 This is an early warning information map of bank collapse in the lower reaches of the Yangtze River in 2020, which is related to an embodiment of the present invention. Detailed Implementation

[0074] The following description, in conjunction with the accompanying drawings, details the riverbank collapse early warning method and apparatus based on multi-source data fusion involved in this invention.

[0075] <Example>

[0076] like Figure 1 As shown, the riverbank collapse early warning method based on multi-source data fusion provided in this embodiment includes the following steps:

[0077] Step 1: Identify areas prone to bank collapse. Investigate and study the monitoring data of the river section, and preliminarily select areas prone to bank collapse based on the direction of the thalweg and historical bank collapse patterns.

[0078] Step 2: Calculate the probability P of the bank collapse.

[0079] Step 2.1: Based on the measured cross-sectional topography, mark whether a specific cross-section in a specific year has experienced bank collapse. Mark the riverbank that has experienced bank collapse as 1, and the other bank as 0.

[0080] Step 2.2: Select water and sediment factors and river boundary condition factors, mainly including the following 13 factors: annual average discharge (Q) and sediment transport rate (Q0). s ), maximum daily average flow (Q) max ) and maximum daily average sediment transport rate (Q s,max ), floodplain duration (T) f River drainage rate (R) d ), the ratio of the slope above water to the slope below water (S) u and S l ), relative position of the deep thalweg (L) t ), whether or not revetment is required (P), number of soil layers (N), difference in elevation between the beach and channel (H), and thickness of the cohesive soil layer (H). cThe data for these factors come from topographic monitoring data of fixed sections in the middle reaches of the Yangtze River and routine monitoring data from various hydrological or water level stations. The data for these 13 factors can constitute a data set for any section (CS) in a specific year, denoted as S. cs-year For example, the data set for the Jing 45R section in 2018 is denoted as S. 45R-2018 Data set S cs-year The data included includes:

[0081] S cs-year ={Q, Q s Q max Q s,max T f R d S u S l L t P, N, H, H c};

[0082] Step 2.3: Construct a random forest model to calculate the probability P of bank collapse under specific water and sediment conditions. The calculation process is illustrated in the diagram below. Figure 2 As shown. The steps are as follows:

[0083] (1) Sampling. Integrate measured water and sediment data and channel boundary condition data from different cross sections and years to form a total sample set, consisting of n S... cs-year Data set. Select 80% of the S data. cs-year The data set is the training set D, and the remaining 20% ​​of the data set is the validation set T. i Based on the Bootstrap sampling method, the training set D is sampled with replacement, and m samples S are randomly selected each time. cs-year The data sets constitute the sub-training sample set D i Each sub-training sample set D i It contains m*13 data points.

[0084] (2) Calculate the Gini impurity factor-wise. For each sub-training sample set D... i A decision tree can be constructed for each of the following: at each node of the decision tree, M' factors are randomly selected from the M factors affecting the collapse, and the Gini impurity (D) of each of the M' factors at that node is calculated. Specifically, for a given factor, a feature value can be randomly selected within its range, and the data can be divided into two classes based on the feature value. For example, suppose the sub-training sample set D... i In the data set, if the value of the influencing factor in b data groups is greater than the value, these data groups are assigned to the right branch and considered as the "predicted result is bank collapse" group. If b1 data groups in this branch have a bank marker of 1, the assignment is correct, and the probability of a correct prediction is p.k1 =b1 / b; If (b-b1) data sets are marked as 0, it indicates a misclassification, and the probability of a misprediction is 1-p. k1 Similarly, if (mb) data sets have a value for the influencing factor less than the value, they are assigned to the left branch's "predicted result: no bank collapse" group. If b² data sets in this group also have a bank marker of 0, the prediction is correct, with a probability of p. k2 = b² / (mb); If (mb - b²) data sets are marked as 1, it indicates a misclassification, and the probability of a misprediction is 1 - p. k2 Therefore, the Gini impurity (D) of this influence factor at this eigenvalue can be calculated:

[0085]

[0086] Within this range, different feature values ​​are selected multiple times and Gini(D) is calculated. The value corresponding to the minimum Gini impurity is selected as the segmentation value of the influence factor.

[0087] (3) Determine the node splitting factor and construct the decision tree. Compare the Gini impurity (D) of the M' influencing factors at a certain node, and select the influencing factor corresponding to the smallest Gini impurity as the splitting factor for that node. Divide the decision tree downwards sequentially until the decision tree splitting depth reaches the preset depth, and the decision tree construction is complete. Generally, the number of decision trees n is basically stable after it is greater than 100-120, and the depth can be roughly determined according to the number of training samples and the number of influencing factors. Among the combinations of these two parameter values, select the set that makes the model's predictive performance the best as the final parameter. After testing, 200 decision trees were constructed in this example, and the depth of each decision tree is 20, which can meet the requirements of the predictive performance for recall.

[0088] (4) Calculate the bank collapse probability. Use the validation set T... i Data group S cs-year Place them in order from D i In the constructed decision tree, each node is segmented, and each node can be classified as either bank collapse or non-bank collapse. The proportion of nodes judged as bank collapse to the total depth is the bank collapse probability of a certain cross-section CS in a certain year, calculated from the sub-training sample set Di.

[0089] (5) Repeat random sampling r times to form r sub-training sample sets D1, D2, D3...D r Build r decision trees and use the validation set T i Calculate the corresponding bank collapse probability p t (c / v)(t=T1,T2……T rThe average of these values ​​yields the final bank collapse probability P of the cross-section CS in that year:

[0090]

[0091] In the formula, c represents the predicted collapse, v is the total sample value, and p(c / v) represents the probability of bank collapse occurring under specific water and sediment conditions.

[0092] Generally, if the recall rates for both collapsed and non-collapsed banks in the validation set are greater than 80%, the trained model can be considered to have good predictive performance. The Random Forest algorithm can be implemented using Python.

[0093] Step 3: Simulate the bank collapse process and calculate the bank collapse width Bd. For details of the one-dimensional water and sediment movement coupled with the bank collapse process model, please refer to reference [1]. Its calculation process can be summarized as follows: First, the water flow elements and riverbed scouring and deposition amplitude of each section are calculated through the one-dimensional water and sediment dynamics module; based on the calculated water flow conditions, the scouring degree of the riverbank toe is calculated, and the seepage module is used to calculate the change process of the groundwater level inside the riverbank soil of each section; finally, the water flow conditions, riverbed scouring and deposition amplitude, soil moisture content and pore water pressure and other parameters are used as input parameters of the bank collapse module to calculate the stability of the riverbank soil, determine whether it will collapse and calculate the bank collapse width Bd.

[0094] Step 4: Calculate the width W of the outer beach area. First, download high-quality (e.g., Landsat Level-2 satellite imagery) datasets of the study area and flood season from the geospatial data cloud; then, calculate the improved water index MNDWI using green light and shortwave infrared surface reflectance images.

[0095] MNDWI=(ρ GREEN -ρ SWIR ) / (ρ GREEN +ρ SWIR );

[0096] In the formula, ρ GREEN With ρ SWIR These are the surface reflectances in the green light and shortwave infrared bands, respectively. In Landsat 5, 7, and 8 satellites, the Green light band corresponds to B2, B2, and B3, respectively, and the SWIR short outer infrared band corresponds to B5, B5, and B6, respectively.

[0097] The MNDWI water index can be used to identify the water body and riverbank in an image, and the riverbank coordinates can be extracted. Then, the coordinates of the dike are imported into the map, and by comparing the riverbank position with the dike position during the flood season, the lateral distance from the riverbank to the dike, i.e., the width W of the outer beach area, can be calculated. Taking the Jing 45R fixed section as an example, the calculation result of the outer beach area width W is shown in the diagram below. Figure 3 As shown.

[0098] Step 5: Using a fixed cross-section within the riverbank collapse-prone area as the center, extend 2km upstream and downstream, then extend 1.5km inland from the riverbank to form a rectangular area. Using this rectangular area as the study region, download the riverside residential housing area (Ar) within the study region from the Globeland30 website (http: / / www.globeland30.org / ). The results are as follows... Figure 3 As shown.

[0099] Step 6: Define each early warning indicator and its warning limit. The early warning indicators are the probability of bank collapse, the width of the bank collapse, the width of the outer dike beach, and the area of ​​riverside residential housing. Each indicator can be divided into four levels (A, D, and E) according to its numerical value, where AC corresponds to warning levels I-III, and D corresponds to no warning. The warning limits for the three indicators—probability of bank collapse P, width of the outer dike beach W, and area of ​​riverside residential housing Ar—are determined using the natural discontinuity grading method. The natural discontinuity grading method can be implemented using Python code.

[0100] For the width of the bank erosion Bd, banks with an average annual erosion width greater than 80m are classified as Grade A; those between 50 and 80m are classified as Grade B; those between 20 and 50m are classified as Grade C; and those less than 20m are classified as Grade D.

[0101] Step 7: Based on Dempster-Shafer (DS) evidence theory, integrate the indicator calculation results. Define a level set U = {A, B, C, D}. Based on the calculated values ​​of the indicators, determine the level P (P ∈ U) to which each indicator belongs, and assign the probability distribution function m under that level. i (P) value is set to 1.0, and m for other levels i (P) is set to 0.0. And the m values ​​for each level of a certain indicator... i The sum of (P) is 1, which satisfies the relation:

[0102]

[0103] Step 8: Assign weights to the four indicators: probability of bank collapse, width of bank collapse, width of riverbank outside the dike, and area of ​​riverside housing. i (i = 1, 2, 3, 4), which can generally be taken as 0.3, 0.2, 0.4 and 0.1.

[0104] Step 9: Determining the Comprehensive Bank Collapse Warning Level. For the comprehensive level P, calculate the probability assignment function m(P) using the following formula. Select the level corresponding to the maximum probability value m(P) as the final comprehensive bank collapse warning level. Combine this with the actual bank collapse situation to determine the reasonableness of the calculated comprehensive warning level. If the deviation is large (for example, the predicted result is Level I but the actual bank collapse is not severe; or the predicted result is Level IV but the actual annual average bank collapse width is greater than 80m, etc.), then adjust the weights.

[0105]

[0106] Step 10: Provide the bank collapse early warning level classification results for the main bank collapse-prone areas of the studied river section, and draw an early warning information map on the river morphology map, as shown below. Figure 4 The image shows a map illustrating the bank collapse early warning information. Table 1 presents the early warning levels for some bank collapse-prone areas in the lower Jingjiang section of the middle reaches of the Yangtze River in 2020, based on the above prediction results. A comparison of these results with the field survey findings shows they are largely consistent.

[0107] Table 1. Classification of Bank Collapse Warning Levels in Major Bank Collapse-Prone Areas of the Lower Jingjiang River in the Middle Reaches of the Yangtze River in 2020

[0108]

[0109] Furthermore, this embodiment also provides a riverbank collapse early warning device based on multi-source data fusion that can automatically implement the above method. The device includes a high-risk area determination unit, a collapse probability calculation unit, a simulation calculation unit, a beach width calculation unit, a housing area calculation unit, a warning limit division unit, a fusion calculation unit, a weight assignment unit, an early warning level determination unit, an early warning unit, an input display unit, and a control unit.

[0110] The section on identifying prone areas will follow the steps described in step 1 above, obtain monitoring data for the study section, and preliminarily identify prone areas based on the direction of the thalweg and historical bank collapse patterns.

[0111] The bank collapse probability calculation unit performs the steps described in step 2 above to calculate the bank collapse probability P.

[0112] The simulation calculation unit simulates the bank collapse process by performing the steps described in step 3 above and calculates the bank collapse width B.

[0113] The beach width calculation unit performs the steps described in step 4 above and calculates the width W of the beach outside the dike based on the water index.

[0114] The Housing Area Calculation Department performs the steps described in step 5 above to calculate the housing area Ar of riverside residents within the study area.

[0115] The warning limit division department executes the steps described in step 6 above, using the probability of bank collapse P, the width of bank collapse Bd, the width of the beach outside the dike W, and the area of ​​riverside residential housing Ar as warning indicators to divide the warning limits for each indicator.

[0116] The fusion computing unit performs the steps described in step 7 above, and calculates the fusion index results based on the Dempster-Shafer evidence theory.

[0117] The weighting process involves performing the steps described in step 8 above, assigning weights to the four indicators: bank collapse probability, bank collapse width, width of the outer dike beach, and area of ​​riverside housing. i , i = 1, 2, 3, 4.

[0118] The warning level determination department shall execute the steps described in step 9 above to determine the comprehensive warning level for bank collapse.

[0119] The early warning department executes the steps described in step 10 above, provides the classification results of bank collapse early warning levels for the main bank collapse-prone areas of the studied river section, and generates early warning information on the river morphology map accordingly, resulting in a river bank collapse early warning information map. The early warning department displays the early warning information in the corresponding areas of the river morphology map according to the bank collapse early warning level classification results, using different colors to distinguish different early warning levels. When the early warning level in an area reaches a preset threshold, the department sends an emergency bank collapse early warning information to the mobile or fixed terminals of users responsible for the safety of engineering structures in that area within a predetermined time, and simultaneously provides a key reminder for that area on the river morphology map.

[0120] The input display unit is connected in communication with the prone area determination unit, the bank collapse probability calculation unit, the simulation calculation unit, the beach width calculation unit, the housing area calculation unit, the warning limit division unit, the fusion calculation unit, the weight assignment unit, the warning level determination unit, and the warning unit. It is used to allow users to input operation commands and display the data and files of the corresponding units in text, table, or graphic form according to the operation commands.

[0121] The control unit is connected in communication with the prone area determination unit, bank collapse probability calculation unit, simulation calculation unit, beach width calculation unit, housing area calculation unit, warning limit division unit, fusion calculation unit, weight assignment unit, early warning level determination unit, early warning unit, and input display unit, and controls their operation.

[0122] The above embodiments are merely illustrative examples of the technical solutions of the present invention. The riverbank collapse early warning method and apparatus based on multi-source data fusion involved in the present invention are not limited to the contents described in the above embodiments, but are defined by the scope of the claims. Any modifications, additions, or equivalent substitutions made by those skilled in the art based on these embodiments are within the scope of protection claimed by the claims of the present invention.

Claims

1. A riverbank collapse early warning method based on multi-source data fusion, characterized in that, The method comprises the following steps: Step 1: determining the bank collapse prone area; Obtain the monitoring data of the study reach, and preliminarily select the bank collapse prone area according to the trend of the thalweg and the historical bank collapse situation; Step 2: calculating the bank collapse probability P, comprising the following sub-steps: Step 2-1: according to the measured section topography, mark whether the specific section will collapse in a specific year, mark the riverbank that will collapse as 1, and vice versa as 0; Step 2-2: Selecting the water and channel boundary condition factors, which include the following 13 factors: annual mean flow Q and sediment discharge Q s , maximum daily mean flow Q max and maximum daily mean sediment discharge Q s,max , flood duration T f , channel recession rate R d , water and underwater slope ratio S u and S l , relative position of the thalweg L t , whether there is a revetment P, number of soil layer N, bank height difference H and thickness of clay layer H c ; the data of these 13 factors can form a data set of a specific cross section CS in a specific year year, denoted as S cs-year , and the data set S cs-year includes the following data: S cs-year = {Q, Q s , Q max , Q s,max , T f , R d , S u , S l , L t , P, N, H, H c}; Step 2-3: constructing a random forest model to calculate the bank collapse probability P under specific water and sediment conditions, which comprises the following steps: (1) Sampling: The measured water and sediment data and the river boundary condition data of different sections and different years are integrated to form a total sample set, n S cs-year Data groups; based on the Bootstrap sampling method, m S cs-year Data groups are randomly selected each time to form a sub-training sample set D i Each sub-training sample set D i Contains m*13 data. (2) Factor-by-factor calculation of Gini impurity: each sub-training sample set D i A decision tree can be constructed, and at each node of the decision tree, M' factors are randomly selected from the M factors, and the Gini impurity Gini(D) of the M' factors at the node is calculated. (3) determining the node segmentation factor and constructing the decision tree: selecting the impact factor corresponding to the minimum Gini impurity as the segmentation factor of the node, until the splitting depth of the decision tree reaches the preset depth, and the decision tree is constructed; (4) Calculate the probability of bank collapse: the verification set T i Data is put into the constructed decision tree, and the classification results of bank collapse and non-bank collapse are obtained by node segmentation. The proportion of nodes judged as bank collapse in the total depth is the proportion of sub-training sample set D i The bank collapse probability of a certain section CS in a certain year year (5) Repeat random sampling r times to form r sub-training sample sets D1, D2, D3...D r Build r decision trees and use the validation set T i Calculate the corresponding bank collapse probability p t (c / v)(t=T1,T2……T r ), and take the average of them to get the final bank collapse probability P: In the formula, c represents the prediction of collapse, v is the total sample value, and p(c / v) represents the probability of bank collapse under specific water and sediment conditions; Step 3: simulating the bank collapse process and calculating the bank collapse width Bd; Step 4: calculating the embankment beach width W according to the water body index; Step 5: calculating the housing area Ar of the residents along the river in the study area; Step 6: taking the bank collapse probability P, the bank collapse width Bd, the embankment beach width W and the housing area Ar of the residents along the river as the early warning indexes, and dividing the early warning warning limits of each index; Step 7: based on the Dempster-Shafer evidence theory, fusing the index calculation results; Step 8: Assign weights w to the four indicators of bank collapse probability, bank collapse width, beach width outside the embankment, and housing area along the river, respectively i , i = 1, 2, 3, 4; Step 9: determining the bank collapse early warning comprehensive grade; Step 10: giving the bank collapse early warning grade division results of the main bank collapse prone area of the study reach, and drawing the early warning information on the river regime map to obtain the river channel bank collapse early warning information map.

2. The river channel bank collapse early warning method based on multi-source data fusion according to claim 1, characterized in that: wherein, In step 3, the water flow elements and the riverbed scouring and silting amplitude of each section are calculated through a one-dimensional water and sediment dynamics module; the scouring degree of the riverbank toe is calculated according to the calculated water flow conditions, and the subsurface water level change process of the soil body inside each section of the riverbank is calculated by using a seepage module; finally, the water flow conditions, the riverbed scouring and silting amplitude, and the soil moisture content and pore water pressure are taken as the input parameters of the bank collapse module to calculate the stability of the riverbank soil body, judge whether it will collapse, and calculate the bank collapse width Bd. 3.The riverbank collapse early warning method based on multi-source data fusion of claim 1, characterized in that: wherein, In step 4, high-quality remote sensing image data sets of the study area and the flood season period are obtained from the geographic spatial data cloud; then, the improved water body index MNDWI is calculated by using the green light and short-wave infrared band surface reflectivity image: MNDWI = (p GREEN + p SWIR ) / (p GREEN + p SWIR ); where ρ GREEN and ρ SWIR are the surface reflectances in the green and shortwave infrared bands, respectively. The water body and riverbank part in the image are judged by the MNDWI water body index, and the riverbank line coordinates are extracted; then, the coordinates of the levee are imported into the map, the position of the riverbank line in the flood season is compared with the position of the levee, and the transverse distance between the riverbank line and the levee is calculated as the embankment beach width W.

4. The river channel bank collapse early warning method based on multi-source data fusion according to claim 1, characterized in that: wherein In step 6, the bank collapse probability P, the bank collapse width Bd, the beach width W outside the embankment and the residential area Ar along the river are taken as the early warning indexes; each index is divided into four levels A-D according to the numerical value, wherein A-C correspond to the early warning levels I-III, and D corresponds to no early warning; the natural breakpoint classification method is adopted to divide the early warning limits of the bank collapse probability P, the beach width W outside the embankment and the residential area Ar along the river; For the bank collapse width, the bank collapse with an annual average bank collapse width greater than 80m is divided into A; when the annual average bank collapse width is between 50m and 80m, it is divided into B; when the annual average bank collapse width is between 20m and 50m, it is divided into C; and when the annual average bank collapse width is less than 20m, it is divided into D.

5. The river bank collapse early warning method based on multi-source data fusion according to claim 1, characterized in that: wherein, In step 7, a set of grades U = {A, B, C, D} is defined, and according to the calculated value of the index, the grade P (P e U) to which each index belongs is determined, and the probability distribution function m i (P) under the grade is assigned i (P) is set to 1.0, and m i (P) under the other grades is set to 0.0; and the sum of m i (P) of each grade of a certain index is 1, i.e. the relationship is satisfied:

6. The river bank collapse early warning method based on multi-source data fusion according to claim 1, characterized in that: wherein, In step 9, for the comprehensive level P, the probability assignment function m(P) is calculated by the following formula: The level corresponding to the maximum probability value m(P) is taken as the final bank collapse early warning comprehensive level.

7. The river bank collapse early warning method based on multi-source data fusion according to claim 1, characterized in that: wherein, In step 9, after obtaining the bank collapse early warning comprehensive level result, the rationality of the early warning comprehensive level calculation result is judged based on the actual bank collapse situation: if the deviation is large, the weight is adjusted, and the calculation and judgment are performed again until the early warning comprehensive level calculation result is reasonable.

8. A riverbank collapse early warning device based on multi-source data fusion, characterized in that, It comprises: An easy-to-happen area determination part that acquires monitoring data of a research river section, and preliminarily determines a bank collapse easy-to-happen area according to the depth-trend direction and historical bank collapse situation; A bank collapse probability calculation part that calculates the bank collapse probability P through the following steps: Step 2-1: According to the measured cross-section topography, mark whether a specific cross-section has bank collapse in a specific year, mark the riverbank with bank collapse as 1, and mark the opposite as 0; Step 2-2: Selecting the water and channel boundary condition factors, which include the following 13 factors: annual mean flow rate Q and sediment transport rate Q s , maximum daily mean flow rate Q max and maximum daily mean sediment transport rate Q s,max , flood duration T f , river water recession rate R d , water and underwater slope ratio S u and S l , relative position of the thalweg L t , whether there is a revetment P, number of soil layering N, bank height difference H and thickness of clay soil layer H c ; the data of these 13 factors can form a data set of a specific cross section CS in a specific year year, denoted as S cs-year , and the data set S cs-year includes the following data: S cs-year = {Q, Q s , Q max , Q s,max , T f , R d , S u , S l , L t , P, N, H, H c}; Step 2-3: Construct a random forest model to calculate the bank collapse probability P under specific water and sediment conditions, and the steps are as follows: (1) Sampling: The measured water and sediment data of different sections in different years and the river boundary condition data are integrated to form a total sample set, n S cs-year Data groups; based on the Bootstrap sampling method, m S cs-year Data groups are randomly selected each time to form a sub-training sample set D i Each sub-training sample set D i contains m*13 data. (2) Factor-by-factor calculation of Gini impurity: each sub-training sample set D i A decision tree can be constructed, and at each node of the decision tree, M' factors are randomly selected from the M factors, and the Gini impurity Gini(D) of the M' factors at the node is calculated. (3) Determine the node segmentation factor and construct the decision tree: select the impact factor corresponding to the minimum Gini impurity as the segmentation factor of the node, and continue until the decision tree splitting depth reaches the preset depth, and the decision tree construction is completed; (4) Calculate the probability of bank collapse: the verification set T i Data is put into the constructed decision tree, and the classification results of bank collapse and non-bank collapse are obtained by node segmentation. The proportion of nodes judged as bank collapse in the total depth is the proportion of sub-training sample set D i The bank collapse probability of a certain section CS in a certain year year (5) Repeat random sampling r times to form r sub-training sample sets D1, D2, D3...D r Build r decision trees and use the validation set T i Calculate the corresponding bank collapse probability p t (c / v)(t=T1,T2……T r ), and take the average of them to get the final bank collapse probability P: In the formula, c represents the prediction of collapse, v is the total sample value, and p(c / v) represents the probability of bank collapse under specific water and sediment conditions; A simulation calculation part that simulates the bank collapse process and calculates the bank collapse width B; A beach width calculation part that calculates the beach width W outside the embankment according to the water body index; A housing area calculation part that calculates the residential housing area Ar along the river in the research area; A warning limit division part that takes the bank collapse probability P, the bank collapse width Bd, the beach width W outside the embankment and the residential area Ar along the river as the early warning indexes, and divides the early warning limits of each index; A fusion calculation part that fuses the index calculation results based on the Dempster-Shafer evidence theory; The weight assignment unit respectively assigns weights w to the four indexes of the bank collapse probability, the bank collapse width, the beach width outside the dike, and the area of the house near the river. i i = 1, 2, 3, 4; An early warning level determination part that determines the bank collapse early warning comprehensive level; An early warning part that gives the bank collapse early warning level division result of the main bank collapse easy-to-happen area of the research river section, generates early warning information on the river regime map according to the result, and obtains the river bank collapse early warning information map. The control unit is connected with the prone area determination unit, the bank collapse probability calculation unit, the simulation calculation unit, the beach width calculation unit, the housing area calculation unit, the warning limit division unit, the fusion calculation unit, the weight assignment unit, the early warning grade determination unit and the early warning unit, and controls the operation of them. 9.The riverbank collapse early warning device based on multi-source data fusion of claim 8, wherein, Further comprising: The input display unit is connected with the prone area determination unit, the bank collapse probability calculation unit, the simulation calculation unit, the beach width calculation unit, the housing area calculation unit, the warning limit division unit, the fusion calculation unit, the weight assignment unit, the early warning grade determination unit, the early warning unit and the control unit, and is used for inputting operation instructions and displaying data and files of the corresponding units in the form of text, table or graph.

10. The river bank collapse early warning device based on multi-source data fusion according to claim 8, characterized in that: wherein, The early warning unit displays the early warning information on the river bank collapse early warning information map according to the early warning grade division result, and displays the early warning information in the corresponding area on the river regime map with corresponding colors, and distinguishes different early warning grades with different colors, and when the early warning grade of an area reaches a preset threshold, sends an emergency bank collapse early warning information to the mobile or fixed terminal of a user responsible for the safety of engineering buildings in the area within a predetermined time, and highlights the area on the river regime map.

Images