Landslide geological disaster multi-index fusion forecasting method based on ERT monitoring

Abstract

The invention discloses a landslide geological disaster multi-index fusion forecasting method based on ERT monitoring. The method comprises the following steps: S1, carrying out long-term ERT monitoring on a slope body and acquiring monitoring data; s2, performing statistical analysis on the monitoring data to obtain the average resistivity, the correlation coefficient, the average resistivity change rate and the correlation coefficient change rate at different moments; s3, performing inversion based on the neural network to obtain an inversion resistivity profile map and an inversion resistivity change rate profile map, and performing water flowing fracture identification on the inversion resistivity change rate profile map; s4, carrying out landslide risk grade division according to the fracture development condition parameters of the section, the average resistivity and the correlation coefficient, and carrying out landslide risk early warning by taking the weighted sum of the obtained three risk grades as a comprehensive prediction index; and carrying out critical sliding prediction according to the absolute value product of the average resistivity change rate and the correlation coefficient change rate. According to the method, more accurate early warning and forecasting of medium-and-long-term, short-term and temporary landslide of the landslide can be realized.

Description

Technical Field

[0001] The invention relates to a landslide geological disaster prediction method, belonging to the technical field of landslide geological disaster monitoring and early warning. Background Art

[0002] Landslide geological disasters are a frequent global natural disaster, which causes more casualties and property losses than natural disasters such as earthquakes, floods, and typhoons. Domestic landslide disasters are characterized by huge numbers, wide distribution, high frequency, and strong suddenness, and the prevention and control situation is severe and complex. Landslide geological disaster monitoring can dynamically grasp the evolution stage and development trend of disasters, evaluate slope stability, and timely discover abnormal changes in landslides, so as to buy precious time for disaster prevention and mitigation. It is an effective means to actively prevent landslide geological disasters. Moreover, long-term monitoring data can provide rich basic information for scientific research on landslide geological disasters, help to reveal the cause mechanism of landslides, and formulate more scientific and reasonable prevention and control measures. At present, landslide geological disaster monitoring mainly focuses on surface displacement monitoring, deep displacement monitoring, meteorological monitoring, etc. Among them, surface deformation monitoring technology mainly monitors the displacement changes of surface feature points, cannot obtain information on geological activities inside landslides, and is easily affected by external factors such as atmospheric conditions, terrain undulations, vegetation coverage, and human activities. The timeliness and reliability of its monitoring and early warning are difficult to guarantee. Deep deformation monitoring mainly measures the horizontal displacement of the slope below the ground. It has high measurement accuracy and can efficiently obtain the sliding characteristics and development trends of the sliding surface, and timely discover the precursors of disasters. However, the monitoring equipment has problems such as high price, limited range and poor environmental adaptability, which restricts its universal application. Meteorological monitoring belongs to external inducement monitoring, with rainfall and rainfall intensity as core data. The meteorological early warning and forecast of landslide disasters adopts the critical precipitation threshold model, which can warn of landslide disasters at the regional scale. However, the early warning model rarely considers the differences in the geological environment of landslides, especially the analysis of the mechanism of instability and destruction of geological bodies, and the accuracy and reliability of the forecast are still limited.

[0003] The high-density resistivity method (ERT) is based on the difference in electrical conductivity of rock and soil in the landslide area. It can effectively identify the stratigraphic structure, sliding surface, and cracks in the landslide area. It has the advantages of being sensitive to low-resistance bodies, good site adaptability, convenient construction, and high cost-effectiveness. It is widely used in landslide disaster investigation. Li Fu et al. [Li Fu, Zhou Hongfu, Ge Hua. Exploration of electrical characteristics of different types of landslide bodies by high-density resistivity method, Geophysical and Geochemical Exploration, 2019, 43(1): 215-221] used the high-density resistivity method to detect soil landslides, rock landslides, and broken rock landslides, and used the resistivity anomaly characteristics to infer the type and material composition of the landslide body. Wang Lei et al. [Wang Lei, Li Xiaobo, Su Zhandong, et al. Application of high-density electrical method in landslide exploration at the loess-mudstone contact surface, Journal of Geomechanics, 2019, 25(4): 536-543] analyzed the resistivity difference of loess with different water contents through two-dimensional resistivity section and inferred the location of loess-mudstone contact surface. Lin Song et al. [Lin Song, Wang Wei, Deng Xiaohu, et al. Geological and geophysical electrical characteristics of typical landslides in the Three Gorges Reservoir area, Journal of Jilin University (Earth Science Edition), 2020, 50(1): 273-284] applied ERT to the landslide investigation in the Three Gorges Reservoir area and obtained the stratigraphic structure and slip surface of the landslide body through two-dimensional inversion and analysis. Liu Dong et al. [Liu Dong, Zhang Fanyu, Chen Li, et al. Application of high-density electrical method in structural detection and three-dimensional modeling of loess landslides, Progress in Geophysics, 2022, 37(4): 1742-1748] used two-dimensional ERT to identify the loess-mudstone stratigraphic structure, loess thickness, water-rich area distribution, and sliding surface spatial characteristics of loess landslides, and constructed a three-dimensional geological model based on borehole lithology data. In order to dynamically obtain geological information within the landslide body, especially the infiltration and migration of rainfall, the evolution of the sliding surface, etc., many scholars have carried out landslide hazard monitoring research based on ERT. Tian Zhongying et al. [Tian Zhongying, Zhang Maosheng, Feng Li, et al. Detection of dominant channels of loess landslide based on comprehensive geophysical exploration, Northwest Geology, 2019, 52(2): 172-180] used high-density electrical method with small point spacing and ground penetrating radar method to achieve fine characterization of dominant channels such as micro-cracks in loess, and discussed the transformation relationship between the gradual development of micro-cracks into cracks and then the formation of sliding surfaces. Yan Yajing et al. [Yan Yajing, Yan Yongshuai, Zhao Guizhang, et al. Study on the law of water migration in natural slopes based on high-density electrical method, Rock and Soil Mechanics, 2019, 40(7): 2807-2814] used high-density electrical method combined with time domain reflection technology to construct the corresponding relationship between slope soil moisture content and resistivity, and analyzed the distribution characteristics and migration law of groundwater in slope media. Xu Tao et al. [Xu Tao, Wang Xiaomao, Chen Shuangshuang, et al. Experimental study on time-shift resistivity detection of seepage and sliding of expansive soil slopes. Yangtze River, 2022, 53(2): 170-175] Based on the basic principle of time-shift resistivity method, the geophysical electrical response characteristics of water infiltrating through cracks and defects in expansive soil slopes were explored.Meng Yongdong et al. [Meng Yongdong, Zhang Weijie, Cai Zhenglong, et al. High-density electrical model test study on rainfall infiltration of landslides. Progress in Geophysics, 2023, 38(4): 1842-1852] established a mathematical relationship between rock and soil resistivity and saturation through saturation measurement tests, and used the high-density resistivity method to monitor the rainfall infiltration process of landslides under multi-period interval rainfall conditions.

[0004] In summary, the high-density resistivity method (ERT) is an effective method for landslide disaster investigation and monitoring. However, the current application is still mainly based on landslide geological disaster investigation based on single measurement, or mainly based on short-term rainfall infiltration law and sliding surface evolution law monitoring. This single measurement or short-term small amount of monitoring data is difficult to support landslide disaster early warning and forecasting. In view of the dynamic evolution characteristics and causal complexity of landslide disasters, to realize the early warning and forecasting of landslide geological disasters based on ERT monitoring, on the one hand, it is necessary to develop long-term, automated, and intelligent ERT monitoring technology to realize long-term data collection, automatic data transmission and preprocessing, intelligent inversion and feature recognition on site; on the other hand, it is necessary to integrate diverse characteristic parameters such as statistical analysis and inversion recognition based on ERT monitoring data to obtain comprehensive prediction indicators that characterize slope deformation, damage and development trends, and realize medium- and long-term, short-term and imminent landslide early warning and forecasting. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a multi-index fusion prediction method for landslide geological disasters based on ERT monitoring. Based on long-term ERT monitoring data, a comprehensive prediction index reflecting the slope damage trend is obtained to realize medium- and long-term, short-term and imminent landslide early warning and forecasting.

[0006] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0007] A multi-index fusion prediction method for landslide geological disasters based on ERT monitoring includes the following steps:

[0008] S1. Carry out long-term ERT monitoring of slopes and obtain monitoring data;

[0009] S2. Perform statistical analysis on the monitoring data to obtain the average resistivity at different times Correlation coefficient γ, average resistivity change rate δ, correlation coefficient change rate η;

[0010] S3. Invert the ERT monitoring data based on the neural network to obtain the resistivity profile reflecting the electrical distribution in the slope at different times and the inverted resistivity change rate profile reflecting the resistivity evolution in the slope, and identify the water-conducting fractures on the inverted resistivity change rate profile;

[0011] S4, respectively, according to the profile fracture development parameters κ, average resistivity The landslide risk level is divided into different levels according to the correlation coefficient γ, and the weighted sum of the three risk levels is used as the comprehensive prediction index to issue landslide risk warnings of different levels; the product of the absolute value of the average resistivity change rate δ and the correlation coefficient change rate η is used as the impending slip prediction index for impending slip prediction. When the impending slip prediction index rises sharply, it indicates that a landslide is about to occur; among them, the profile crack development parameter κ is obtained by level classification based on the conditions of the rear edge tension cracks, flank shear cracks and front edge swelling cracks in the identified water-conducting cracks.

[0012] Furthermore, before statistical analysis, the monitoring data is preprocessed including temperature correction, electrode position update, and random noise removal; wherein the temperature correction formula is: The resistivity ρ observed at temperature T is expressed as T Converted to resistivity at reference temperature T0 α is the rate of change of resistivity caused by unit temperature change.

[0013] More preferably, the value of α is 0.025°C -1 .

[0014] Further preferably, the random noise removal is performed using a wavelet threshold denoising method.

[0015] Preferably, automatic identification of water-conducting fractures based on deep learning is performed on the inverted resistivity change rate profile, as follows:

[0016] A training data set is constructed by mixing synthetic inversion resistivity change profile data containing water-conducting fractures and actual inversion resistivity change profile data;

[0017] Label the water-conducting fractures to obtain a labeled training data set;

[0018] Train the deep learning model using labeled training datasets;

[0019] The trained deep learning model is used to perform automatic identification of water-conducting fractures based on deep learning on the inverted resistivity change rate profile.

[0020] Preferably, the profile fracture development parameter κ is obtained by dividing the identified water-conducting fractures into five levels according to the conditions of the rear-edge tensile fractures, flank shear fractures and front-edge swelling fractures. The specific basis for the level division is as follows: Level 1: rear-edge tensile fractures appear on the inverted resistivity change rate profile, but the number of fractures is small, and the fracture width and depth are limited; no flank shear fractures and front-edge swelling fractures appear;

[0021] Level 2: The number of rear-edge tension cracks on the inverted resistivity change rate profile increases, and the crack width and depth increase; no flank shear cracks and front-edge expansion cracks appear;

[0022] Level 3: There are flank shear cracks on the inverted resistivity change rate profile, but the number of cracks is small, and the width and depth of the cracks are limited; there are no front edge expansion cracks;

[0023] Level 4: The number of flank shear cracks on the inverted resistivity change rate profile increases, and the crack width and depth increase; no front edge expansion cracks appear;

[0024] Level 5: The trailing edge tension cracks, flank shear cracks and leading edge uplift cracks have all appeared on the inverted resistivity change rate profile.

[0025] Preferably, the average resistivity is calculated according to the following method: Landslide risk level classification:

[0026] Level 1:

[0027] Level 2:

[0028] Level 3:

[0029] Level 4:

[0030] Level 5:

[0031] Among them, the average resistivity increases is the average resistivity threshold when triggering landslide events.

[0032] Preferably, the landslide risk level is classified according to the correlation coefficient γ in the following manner:

[0033] Level 1: (γ θ +3Δγ)<γ≤(γ θ +4Δγ);

[0034] Level 2: (γ θ +2Δγ)<γ≤(γ θ +3Δγ);

[0035] Level 3: (γ θ +Δγ)<γ≤(γ θ +2Δγ);

[0036] Level 4: Gamma θ <γ≤(γ θ +Δγ);

[0037] Level 5: γ≤γθ ;

[0038] Among them, the incremental step length of the correlation coefficient Δγ=(1-γ θ ) / 5,γ θ is the correlation coefficient threshold for triggering landslide events.

[0039] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0040] Based on long-term ERT monitoring data, the present invention performs real-time statistical analysis and inversion identification of ERT monitoring data to obtain key parameters reflecting the evolution of slope cracks and water migration. Through parameter classification and weighted superposition, a comprehensive prediction index reflecting the slope damage trend is obtained, thereby realizing mid- to long-term, short-term and imminent landslide early warning and forecasting with higher forecast accuracy.

[0041] The present invention performs ERT monitoring data inversion based on a neural network, and further adopts an automatic identification technology of water-conducting fissures based on deep learning, which has a high degree of automation and intelligence, and effectively improves the real-time performance of disaster early warning and forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a framework diagram of the implementation process of the multi-index fusion prediction method for landslide geological disasters based on ERT monitoring proposed in the present invention;

[0043] Figure 2 It is a schematic diagram of the layout structure of the ERT automatic monitoring system in a specific embodiment. DETAILED DESCRIPTION

[0044] In view of the shortcomings of the existing technology, the solution of the present invention is to perform real-time statistical analysis and inversion identification of ERT monitoring data based on long-term ERT monitoring data, obtain key parameters reflecting the evolution of slope cracks and water migration, and obtain comprehensive prediction indicators reflecting the slope failure trend through parameter classification and weighted superposition, thereby achieving more accurate medium- and long-term, short-term and imminent landslide early warning and forecasting.

[0045] The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring proposed in the present invention comprises the following steps:

[0046] S1. Carry out long-term ERT monitoring of slopes and obtain monitoring data;

[0047] S2. Perform statistical analysis on the monitoring data to obtain the average resistivity at different times Correlation coefficient γ, average resistivity change rate δ, correlation coefficient change rate η;

[0048] S3. Invert the ERT monitoring data based on the neural network to obtain the resistivity profile reflecting the electrical distribution in the slope at different times and the inverted resistivity change rate profile reflecting the resistivity evolution in the slope, and identify the water-conducting fractures on the inverted resistivity change rate profile;

[0049] S4, respectively, according to the profile fracture development parameters κ, average resistivity The landslide risk level is divided into different levels according to the correlation coefficient γ, and the weighted sum of the three risk levels is used as the comprehensive prediction index to issue landslide risk warnings of different levels; the product of the absolute value of the average resistivity change rate δ and the correlation coefficient change rate η is used as the impending slip prediction index for impending slip prediction. When the impending slip prediction index rises sharply, it indicates that a landslide is about to occur; among them, the profile crack development parameter κ is obtained by level classification based on the conditions of the rear edge tension cracks, flank shear cracks and front edge swelling cracks in the identified water-conducting cracks.

[0050] In order to facilitate public understanding, the following takes rainfall-type landslide geological disaster forecasting as an example and combines the accompanying drawings to explain the technical solution of the present invention in detail:

[0051] The multi-index fusion prediction method for landslide geological disasters in this embodiment is implemented as follows: Figure 1 As shown, the details are as follows:

[0052] (1) Analyze and determine the basic characteristics of landslides

[0053] Based on field investigation and collection and collation of existing geological, hydrological, geophysical, drilling and other data, the stratigraphic lithology, geological structure, hydrogeological conditions, etc. of the unstable slope are identified, and the type, scale, perimeter, depth of sliding surface, thickness of covering layer, etc. of the landslide are analyzed and determined.

[0054] (2) ERT automated monitoring system architecture

[0055] The ERT automatic monitoring system used in this embodiment is as follows Figure 2 As shown in the figure, the system is mainly composed of a collection unit, a communication unit and a power supply unit, wherein the collection unit includes an ERT host, a data cable, a converter, an electrode, etc.; the communication unit is a 4G network transmission module; and the power supply unit is a solar panel power supply or a mains power supply module. The system can realize the automatic collection and transmission of ERT apparent resistivity data, and meet the real-time, continuous, and online technical requirements and safety requirements of the monitoring project.

[0056] (3) Layout of ERT monitoring profiles

[0057] The layout of ERT monitoring profiles should fully consider factors such as landslide type, landslide perimeter, sliding surface depth, and testing conditions. The length of the monitoring profile should cover the landslide body and part of the landslide bed around it, and meet the detection depth requirements. When conditions permit, ERT monitoring generally lays out longitudinal and transverse profiles at the same time, and the number of profiles should not be less than 2. The longitudinal profile is arranged along the main deformation direction of the landslide; the transverse profile is arranged perpendicular to the main deformation direction of the landslide. A typical ERT monitoring profile layout is as follows: Figure 2 As shown, P1, P2 and P3 are longitudinal ERT sections, and S1 and S2 are transverse ERT sections.

[0058] (4) ERT monitoring apparent resistivity data preprocessing

[0059] Since the ERT monitoring profile is deployed on unstable slopes for a long time, the monitored apparent resistivity data will be affected by non-geological factors such as soil temperature changes, electrode position movement and random noise. Therefore, it is necessary to perform soil temperature correction, electrode position update and noise removal on the monitored apparent resistivity data to obtain the corrected monitored apparent resistivity data.

[0060] (a) Temperature correction

[0061] Temperature correction uses the following formula:

[0062] ρ T0 =ρ T (1+α(T-T0))

[0063] The resistivity ρ observed at temperature T is expressed as T Calculated as the resistivity ρ at the reference temperature T0 T0 , α is the rate of change of resistivity caused by unit temperature change.

[0064] The reference temperature T0 can usually be set to 20°C to 25°C. In this embodiment, 25°C is used. The temperature correction formula is: 25 =ρ T (1+α(T-25))

[0065] In the formula, ρ T is the resistivity observed at temperature T; ρ 25 is the resistivity at 25°C; α is the test constant, usually 0.025°C -1 , that is, a change of 1°C in temperature will result in a change of 2.5% in resistivity. The soil resistivity obtained by the temperature sensor placed in the soil can be used to uniformly calibrate the monitored apparent resistivity to the resistivity at 25°C, thereby eliminating the impact of temperature fluctuations.

[0066] (b) Electrode position update

[0067] During the ERT monitoring process, the position of the electrode needs to be measured regularly. Especially when the slope is significantly deformed, the current electrode position should be measured in time, and the three-dimensional coordinates of the electrode in the system should be updated in time to eliminate the resistivity error caused by the inaccurate spatial position of the electrode. When the electrode position on the landslide body changes, but the electrode position on the landslide bed does not change, only the electrode position on the landslide body can be updated to improve work efficiency.

[0068] (c) Random noise removal

[0069] For the random noise interference generated during the observation process, common data denoising methods such as wavelet denoising, mean filtering denoising, and sliding average denoising can be used. In this embodiment, the wavelet threshold denoising method is used to remove noise to reduce the impact of random noise on subsequent processing results.

[0070] (5) Calculation of statistical parameters of apparent resistivity in ERT monitoring

[0071] The corrected apparent resistivity data were statistically calculated to obtain the average resistivity reflecting the change of total water content in the slope. The correlation coefficient γ reflects the temporal and spatial evolution of the wetting process in the slope, and the average resistivity change rate δ and correlation coefficient change rate η are sensitive to the hydrostatic pressure in the slope and can reflect the change of slope water content in a short period of time.

[0072] Assume that T0 is a reference time, such as a time before a rainfall event; T1 and T2 are two times in a rainfall event.

[0073] The apparent resistivity observed at time T0 is: ρ0={ρ 01 , ρ 02 , ρ 03 ,…,ρ 0n}, n is the total number of profile data.

[0074] The apparent resistivity observed at time T1 is: ρ1={ρ 11 , ρ 12 , ρ 13 ,…,ρ 1n}, n is the total number of profile data.

[0075] The apparent resistivity observed at time T2 is: ρ2 = {ρ 21 , ρ 22 , ρ 23 ,…,ρ 2n}, n is the total number of profile data.

[0076] Then: The average resistivity at T0, T1, and T2 and They are

[0077]

[0078] The correlation coefficient γ between time T1 and reference time T0 01 for:

[0079]

[0080] The correlation coefficient γ between time T2 and reference time T0 02 for:

[0081]

[0082] Average resistivity change rate δ at T1 and T2 12 for:

[0083]

[0084] The change rate of correlation coefficient at time T1 and T2 η 12 for:

[0085]

[0086] (6) Inversion imaging of ERT monitoring data based on neural network

[0087] The principle of ERT monitoring data inversion based on neural network is to generate a series of samples for learning through forward algorithm, and then use machine learning model to learn the samples, continuously adjust the structure and parameters of the model through learning, and finally generate an inversion model that can correctly interpret ERT monitoring data. Back propagation neural network (BP), radial basis function neural network (RBF), extreme learning machine (ELM) or its optimization algorithm PSO-BP, ABP, GA-BP, POA-ELM, etc. can be used to realize two-dimensional nonlinear inversion of ERT monitoring, and obtain resistivity profiles reflecting the electrical distribution in the slope at different times; at the same time, by comparing the inverted resistivity at a certain moment in the rainfall event with the inverted resistivity at a certain initial moment before the rainfall, the resistivity change rate profile reflecting the resistivity evolution in the slope can be obtained.

[0088] The learning samples of landslide ERT monitoring are generated by finite element forward modeling. Based on high-precision terrain measurement data, a two-dimensional slope model with terrain is constructed, including layered strata, water-conducting fractures, sliding surfaces, blocks and other models. Forward calculations are performed on them to obtain the apparent resistivity data of the sample model. The inherent relationship between the landslide ERT forward modeling apparent resistivity and model parameters is fully considered, and the forward modeling apparent resistivity of each model is used as the input node and the model parameters as the output node for network training.

[0089] When the number of electrodes and model parameters is large, the number of neural network input and output nodes is huge, and the hidden layer structure is complex, resulting in a lot of time required for training and testing. Therefore, in this embodiment, before the neural network inversion calculation, it is necessary to first perform dimensionality reduction processing on the high-dimensional training set data. Dimensionality reduction refers to the elimination of unimportant information while retaining the important features of the data, so as to achieve the purpose of reducing the number of neural network input nodes and improving the efficiency of inversion calculation. In view of the large amount of apparent resistivity data obtained by landslide ERT monitoring, this embodiment adopts principal component analysis (PCA) for dimensionality reduction, selects principal components with a cumulative contribution rate of more than 95% for neural network training, and selects the appropriate number of principal components by evaluating the prediction accuracy of the test set.

[0090] Assuming that the inversion resistivity of a certain inversion data point (x, y) in the slope at the reference time T0 before the rainfall event is ρ0, and the inversion resistivity at the time T1 during the rainfall event is ρ1, the change rate of the inversion resistivity at the time T1 relative to the time T0 is defined as:

[0091]

[0092] By calculating the inversion resistivity change rate of all inversion data points in the slope respectively, we can obtain the inversion resistivity change rate profile of the entire slope at time T1 relative to the reference time T0.

[0093] (7) Automatic identification of water-conducting fractures based on deep learning

[0094] The identification of water-conducting fissures on the ERT monitoring profile is a key task in landslide monitoring. The traditional identification method is mainly carried out in a human-computer interaction manner, which cannot be automated and has low efficiency. The identification accuracy is heavily dependent on the professional knowledge, interpretation experience and subjective understanding of the technicians. Therefore, it is not conducive to the long-term automated monitoring of ERT. To this end, the present invention further proposes a water-conducting fissure identification method based on deep learning, and performs automatic identification of water-conducting fissures based on deep learning on the inverted resistivity change rate profile. Deep learning models such as convolutional neural networks (CNN), deep residual networks (ResNet), and U-Net networks are used to realize the automatic identification of cracks such as rear edge tensile cracks, flank shear cracks, and front edge swelling cracks on the resistivity change rate profile.

[0095] The deep learning-based water-conducting fracture identification method uses forward synthetic data and actual data to jointly train the deep learning model. First, the synthetic inversion resistivity change rate profile data containing water-conducting fractures and the actual inversion resistivity change rate profile data are separated into fixed-size data blocks for making training data sets and verification data sets. The data sets contain linear and nonlinear water-conducting fractures. Secondly, the water-conducting fractures are annotated to obtain a labeled data set, and the deep learning model is trained with the labeled training data set, and the model prediction performance is verified with the labeled verification data set. Finally, the trained deep learning model is used to automatically identify the linear and nonlinear water-conducting fractures on the inversion resistivity change rate profile obtained by ERT monitoring. The identification process is completed automatically without setting any parameters.

[0096] (8) Obtaining medium-term, long-term and short-term comprehensive forecast indicators

[0097] Profile fracture development parameters κ, average resistivity The correlation coefficient γ mainly reflects the deformation behavior of the landslide, the development of the dominant channel in the slope, and the spatial and temporal distribution of water. In this embodiment, each parameter is divided into five levels according to the corresponding threshold when triggering the landslide, and the weight of each indicator is determined according to the dominant indicator. After weighted superposition, a medium-term, long-term, and short-term comprehensive prediction index Ψ reflecting the slope damage trend and damage time can be obtained. According to the comprehensive index Ψ, a blue, yellow, orange, and red four-level warning can be issued.

[0098] Through numerical simulation, physical simulation or statistics of monitoring data of similar landslides, the average resistivity threshold that triggers landslide events can be obtained. and correlation coefficient threshold γ θ .

[0099] (a) Average resistivity Classification: The average resistivity is defined as In general, take: According to the average resistivity threshold and progressive growth The average resistivity can be obtained The classification of levels is shown in Table 1.

[0100] (b) Classification of correlation coefficient γ: The progressive step length of the correlation coefficient is defined as Δγ. In general, Δγ=(1-γ θ ) / 5, according to the threshold value of the correlation coefficient γ θ The level classification of the correlation coefficient γ can be obtained by adding the incremental step length Δγ, see Table 1.

[0101] (c) Classification of the parameter κ of the profile crack development: Before the overall instability of the push-type landslide, it generally goes through a long deformation development evolution stage. In different evolution stages, local stress concentrations such as tensile stress, compressive stress, and shear stress will be generated in different parts of the landslide body, and cracks corresponding to their mechanical properties will be generated in the corresponding parts, including rear edge tensile cracks, flank shear cracks, and front edge swelling cracks. According to the staged matching characteristics of the crack system of the push-type landslide, after the slope deformation begins, the order of crack generation is rear edge tensile cracks, flank shear cracks, and front edge swelling cracks. When all three types of cracks have appeared and formed a basically closed surface crack morphology, it indicates that the sliding surface of the slope has been basically connected, the conditions for the overall instability and destruction of the slope have been met, and the landslide is about to occur. The generation and expansion of the rear edge tensile cracks, flank shear cracks, and front edge swelling cracks can be dynamically monitored through the ERT inversion resistivity change rate profile.

[0102] The occurrence and expansion of cracks in push-type landslides correspond to the evolution stage of the slope. According to the generation and expansion rules of tensile cracks at the rear edge of the landslide, shear cracks at the flanks, and swelling cracks at the front edge on the ERT monitoring section, the parameter κ can be graded, as shown in Table 1.

[0103] Table 1. List of prediction parameter classification

[0104]

[0105] is κ, γ is assigned weight parameters w1, w2, and w3 respectively, and weighted superposition is performed to obtain the comprehensive prediction index Ψ. The sum of weight parameters w1, w2, and w3 is 1, that is, w1+w2+w3=1. Generally speaking, the weight of the prediction parameter κ is relatively greater than and γ. Assume that at time t, The prediction parameter levels of γ are κ t , γ t , then the calculation formula of the comprehensive prediction index Ψ is as follows:

[0106]

[0107] According to the comprehensive indicators, four levels of warning can be issued: blue, yellow, orange, and red:

[0108] Blue warning: 1≤Ψ<2

[0109] Yellow warning: 2≤Ψ<3

[0110] Orange warning: 3≤Ψ<4

[0111] Red warning: 4≤Ψ

[0112] (9) Obtaining slip prediction indicators

[0113] The average resistivity change rate δ and the correlation coefficient change rate η are sensitive to the hydrostatic pressure in the slope and can accurately reflect the change of slope water content in a short period of time. The impending landslide prediction index Φ can be obtained from the average resistivity change rate δ and the correlation coefficient change rate η:

[0114] Φ=|δ|×η|

[0115] When the imminent landslide prediction index Φ changes gently, it means that the water content in the slope changes slowly; when the imminent landslide prediction index Φ rises sharply, it means that a large number of cracks in the slope are rapidly expanding and the sliding surface is connected, indicating that a landslide is about to occur.

Claims

1. A multi-index fusion prediction method for landslide geological disasters based on ERT monitoring, characterized in that: The following steps are involved: S1. Carry out long-term ERT monitoring of slopes and obtain monitoring data; S2. Perform statistical analysis on the monitoring data to obtain the average resistivity at different times Correlation coefficient γ, average resistivity change rate δ, correlation coefficient change rate η; S3. Invert the ERT monitoring data based on the neural network to obtain the resistivity profile reflecting the electrical distribution in the slope at different times and the inverted resistivity change rate profile reflecting the resistivity evolution in the slope, and identify the water-conducting fractures on the inverted resistivity change rate profile; S4, respectively, according to the profile fracture development parameters κ, average resistivity The landslide risk level is divided by the correlation coefficient γ, and the weighted sum of the three risk levels is used as a comprehensive prediction index to carry out landslide risk warning of different levels; The product of the absolute value of the average resistivity change rate δ and the correlation coefficient change rate η is used as the impending landslide prediction index for landslide prediction. When the impending landslide prediction index rises sharply, it indicates that a landslide is about to occur. Among them, the profile crack development parameter κ is obtained by graded classification based on the conditions of the rear edge tensile cracks, flank shear cracks and front edge swelling cracks in the identified water-conducting cracks.

2. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 1 is characterized in that: Before statistical analysis, the monitoring data is preprocessed including temperature correction, electrode position update, and random noise removal; the temperature correction formula is: The resistivity ρ observed at temperature T is expressed as T Converted to resistivity at reference temperature T0 α is the rate of change of resistivity caused by unit temperature change.

3. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 2 is characterized in that: The value of α is 0.025℃ -1 .

4. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 2 is characterized in that: The random noise removal is performed using a wavelet threshold denoising method.

5. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 1 is characterized in that: Automatic identification of water-conducting fractures based on deep learning is performed on the inverted resistivity change rate profile, as follows: A training data set is constructed by mixing synthetic inversion resistivity change profile data containing water-conducting fractures and actual inversion resistivity change profile data; Label the water-conducting fractures to obtain a labeled training data set; Train the deep learning model using labeled training datasets; The trained deep learning model is used to perform automatic identification of water-conducting fractures based on deep learning on the inverted resistivity change rate profile.

6. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 1 is characterized in that: The profile fracture development parameter κ is obtained by dividing the identified water-conducting fractures into five levels based on the conditions of the rear-edge tension fractures, flank shear fractures, and front-edge swelling fractures. The specific basis for the level division is as follows: Level 1: The inverted resistivity change rate profile shows rear-edge tension cracks, but the number of cracks is small, and the crack width and depth are limited; no flank shear cracks and front-edge expansion cracks appear; Level 2: The number of rear-edge tension cracks on the inverted resistivity change rate profile increases, and the crack width and depth increase; no flank shear cracks and front-edge expansion cracks appear; Level 3: There are flank shear cracks on the inverted resistivity change rate profile, but the number of cracks is small, and the width and depth of the cracks are limited; there are no front edge expansion cracks; Level 4: The number of flank shear cracks on the inverted resistivity change rate profile increases, and the crack width and depth increase; no front edge expansion cracks appear; Level 5: The trailing edge tension cracks, flank shear cracks and leading edge uplift cracks have all appeared on the inverted resistivity change rate profile.

7. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 1 is characterized in that: The average resistivity was calculated as follows: Landslide risk level classification: Level 1: Level 2: Level 3: Level 4: Level 5: Among them, the average resistivity increases is the average resistivity threshold that triggers landslide events.

8. The multi-index fusion prediction method for landslide geological disasters based on ERT monitoring as claimed in claim 1 is characterized in that: The landslide risk level is divided according to the correlation coefficient γ in the following way: 1st grade: (c θ +3Δγ)<γ≤(γ θ +4Δg); 2nd grade: (c θ +2Δγ)<γ≤(γ θ +3Δg); 3rd grade: (c θ +Δγ)<γ≤(γ θ +2Δg); 4th grade: c θ <γ≤(γ θ +Δg); Level 5: γ≤γ θ ; Among them, the incremental step length of the correlation coefficient Δγ=(1-γ θ ) / 5,γ θ is the correlation coefficient threshold for triggering landslide events.

Images