A Slope Risk Early Warning and Sliding Volume Prediction Method Based on Prophet-LSTM

Abstract

The present invention discloses a slope risk early warning and sliding volume prediction method based on Prophet-LSTM, which includes the steps: S1: Collect slope monitoring data; S2: Preprocess the slope monitoring data collected in step S1; S3: Based on the Prophet-LSTM algorithm, establish a displacement cycle prediction model and a trend prediction model to predict the slope displacement, and conduct real-time early warning on the slope displacement value and the slope prediction value; S4: Use the displacement monitoring data at different depths of the slope to analyze the depth of the potential sliding surface of the slope, connect the positions of multiple dangerous points to form a slope sliding surface, and calculate the sliding volume; S5: Establish a multi-parameter early warning framework for the slope to conduct a comprehensive evaluation of multiple indicators for the slope. The method in the present invention can predict and early warn the slope displacement, estimate the volume and position of potential landslides, so as to determine emergency measures according to the risk loss assessment, help to quickly and effectively identify landslide risks based on on-site monitoring, improve the understanding of landslide mechanisms and slope risk management.

Description

Technical Field

[0001] The present invention relates to the technical field of slope engineering, and particularly to a slope risk early warning and sliding volume prediction method based on Prophet-LSTM. Background Technique

[0002] Landslides, as a natural disaster with a relatively high occurrence frequency, have caused huge economic losses to countries around the world and seriously threatened people's lives and property safety. There are many factors causing landslides, among which excavation is an important factor in slope failure. The slope unloading during the excavation process will cause a rapid adjustment of the slope stress field and a gradual deformation of the geotechnical materials, resulting in the continuous expansion of the original cracks and the gradual generation of new cracks. At the same time, the influence of external environmental factors such as rainfall and groundwater level exacerbates the expansion of the cracks and reduces the shear strength of geotechnical materials such as soil and soft rock, making the slope more prone to failure. Traditional slope risk early warning methods mainly include physical model method and numerical simulation method, but they are usually limited by the complex geological characteristics of the slope. With the development of monitoring technologies and equipment, data-driven models based on deep learning are widely used in slope risk early warning, such as grey system models, regression models, chaos models, etc., to predict slope displacement and capture the non-linear dynamic behavior of slope displacement. However, the traditional data-driven model method has problems such as long prediction time, inability to evaluate in real time and dynamically, and inability to consider environmental factors such as rainfall and groundwater level changes. Therefore, how to conduct real-time dynamic evaluation and early warning of slope risk considering the influence of environmental factors such as rainfall and groundwater level is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0003] Aiming at the above existing problems, the present invention aims to provide a slope risk early warning and sliding volume prediction method based on Prophet-LSTM. After preprocessing the slope monitoring data, a slope displacement prediction model is established based on the Prophet-LSTM algorithm, and slope displacement prediction and early warning are carried out according to the model. A multi-parameter early warning framework is established to comprehensively evaluate the slope, estimate the volume and location of potential landslides, so as to determine emergency measures according to the risk loss assessment, which helps to quickly and effectively identify landslide risks based on on-site monitoring, improve the understanding of landslide mechanisms and slope risk management.

[0004] To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] A slope risk early warning and sliding volume prediction method based on Prophet-LSTM, characterized by including the following steps,

[0006] S1: Collect slope monitoring data;

[0007] S2: Preprocess the slope monitoring data collected in step S1;

[0008] S3: Based on the Prophet-LSTM algorithm, establish a displacement periodic prediction model and a trend prediction model to predict the slope displacement, and conduct real-time early warning on the slope displacement value and the slope prediction value;

[0009] S4: Use the displacement monitoring data at different depths of the slope to analyze the depth of the potential slip surface of the slope, connect the positions of multiple dangerous points to form a slope slip surface, and calculate the sliding volume;

[0010] S5: Establish a multi-parameter early warning framework for the slope and conduct a comprehensive evaluation of multiple indicators for the slope.

[0011] Further, the slope monitoring data described in step S1 includes slope displacements at different depths, groundwater levels, and rainfall amounts.

[0012] Further, the specific operations of step S2 include

[0013] S201: Adopt the 3σ criterion to delete outliers from the original slope monitoring data collected in step S1;

[0014] S202: Adopt the discrete wavelet transform method to denoise the slope monitoring data after deleting outliers, making the displacement time series curve smooth and orderly.

[0015] Further, the specific operations of step S202 include the following steps

[0016] S2021: Adopt the empirical mode decomposition method to decompose the noise signal of the slope monitoring data into high-frequency intrinsic mode functions and low-frequency intrinsic mode functions;

[0017] S2022: Conduct wavelet decomposition on the high-frequency intrinsic mode functions obtained in S2021 to obtain high-frequency wavelet coefficients and low-frequency wavelet coefficients;

[0018] S2023: Directly retain the numerical values of the low-frequency wavelet coefficients obtained in S2022; set a threshold to perform non-linear processing on the high-frequency wavelet coefficients to suppress the noise signal;

[0019] S2024: Perform the inverse operation of wavelet transform on the high-frequency wavelet coefficients after threshold processing in S2023 and the low-frequency wavelet coefficients with the original values retained to obtain a wavelet reconstruction signal;

[0020] S2025: Reconstruct the signal again for the low-frequency intrinsic mode functions obtained in S2021 and the wavelet reconstruction signal after wavelet reconstruction in S2024 to obtain a denoised signal.

[0021] Further, the specific operations of step S3 include the following steps

[0022] S301: Use the Prophet algorithm to decompose the time series of slope monitoring data into a periodic term and a trend term

[0023] y(t) = g(t) + s(t) + ∈ t

[0024] where y(t) is the time series of slope monitoring data, g(t) is the trend term, s(t) is the periodic term, and ∈ t is the error term, which follows a normal distribution;

[0025] S302: Use the Prophet algorithm to predict the trend term;

[0026] S303: Use the LSTM algorithm to predict the periodic term and establish an LSTM neural network model between the groundwater level, rainfall, and periodic displacement of the slope;

[0027] S304: Add the results of the trend term and the periodic term to obtain the predicted value of the slope displacement;

[0028] S305: According to expert experience, set the risk thresholds of slope displacement for low risk, medium risk, and high risk respectively, and conduct real-time early warning of the slope displacement value and the predicted value;

[0029] S306: According to the creep theory of rock and soil masses, divide the slope movement mode into the initial deformation stage, uniform deformation stage, and accelerated deformation stage, set the risk thresholds of slope displacement rates for each stage respectively, and conduct real-time early warning of the slope displacement rate.

[0030] Further, the specific operations of step S301 include the following steps

[0031] S3011: Based on the mutation point theory, use Prophet to automatically identify the mutation points in the time series, use a piecewise linear model to fit the deformation time series between two adjacent mutation points, and extract the trend term. Then the trend term g(t) is expressed as

[0032] g(t) = (k + a(t) T δ)t + (m + a(t) T Y)

[0033] where k represents the growth rate, a(t) represents the number of changes in the mutation points before time t, δ represents the change amount of the growth rate, m is the offset parameter, T is the transpose symbol, and δ and γ are fitness;

[0034] S3012: Based on the Fourier series, use the Prophet algorithm to establish a periodic term model. Then the periodic term s(t) is expressed as

[0035]

[0036] In the formula, N represents the number of cycles used in the model, P represents the period length of the expected time series, 2n represents the number of parameters to be estimated for the fitting period term, a n and b n are constants, which are obtained by fitting the periodic formula.

[0037] Furthermore, the LSTM neural network model among the groundwater level, rainfall, and slope periodic displacement in step S303 is as follows:

[0038] f t = σ(W fx x t + W fh h t-1 + b f )

[0039] i t = σ(W ix x t + W ih h t-1 + b i )

[0040] o t = σ(W ox x t + W oh h t-1 + b o )

[0041]

[0042]

[0043] h t = o t ⊙ tanh(C t )

[0044] In the formula, f t , i t , o t respectively represent the forgetting gate, input gate, and output gate; σ represents the sigmoid activation function; W fh , W ih , W oh , W Ch respectively represent the association weight matrices of the forgetting gate, input gate, output gate, and neuron state; W fx , W ix , W ox , W Cx respectively represent the input weight matrices of the forgetting gate, input gate, output gate, and neuron state; b f, b i , b o , b C represent the corresponding deviation terms; C t , C t-1 , respectively represent the state of the neuron at the current moment, the state of the neuron at the previous moment, and the temporary state of the neuron at the current moment; x t represents the input value at the current moment, that is, the groundwater level and rainfall; h t represents the output value of the neuron at the current moment, that is, the periodic displacement of the slope; ⊙ represents the Hadamard product of matrices.

[0045] Furthermore, the specific operations in step S4 include the following steps,

[0046] S401: Calculate the displacement difference between adjacent depths of the multi-depth displacement monitoring holes, organize the data into relative displacement time series data of adjacent depths, and draw the relative displacement time series curve of adjacent depths;

[0047] S402: Respectively set the relative displacement risk thresholds for adjacent depths of low risk, medium risk, and high risk, analyze the development trend of the relative displacement of adjacent depths over time, determine the risk level, and evaluate the depth range where landslides may occur and the landslide risk points;

[0048] S403: Connect the positions of multiple landslide risk points to form a slope sliding surface, and calculate the sliding volume according to the slope sliding surface.

[0049] Furthermore, the specific operations in step S5 include the following steps,

[0050] S501: Establish a factor set, an evaluation set, an evaluation matrix, and a factor weight vector for the comprehensive evaluation of the slope;

[0051] S502: Adopt the fuzzy comprehensive evaluation method to establish a comprehensive evaluation model of the slope, and calculate the risk evaluation result and the total score of the system.

[0052] Furthermore, the specific operations in step S501 include the following steps,

[0053] S5011: Establish a factor set including three influence indicators: the degree of slope displacement, the rate of slope displacement, and the degree of risk of the sliding risk point;

[0054] S5012: Establish an evaluation set including three evaluation indicators: low risk, medium risk, and high risk;

[0055] S5013: Conduct single-factor fuzzy evaluation, and establish a single-factor judgment matrix according to the membership degree of the elements in the factor set to the elements in the evaluation set;

[0056] S5014: Establish a fuzzy set of the weight sets of each factor, i.e., the factor weight vector, according to the importance of the slope displacement degree, the slope displacement rate, and the risk degree of the sliding risk point to the slope stability.

[0057] Further, the specific operations of step S502 include the following steps

[0058] S5021: Establish a slope fuzzy evaluation model through fuzzy transformation for the single-factor judgment matrix determined in S5013 and the factor weight vector determined in S5014 to obtain the risk evaluation result;

[0059] S5022: Determine the grade points of each factor in the evaluation set obtained in S5012 according to the slope risk classification level, and calculate the total system score.

[0060] The beneficial effects of the present invention are:

[0061] 1. The slope risk early warning and sliding volume prediction method based on Prophet-LSTM in the present invention can perform real-time dynamic evaluation and early warning of slope risks considering the influence of environmental factors such as rainfall and groundwater level. First, based on the Internet of Things technology, the slope displacement, groundwater level, and rainfall at different depths are monitored and data is collected in real time. Then, preprocessing operations such as eliminating outliers and denoising are performed on the collected slope monitoring data. The Prophet algorithm is used to decompose the time series into a periodic term and a trend term. Based on the Prophet-LSTM algorithm, a slope displacement periodic prediction model and a trend prediction model are established to predict the slope displacement, and real-time early warning is performed on the slope displacement value and the predicted value. Then, using the displacement monitoring data at different depths of the slope, the depth of the potential sliding surface of the slope is analyzed, the positions of multiple dangerous points are connected to form a slope sliding surface, and the sliding volume is calculated. Finally, a multi-parameter early warning framework for the slope is established to perform a comprehensive evaluation of multiple indicators of the slope. The proposed method can estimate the volume and location of potential landslides, so as to determine emergency measures according to the risk loss assessment, which helps to quickly and effectively identify landslide risks based on on-site monitoring, improve the understanding of landslide mechanisms, and slope risk management.

[0062] 2. A set of monitoring data preprocessing processes are proposed in the present invention. The 3σ criterion is used to delete outliers from the original monitoring data, and the discrete wavelet transform (DWT) is used to denoise the data. This method makes the monitoring data time series curve smooth and orderly, can improve the performance of subsequent prediction algorithms, and effectively solves the problem of poor performance of prediction methods caused by abnormal monitoring data.

[0063] 3. In the present invention, the Prophet algorithm is used to decompose the time series into a periodic term and a trend term, and based on the Prophet-LSTM algorithm, a periodic term prediction model and a trend term prediction model are established to predict the slope displacement. This method integrates the advantages of the Prophet algorithm and the LSTM algorithm, overcomes the disadvantages of the Prophet algorithm such as single variable, low prediction accuracy, and the LSTM algorithm such as long training time and poor generalization ability for non-periodic data, making the new method have the advantages of time series decomposition, real-time dynamic prediction, multi-variable input and output, and high prediction accuracy.

[0064] 4. In the present invention, the slope sliding surface is determined by analyzing the displacement mutation data at different depths, the volume of the sliding body is calculated, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the risk level of the slope monitoring points, and the risk evaluation results and the total system scores are given, providing a direct theoretical basis for quickly identifying landslide risks, taking protective measures in a timely manner, and avoiding large-scale landslides. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is the overall flowchart of the slope risk early warning and sliding volume prediction method based on Prophet-LSTM of the present invention;

[0066] Figure 2 is the comparison chart of the original monitoring data, the data after removing outliers by 3σ, and the data after DWT denoising in Embodiment 1 of the present invention;

[0067] Figure 3 is the operation flowchart of denoising processing using DWT in Embodiment 1 of the present invention;

[0068] Figure 4 is the result chart of decomposing the time series into a periodic term and a trend term using Prophet in Embodiment 1 of the present invention;

[0069] Figure 5 is the analysis result chart of the displacement change mutation points in Embodiment 1 of the present invention;

[0070] Figure 6 is the trend term prediction result chart in Embodiment 1 of the present invention;

[0071] Figure 7 is the periodic term prediction result chart in Embodiment 1 of the present invention.

[0072] Figure 8 is the slope displacement prediction result chart obtained by adding the trend term and the periodic term in Embodiment 1 of the present invention;

[0073] Figure 9 is the slope sliding surface chart formed by connecting the positions of multiple landslide risk points in Embodiment 1 of the present invention;

[0074] Figure 10 This is a graph showing the performance comparison results of various prediction methods for slope displacement in the simulation experiment of the present invention. Detailed implementation manners

[0075] In order to enable ordinary technicians in the field to better understand the technical solution of the present invention, the technical solution of the present invention will be further described below in conjunction with the drawings and embodiments.

[0076] Embodiment 1:

[0077] As shown in the attached Figure 1 figures, a slope risk early warning and sliding volume prediction method based on Prophet-LSTM includes the following steps:

[0078] S1: Collect slope monitoring data;

[0079] Specifically, the slope monitoring data includes slope displacements at different depths, groundwater levels, and rainfall amounts. Through an automatic monitoring platform for multi-depth displacements, groundwater levels, and rainfall amounts, real-time monitoring of slope displacements at different depths, groundwater levels, and rainfall amounts can be achieved based on Internet of Things technology. It should be noted that the automatic monitoring platform is an existing technology and can be directly used for real-time monitoring of slope displacements at different depths, groundwater levels, and rainfall amounts. The slope monitoring data in this application specifically adopts the slope monitoring data of the Shaojiazhuang accumulation body slope in Kaili, Guizhou, including rainfall amounts, groundwater levels, and displacement data at different depths.

[0080] Further, S2: Preprocess the slope monitoring data collected in step S1;

[0081] Specifically, S201: Adopt the 3σ criterion to delete outliers from the original slope monitoring data collected in step S1;

[0082] Due to external interference or short-term instrument failures, etc., there are many abnormal data points in the original monitoring data collected by the automatic monitoring instrument, far exceeding the reasonable data range. As shown in the attached Figure 2 figures, there are many discrete points in the original slope displacement data collected, and they exist in the form of mutations. If the original monitoring data is used for prediction, it will lead to problems with poor performance of some prediction algorithms. In order to improve the algorithm performance, it is necessary to delete outliers from the original monitoring data. In the present invention, the 3σ criterion is adopted to delete outliers from the original monitoring data.

[0083] The 3σ criterion is used to process sample data that follows a normal or approximately normal distribution and has a sufficiently large number of measurements. For a set of original monitored slope displacement data, assuming it only contains random errors, it is calculated and processed to obtain the standard deviation, and an interval is determined based on a certain probability. When the absolute value of the residual error of the monitored value in the data exceeds the interval error, the monitored value is determined as an invalid value and removed. Usually, the error of ±3σ is taken as the limit error, and the probability that the valid values are distributed within the range of (μ - 3σ, μ + 3σ) is 0.9974, and the possibility of exceeding this range is less than 0.3%.

[0084] S202: Use the discrete wavelet transform method (DWT) to denoise the slope monitoring data after removing outliers, making the displacement time series curve smooth and orderly;

[0085] After removing outliers using the 3σ criterion, there are still large fluctuations in the monitoring data. This phenomenon belongs to the weak noise during the monitoring process. As shown in the appendix Figure 2 This has a great interference on the establishment of subsequent prediction models. Therefore, it is necessary to use discrete wavelet transform to perform secondary smoothing processing on the monitoring data to remove more noise. The comparison of the original monitoring data, the data after removing outliers using 3σ, and the data after DWT denoising is shown in the appendix Figure 2 as shown

[0086] In step S202, the operation process of using the discrete wavelet transform method (DWT) to denoise the slope monitoring data after removing outliers is as shown in the appendix Figure 3 as shown, and specifically includes the following steps

[0087] S2021: Use the empirical mode decomposition method (EMD) to decompose the noise signal of the slope monitoring data into high-frequency intrinsic mode functions and low-frequency intrinsic mode functions;

[0088] S2022: Perform wavelet decomposition on the high-frequency intrinsic mode functions obtained in S2021 to obtain high-frequency wavelet coefficients and low-frequency wavelet coefficients;

[0089] S2023: Directly retain the numerical values of the low-frequency wavelet coefficients obtained in S2022; set a threshold and perform non-linear processing on the high-frequency wavelet coefficients to suppress the noise signal;

[0090] S2024: Perform the inverse operation of wavelet transform on the high-frequency wavelet coefficients after threshold processing in S2023 and the low-frequency wavelet coefficients whose original values are maintained to obtain the wavelet reconstruction signal;

[0091] S2025: Reconstruct the signal again for the low-frequency intrinsic mode functions obtained in S2021 and the wavelet reconstruction signal after wavelet reconstruction in S2024 to obtain the denoised signal.

[0092] Further, S3: Based on the Prophet-LSTM algorithm, establish a displacement period prediction model and a trend prediction model to predict the slope displacement, and conduct real-time early warning on the slope displacement value and the slope prediction value;

[0093] Specifically, S301: Use the Prophet algorithm to decompose the time series of slope monitoring data into a periodic term and a trend term

[0094] y(t) = g(t) + s(t) + ∈ t

[0095] where y(t) is the time series of slope monitoring data, g(t) is the trend term, s(t) is the periodic term, and ∈ t is the error term, which follows a normal distribution;

[0096] Considering the interference of rainfall and groundwater level on slope deformation, extract the periodic displacement caused by rainfall and groundwater level with 365d as the prior condition. The decomposition result is as shown in the appendix Figure 4 shown. From the appendix Figure 4 it can be seen that the trend term and periodic term extracted by Prophet can intuitively reflect the trend and periodic fluctuations of the displacement time series relative to the original time series.

[0097] More specifically, S3011: Based on the mutation point theory, use Prophet to automatically identify the mutation points in the time series, and use a piecewise linear model (linear growth) to fit the deformation time series between two adjacent mutation points to extract the trend term. Then the trend term g(t) is expressed as

[0098] g(t) = (k + a(t) T δ)t + (m + a(t) T γ)

[0099] where k represents the growth rate, a(t) represents the number of changes in the mutation points before time t, δ represents the change amount of the growth rate, m is the offset parameter, T is the transpose symbol, and δ and γ are fitness;

[0100] Based on the mutation point theory, identify the potential mutation points within the entire monitoring data range, and mark the points with a change rate greater than or equal to 0.5mm / d as mutation points. The analysis result of the displacement change mutation points is as shown in the appendix Figure 5 shown. The appendix Figure 5 shows the relationship curve between the mutation points and the trend term. The black curve is the cumulative curve of the time series displacement after preprocessing, and the focus of the virtual vertical line and the time axis represents the number of mutations.

[0101] S3012: Based on the Fourier series, use the Prophet algorithm to establish a periodic term model. Then the periodic term s(t) is expressed as

[0102]

[0103] In the formula, N represents the number of cycles used in the model, P represents the cycle length of the expected time series, 2n represents the number of parameters to be estimated for the fitting cycle term, a n and b n are constants, which are obtained by fitting the periodic formula.

[0104] S302: Predict the trend term using the Prophet algorithm;

[0105] The trend term reflects the future displacement change trend and has a guiding role in slope risk early warning. Prophet can achieve the dynamic decomposition of real-time data to ensure the real-time update of the trend term. According to the cumulative curve of slope displacement monitoring data, during the excavation process, the cumulative curve of slope displacement time series data mostly shows an upward trend and is a non-saturated curve. Therefore, based on the Prophet algorithm, a multi-level linear fitting method is used to fit the change curve of the trend term, and the trend term displacement values at different time periods can be predicted. The prediction results of the trend term are as shown Figure 6 below, and the prediction results are basically consistent with the trend direction and the original trend term.

[0106] S303: Predict the cycle term using the LSTM algorithm, establish an LSTM neural network model between the groundwater level, rainfall, and slope periodic displacement, and train it;

[0107] The LSTM neural network model between the groundwater level, rainfall, and slope periodic displacement is:

[0108] f t = σ(W fx x t + W fh h t-1 + b f )

[0109] i t = σ(W ix x t + W ih h t-1 + b i )

[0110] o t = σ(W ox x t + W oh h t-1 + b o )

[0111]

[0112]

[0113] h t = o t ⊙tanh(C t )

[0114] In the formula, f t , i t , o t respectively represent the forget gate, input gate, and output gate; σ represents the sigmoid activation function; W fh , W ih , W oh , W Ch respectively represent the associated weight matrices of the forget gate, input gate, output gate, and neuron state; W fx , W ix , W ox , W Cx respectively represent the input weight matrices of the forget gate, input gate, output gate, and neuron state; b f , b i , b o , b C represent the corresponding bias terms; C t , C t-1 , respectively represent the neuron state at the current moment, the neuron state at the previous moment, and the temporary neuron state at the current moment; x t represents the input value at the current moment, that is, the groundwater level and rainfall; h t represents the output value of the neuron at the current moment, that is, the periodic displacement of the slope; ⊙ represents the Hadamard product (or element-wise product) of matrices.

[0115] LSTM is a commonly used algorithm for time series data prediction, suitable for periodic data, and has the advantages of high prediction accuracy, multi-variable input and output. Therefore, the LSTM algorithm is used to predict the periodic term curve. The time series water level, time series rainfall data, and periodic displacement are used as input items for prediction, the correlation relationship between rainfall, water level data, and the periodic term is established, and a periodic term prediction model is established. The prediction result of the periodic term is as shown in the appendix Figure 7 As shown, the prediction result curve is basically consistent with the original data curve, showing perfect prediction performance.

[0116] S304: Add the results of the trend term and the periodic term to obtain the predicted value of the slope displacement, and the prediction curve is as shown in the appendix Figure 8 As shown;

[0117] S305: According to expert experience, respectively set the slope displacement risk thresholds for low risk, medium risk, and high risk, and conduct real-time early warning of the slope displacement value and the predicted value;

[0118] Slope displacement is a commonly used indicator for slope risk assessment. When the displacement reaches the threshold, risk early warning will be carried out. The slope displacement risk early warning level is divided into three stages: low risk (within 19 mm), medium risk (19 mm - 25 mm), and high risk (above 25 mm). Using Internet of Things technology, real-time monitoring of slope displacement is carried out; using the Prophet-LSTM real-time dynamic displacement prediction model established in the above steps to predict slope displacement, and realizing real-time early warning of slope displacement values and slope prediction values.

[0119] S306: According to the creep theory of rock and soil masses, the slope movement mode is divided into the initial deformation stage, uniform deformation stage, and accelerated deformation stage. The risk thresholds of slope displacement rates in each stage are set respectively, and real-time early warning of slope displacement rates is carried out.

[0120] Furthermore, S4: Using the displacement monitoring data at different depths of the slope, analyze the depth of the potential slip surface of the slope, connect the positions of multiple dangerous points to form a slope slip surface, and calculate the sliding volume;

[0121] Specifically, S401: Calculate the displacement difference between adjacent depths of the multi-depth displacement monitoring holes, organize the data into relative displacement time series data of adjacent depths, and draw the relative displacement time series curve of adjacent depths;

[0122] S402: Respectively set the relative displacement risk thresholds of adjacent depths for low risk, medium risk, and high risk, analyze the development trend of the relative displacement of adjacent depths over time, determine the risk level, and evaluate the depth range where landslides may occur and the landslide risk points;

[0123] S403: Connect the positions of multiple landslide risk points to form a slope slip surface, calculate the sliding volume according to the slope slip surface, and the slope slip surface is as shown in the appendix Figure 9 as follows.

[0124] Furthermore, S5: Establish a multi-parameter early warning framework for the slope and conduct a comprehensive evaluation of multiple indicators for the slope.

[0125] Specifically, S501: Establish a factor set, evaluation set, evaluation matrix, and factor weight vector for slope comprehensive evaluation;

[0126] S502: Adopt the fuzzy comprehensive evaluation method to establish a slope comprehensive evaluation model, and calculate the risk evaluation result and the total system score.

[0127] Among them, the specific operations of step S501 include the following steps,

[0128] S5011: Establish a factor set including three influence indicators of slope displacement degree, slope displacement rate, and risk degree of sliding risk points;

[0129] S5012: Establish an evaluation set that includes three evaluation indicators of low risk, medium risk, and high risk;

[0130] S5013: Conduct single-factor fuzzy evaluation. Based on the membership degree of the elements in the factor set to the elements in the evaluation set, establish a single-factor judgment matrix;

[0131] S5014: According to the importance of the slope displacement degree, slope displacement rate, and the risk degree of the sliding risk point to the slope stability, establish a fuzzy set of the weight set of each factor, that is, the factor weight vector.

[0132] The specific operations of step S502 include the following steps.

[0133] S5021: For the single-factor judgment matrix determined by S5013 and the factor weight vector determined by S5014, establish a slope fuzzy evaluation model through fuzzy transformation to obtain the risk evaluation result;

[0134] S5022: According to the slope risk classification level, determine the grade points of each factor obtained from S5012, and calculate the total system score.

[0135] Simulation experiment:

[0136] Verify the Prophet-LSTM algorithm proposed in the present invention through engineering verification. The data for engineering verification comes from the monitoring data of the Shaojiazhuang accumulation body slope in Kaili, Guizhou, including rainfall, groundwater level, and displacement data at different depths. The specific verification method is as follows: Without considering rainfall and groundwater level, use the LSTM algorithm, SVR algorithm, and Prophet algorithm respectively to predict the slope displacement. When considering rainfall and groundwater level, use the LSTM algorithm and SVR algorithm respectively to predict the slope displacement. Finally, compare and analyze the results with the results of predicting the slope displacement using the Prophet-LSTM algorithm proposed in the present invention. The results are as shown in the appendix Figure 10 shown. In the appendix Figure 10 , the data on the left side of the virtual vertical line is the training set, and the data on the right side is the prediction results of various algorithms. The models with the "_rw" suffix consider the influence of rainfall and groundwater level. The appendix Figure 10 shows that without considering rainfall and groundwater level, the LSTM algorithm has a high prediction accuracy, which also verifies the disadvantage of the poor generalization ability of the LSTM algorithm for non-periodic data. Therefore, when considering environmental factors, the Prophet-LSTM algorithm proposed in the present invention has good prediction performance.

[0137] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM, characterized in that, It includes the following steps, S1: Collect slope monitoring data; S2: Preprocess the slope monitoring data collected in step S1; S3: Based on the Prophet-LSTM algorithm, establish a displacement period prediction model and a trend prediction model to predict the slope displacement, and conduct real-time early warning on the slope displacement value and the slope prediction value; S4: Use the displacement monitoring data at different depths of the slope to analyze the depth of the potential sliding surface of the slope, connect the positions of multiple dangerous points to form a slope sliding surface, and calculate the sliding volume; S5: Establish a multi-parameter early warning framework for the slope to conduct a comprehensive evaluation of multiple indicators of the slope; Among them, the specific operations of step S3 include the following steps, S301: Use the Prophet algorithm to decompose the time series of slope monitoring data into a periodic term and a trend term In the formula, is the time series of slope monitoring data, is the trend term, is the periodic term, is the error term, which follows a normal distribution; S302: Use the Prophet algorithm to predict the trend term; S303: Use the LSTM algorithm to predict the periodic term, and establish an LSTM neural network model between the groundwater level, rainfall, and slope periodic displacement; S304: Add the results of the trend term and the periodic term to obtain the predicted value of the slope displacement; S305: According to expert experience, respectively set the slope displacement risk thresholds for low risk, medium risk, and high risk, and conduct real-time early warning on the slope displacement value and the predicted value; S306: According to the creep theory of rock and soil masses, divide the slope movement mode into an initial deformation stage, a uniform deformation stage, and an accelerated deformation stage, respectively set the slope displacement rate risk thresholds for each stage, and conduct real-time early warning on the slope displacement rate.

2. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 1, characterized in that The slope monitoring data described in step S1 includes slope displacements at different depths, groundwater levels, and rainfall.

3. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 2, characterized in that, The specific operations of step S2 include, S201: Use the 3σ criterion to delete outliers from the original slope monitoring data collected in step S1; S202: Use the discrete wavelet transform method to denoise the slope monitoring data after deleting outliers, making the displacement time series curve smooth and orderly.

4. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 3, characterized in that The specific operations of step S202 include the following steps, S2021: Use the empirical mode decomposition method to decompose the noise signal of the slope monitoring data into high-frequency intrinsic mode functions and low-frequency intrinsic mode functions; S2022: Conduct wavelet decomposition on the high-frequency intrinsic mode functions obtained in S2021 to obtain high-frequency wavelet coefficients and low-frequency wavelet coefficients; S2023: Directly retain the numerical values of the low-frequency wavelet coefficients obtained in S2022; set a threshold to perform non-linear processing on the high-frequency wavelet coefficients to suppress the noise signal; S2024: Perform the inverse operation of wavelet transform on the high-frequency wavelet coefficients after threshold processing in S2023 and the low-frequency wavelet coefficients that maintain their original values to obtain a wavelet reconstruction signal; S2025: Reconstruct the signal again for the low-frequency intrinsic mode functions obtained in S2021 and the wavelet reconstruction signal after wavelet reconstruction in S2024 to obtain a denoised signal.

5. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 4, characterized in that, The specific operations of step S301 include the following steps, S3011: Based on the mutation point theory, Prophet is used to automatically identify the mutation points in the time series. A piecewise linear model is used to fit the deformation time series between two adjacent mutation points, and the trend term is extracted. Then the trend term is expressed as In the formula, k represents the growth rate, represents t the number of times the mutation point changes before the moment, represents the change in the growth rate, m is the offset parameter, T is the transpose symbol, and are the fitness; S3012: Based on the Fourier series, use the Prophet algorithm to establish a periodic term model, and the periodic term is expressed as Wherein, N represents the number of cycles used in the model, P represents the period length of the expected time series, 2n represents the number of parameters to be estimated for fitting the periodic term, and are constants, which are obtained by fitting the periodic formula.

6. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 5, characterized in that, The LSTM neural network model between the groundwater level, rainfall, and slope periodic displacement in step S303 is: Wherein, respectively represent the forget gate, the input gate, and the output gate; represents the sigmoid activation function; respectively represent the associated weight matrices of the forget gate, the input gate, the output gate, and the neural unit state; respectively represent the input weight matrices of the forget gate, the input gate, the output gate, and the neural unit state; represents the corresponding bias term; respectively represent the neural unit state at the current moment, the neural unit state at the previous moment, and the temporary neural unit state at the current moment; represents the input value at the current moment, that is, the groundwater level and rainfall; represents the output value of the neural unit at the current moment, that is, the periodic displacement of the slope; represents the Hadamard product of the matrices.

7. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 5, characterized in that The specific operations of step S4 include the following steps, S401: Calculate the displacement difference between adjacent depths of multi-depth displacement monitoring holes, organize the data into relative displacement time series data of adjacent depths, and plot the relative displacement time series curve of adjacent depths; S402: Respectively set the relative displacement risk thresholds for adjacent depths of low risk, medium risk, and high risk, analyze the development trend of the relative displacement of adjacent depths over time, determine the risk level, and evaluate the depth range where landslides may occur and the landslide risk points; S403: Connect the positions of multiple landslide risk points to form a slope sliding surface, and calculate the sliding volume according to the slope sliding surface.

8. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 7, characterized in that The specific operations of step S5 include the following steps, S501: Establish a factor set, evaluation set, evaluation matrix, and factor weight vector for slope comprehensive evaluation; S502: Adopt the fuzzy comprehensive evaluation method to establish a slope comprehensive evaluation model, and calculate the risk evaluation result and the total system score.

9. A slope risk early warning and sliding volume prediction method based on Prophet-LSTM according to claim 8, characterized in that, The specific operations of step S501 include the following steps, S5011: Establish a factor set including three influence indicators: the degree of slope displacement, the rate of slope displacement, and the degree of risk of sliding risk points; S5012: Establish an evaluation set including three evaluation indicators: low risk, medium risk, and high risk; S5013: Conduct single-factor fuzzy evaluation, and establish a single-factor judgment matrix according to the membership degree of the elements in the factor set to the elements in the evaluation set; S5014: According to the importance of the degree of slope displacement, the rate of slope displacement, and the degree of risk of sliding risk points to slope stability, establish a fuzzy set of the weight set of each factor, that is, the factor weight vector.

Images