Model prediction method and system for slope deformation displacement considering environmental characteristic factors

Abstract

The present invention discloses a method and system for predicting the slope deformation displacement amount considering environmental characteristic factors. The method includes: S1, constructing a historical sample set of slope deformation in the study area; S2, processing by a convolutional neural network model to obtain feature vectors; S3, training the LSTM model for model learning according to time series; S4, collecting future slope environmental monitoring data in the study area at time series and inputting them into the convolutional neural network model to obtain feature vectors, and the LSTM model outputs the preliminary predicted values of slope displacement at time series based on the feature vectors #imgabs0# S5, constructing a linear regression model to fit and obtain the fitted slope deformation displacement amount y t Taking the difference with the preliminary predicted value of slope displacement #imgabs1# to obtain a residual sequence, and using a sliding window to take the average value of the preliminary predicted values of slope displacement before and after to fill the deleted data; S6, optimizing and fitting according to time series by using an SG filter. The present invention can obtain high-precision slope displacement prediction data, providing technical support for the accurate prediction of slope deformation displacement amount.

Description

Technical Field

[0001] The present invention relates to the field of slope deformation monitoring, and particularly to a method and system for predicting the slope deformation displacement model considering environmental characteristic factors. Background Technique

[0002] The prediction of slope SAR displacement can identify potential unstable areas of slopes in advance, help decision-makers take timely measures such as reinforcement and dredging to avoid the occurrence of sudden geological disasters. Especially in high-risk areas such as open-pit mining areas and infrastructure construction areas, the prediction of slope SAR displacement is more critical; however, there are many drawbacks in the previous slope SAR displacement prediction technologies. The past slope SAR displacement prediction methods mainly relied on single-factor analysis, ignoring the complexity of multi-factor interaction, resulting in low accuracy of prediction results. In addition, although many traditional machine learning methods have begun to be applied to slope SAR displacement prediction, they usually adopt a single-layer machine learning mode and do not consider the characteristic differences between different influencing factors, resulting in the underestimation or neglect of the influence of some key environmental factors, thus affecting the prediction accuracy. At the same time, the previous slope SAR displacement prediction did not further optimize the prediction, lacked effective prediction refinement measures, and could not effectively detect and remove outliers in the preliminary prediction, which made the prediction results often be interfered by data quality, affecting the stability and reliability of the model. Summary of the Invention

[0003] The purpose of the present invention is to solve the technical problems pointed out in the background technique, and provide a method and system for predicting the slope deformation displacement model considering environmental characteristic factors. The method extracts and identifies the characteristics of slope environmental monitoring data through a convolutional neural network model, and uses the long short-term memory network to fuse the characteristics of the long-term dependence of time-series data. The difference between the data results of the LSTM model and the linear regression model is further used to obtain the deviation degree, and the outliers are judged through the deviation degree to realize the elimination and filling of outliers, which is beneficial to eliminating the interference of outliers on the prediction results. Finally, the preliminary predicted value of the slope displacement after filling is optimized and fitted by the SG filter according to the time series, and high-precision slope displacement prediction data can be obtained.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A method for predicting the slope deformation displacement model considering environmental characteristic factors, the method comprising:

[0006] S1. Construct a historical sample set of slope deformation in the study area, which includes slope environmental monitoring data and slope deformation displacement and is associated in time series;

[0007] S2. The convolutional neural network model extracts the feature map of environmental feature factors from the slope environmental monitoring data in the slope deformation history sample set and flattens it into a feature vector through a fully connected layer;

[0008] S3. The LSTM model conducts model learning and training on the feature vector and the slope deformation displacement amount according to the time sequence;

[0009] S4. Collect the slope environmental monitoring data of the study area in the future according to the time sequence and input it into the convolutional neural network model to obtain a feature vector. The LSTM model outputs the preliminary prediction value of the slope displacement according to the time sequence based on the feature vector

[0010] S5. Construct a linear regression model with time as the independent variable and the slope deformation displacement amount as the dependent variable for regression fitting to obtain the fitted slope deformation displacement amount y t , and use the fitted slope deformation displacement amount y t and the preliminary prediction value of the slope displacement to calculate the difference to obtain a residual sequence, and calculate the deviation degree Z of the residual sequence t , and use the deviation degree Z t to eliminate the preliminary prediction value of the slope displacement whose absolute value is greater than 2, and use a sliding window to take the average value of the preliminary prediction values of the slope displacement before and after to fill the eliminated data;

[0011] S6. Use the SG filter to optimize and fit the curve of the preliminary prediction value of the slope displacement after filling according to the time sequence to obtain the optimized and fitted slope displacement prediction curve and slope displacement prediction data.

[0012] To better implement the present invention, the linear regression fitting expression of the linear regression model is as follows:

[0013] y t = β0 + β1t + ε t , where y t is the fitted slope deformation displacement amount of the linear regression, t is the continuous time including history and future in the study area, β0 and β1 are regression coefficients, and ε t is the residual term;

[0014] The deviation degree Z t has the following expression:

[0015] where Z t represents the deviation of time t, μ represents the mean of the residual sequence, σ represents the standard deviation of the residual sequence, and r t represents the residual of time t in the residual sequence.

[0016] Preferably, the LSTM model includes a forget gate, an input gate, a cell state, and an output gate. The forget gate outputs a real number vector between 0 and 1 through the sigmoid function, and the expression is as follows:

[0017] f = σ(W * [h t-1 , x t + b), where σ is the activation function, W and b represent the weight and the bias term respectively, and [h t-1 , x t represents the concatenation of the cell state of the (t - 1)-th layer and the input data of the t-th layer, and h t-1 represents the cell state of the (t - 1)-th layer, that is, the hidden state of the (t - 1)-th layer;

[0018] The expression of the input gate is as follows:

[0019] i t = σ(W xi x t + W hi h t-1 + b i

[0020]

[0021] x t is the input data, W hi , W xi , W xg , W hg represent the weights respectively, b i , b g represent the bias values respectively, tanh is the activation function, and i t is the information passed into the cell state, is the content to be updated; The output gate of the LSTM model outputs the preliminary predicted value of the slope displacement The expression is as follows:

[0022] Where W OUT is the weight of the output layer, b OUT is the bias term, and h t represents the hidden state of the t-th layer.

[0023] Preferably, the slope environmental monitoring data includes surface environmental temperature data, rainfall data, atmospheric refractive index data, atmospheric pressure data, and atmospheric relative humidity data, and the environmental characteristic factors include surface environmental temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor; The slope environmental monitoring data in the slope deformation historical sample set and the future slope environmental monitoring data in the study area are both normalized as follows:

[0024] P is the characteristic data of the slope environment monitoring data, P min is the minimum value of the characteristic data of the slope environment monitoring data, P max is the maximum value of the characteristic data of the slope environment monitoring data, and P is the normalized value of the characteristic data of the slope environment monitoring data.

[0025] Preferably, the filling method for the preliminary prediction value of the slope displacement after elimination is as follows: Use a sliding window to take out the preliminary prediction values of the slope displacement before and after the position of the eliminated preliminary prediction value of the slope displacement and calculate the average value, and fill the average value to the position of the eliminated preliminary prediction value of the slope displacement.

[0026] Preferably, in step S2, the convolutional neural network model outputs the feature map Z through the convolutional layer, Z = W * A + B, where W represents the convolutional kernel, which is used to slide and extract the features of all environmental feature factors from the input data A; in the convolutional operation, the convolutional kernel W performs an element-wise multiplication and summation operation with the input data A, and B is the bias term; the input data A is the slope environment monitoring data, and the sequence matrix expression of the input data A is as follows: where A n，m represents the m-th feature of the n-th sample, N is the total number of samples, and M is the total number of features; the feature map Z is processed by pooling to obtain the feature map Z′, and the feature map Z′ is the feature map of the environmental feature factors.

[0027] Preferably, the SG filter uses a sliding window to optimize and fit using the polynomial least squares method, and the expression is as follows:

[0028] where y 最终 represents the slope displacement prediction data after SG filtering, represents the preliminary prediction value of the slope displacement after filling, c λ represents the coefficient obtained by fitting the polynomial using the least squares method, λ1 represents the data points on one side of the sliding window boundary centered on the sliding window, and λ2 represents the data points on the other side of the sliding window boundary centered on the sliding window.

[0029] Preferably, the residual expression of time t in the residual sequence is as follows:

[0030] A slope deformation displacement model prediction system considering environmental characteristic factors, including a slope deformation historical sample set, a data acquisition module, a convolutional neural network model, an LSTM model, and a linear regression model. The slope deformation historical sample set includes slope environmental monitoring data and slope deformation displacement amounts associated in time series; the convolutional neural network model extracts feature maps of environmental characteristic factors from the slope environmental monitoring data in the slope deformation historical sample set and flattens them into feature vectors through a fully connected layer; the LSTM model performs model learning and training on the feature vectors and slope deformation displacement amounts in time series; the data acquisition module is used to collect future slope environmental monitoring data in time series in the study area, input it into the convolutional neural network model to obtain feature vectors, and the LSTM model outputs preliminary slope displacement prediction values in time series based on the feature vectors. The linear regression model constructs a regression fit with time as the independent variable and slope deformation displacement amount as the dependent variable to obtain the fitted slope deformation displacement amount y. t , the fitted slope deformation displacement amount y t and the preliminary slope displacement prediction value are subtracted to obtain a residual sequence, and the deviation degree Z of the residual sequence is calculated. t , the deviation degree Z t The preliminary slope displacement prediction values with an absolute value greater than 2 are removed, and the average values of the preliminary slope displacement prediction values before and after are taken using a sliding window to fill the removed data; the linear regression model includes an SG filter, and the linear regression model uses the SG filter to perform curve optimization fitting on the filled preliminary slope displacement prediction values in time series to obtain an optimized fitting slope displacement prediction curve and slope displacement prediction data.

[0031] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0032] (1) The present invention uses a convolutional neural network model to extract and identify features from slope environmental monitoring data, and uses a long short-term memory network to fuse the features of long-term dependence of time series data. The difference between the data results of the LSTM model and the linear regression model is further used to obtain the deviation degree, and the outliers are judged through the deviation degree to realize the removal and filling of outliers, which is beneficial to eliminating the interference of outliers on the prediction results. Finally, the SG filter is used to perform curve optimization fitting on the filled preliminary slope displacement prediction values in time series, and high-precision slope displacement prediction data can be obtained, which not only effectively enhances the stability and accuracy of the model, but also provides a solid technical guarantee for the accurate identification of the change trend of slope deformation displacement amounts, and has important practical value and forward-looking significance in the application of slope monitoring.

[0033] (2) The present invention fully considers various factors such as environmental characteristic factors and temporal characteristics, extracts comprehensive features through a convolutional neural network model, and can accurately predict the preliminary prediction data of slope displacement through the learning and training of the LSTM model. It also adopts rigorous outlier processing and missing value filling processing, and then introduces SG filtering to optimize the data. The present invention adopts a refined and rigorous prediction method for the slope deformation displacement amount model, significantly improving the data quality of the time series of the slope deformation displacement amount, thereby greatly improving the accuracy and stability of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a structural schematic diagram of the present invention;

[0035] Figure 2 is a schematic diagram of the process principle of the embodiment;

[0036] Figure 3 is a schematic diagram of the time series in which the slope deformation historical sample set in the embodiment is associated with the environmental characteristic factors and the slope deformation displacement amount in chronological order;

[0037] Figure 4 is the preliminary prediction data of slope displacement in the study area of the embodiment, including future time series and in chronological order;

[0038] Figure 5 is the optimized prediction data of slope displacement after outlier removal and sliding filling in the embodiment;

[0039] Figure 6 is the prediction data of slope displacement after filtering, fusion, optimization and fitting in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The present invention will be further described in detail below with reference to the embodiments:

[0041] Embodiment

[0042] As Figure 1 shown, a prediction method for a slope deformation displacement amount model considering environmental characteristic factors includes:

[0043] S1. Construct a slope deformation historical sample set in the study area that includes slope environmental monitoring data and slope deformation displacement amount and is associated in chronological order. As Figure 2 shown, the slope environmental monitoring data in the slope deformation historical sample set of this embodiment includes surface environmental temperature data, rainfall data, atmospheric refractive index data, atmospheric pressure data, and atmospheric relative humidity data. The slope environmental monitoring data in the slope deformation historical sample set is normalized to eliminate the scale difference between data with different dimensions, facilitating the extraction and identification of subsequent environmental characteristic factors.

[0044] For example, take an open-pit coal mine in Xinjiang as the research area. The monitoring radar is IBIS360ArcSAR (used to detect the slope deformation displacement of the measuring points). The slope deformation historical sample set is based on obtaining 300 hours of historical data hourly. For better principle illustration, the environmental characteristic factors in the model include, for example, the surface environmental temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor. The slope environmental monitoring data can be multi-source data. The convolutional neural network model obtains the time-series data of environmental characteristic factors and slope deformation displacement according to the time series (the total time series is 300 hours) based on the slope environmental monitoring data. For example Figure 3 as shown

[0045] S2. The convolutional neural network model extracts the feature map of environmental characteristic factors from the slope environmental monitoring data of the slope deformation historical sample set and flattens it into a feature vector through the fully connected layer. As Figure 2 shown, extract the features that are associated with the slope deformation displacement and have high importance from the slope environmental monitoring data (such as environmental characteristic factors including the surface environmental temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor). These features are flattened into a feature vector according to the time series associated with the slope deformation displacement for model training, facilitating the subsequent LSTM model to output the preliminary prediction value of the slope displacement according to the time series based on the feature vector

[0046] In some embodiments, the convolutional neural network model outputs the feature map Z through the convolutional layer, Z = W * A + B, where W represents the convolutional kernel, which is used to slide and extract the features of all environmental characteristic factors from the input data A. In the convolutional operation (the convolution can be set to 3x3), the convolutional kernel W performs an element-wise multiplication and summation operation with the input data A, and B is the bias term. The input data A is the slope environmental monitoring data, and the sequence matrix expression of the input data A is as follows where A n，m represents the m-th feature of the n-th sample. N is the total number of samples (for example, N slope environmental monitoring data and N slope deformation displacements associated with the research area are collected hourly. N is the total amount of sample data collected hourly. In the case of selecting an open-pit coal mine in Xinjiang in the research area, N is 300), and M is the total number of features (the total number of features including the surface environmental temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor, etc.). The feature map Z (after the convolutional operation, the generated feature map Z contains a large amount of feature information, which significantly increases the data dimension and is beneficial to the significant improvement of feature extraction according to the time series) is obtained after pooling processing (reducing the size of the feature map and simultaneously retaining the most critical information related to the prediction task, which can effectively reduce the computational amount and improve the processing efficiency) to obtain the feature map Z'. The feature map Z' is the feature map of environmental characteristic factors

[0047] S3. The LSTM model performs model learning and training on the feature vectors and slope deformation displacement amounts in chronological order.

[0048] S4. Collect the slope environment monitoring data of the study area in the future in chronological order (in the case of selecting an open-pit coal mine in Xinjiang in the study area, the initial sample size is 300, and the future chronological order is 200 immediately after. In this way, the total number of historical samples and future chronological order is 500). Input the data into the convolutional neural network model to obtain feature vectors, and the LSTM model outputs the preliminary predicted values of the slope displacement in chronological order based on the feature vectors.

[0049] In some embodiments, the slope environment monitoring data includes surface environment temperature data, rainfall data, atmospheric refractive index data, atmospheric pressure data, and atmospheric relative humidity data. The environmental characteristic factors include surface environment temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor. The slope environment monitoring data of the slope deformation historical sample set and the future slope environment monitoring data of the study area are both normalized as follows:

[0050] P is the characteristic data of the slope environment monitoring data, P min is the minimum value of the characteristic data of the slope environment monitoring data, P max is the maximum value of the characteristic data of the slope environment monitoring data, and P' is the normalized value of the characteristic data of the slope environment monitoring data. Taking the surface environment temperature data as an example, if the surface environment temperature is 27°C, the maximum value of the surface temperature characteristic is 33°C, and the minimum value of the surface temperature characteristic is 17°C, then the normalization of the surface temperature of 27°C is:

[0051] In this embodiment, the LSTM model includes a forget gate, an input gate, a cell state (or called an update memory unit), and an output gate. The forget gate outputs a real number vector between 0 and 1 through the sigmoid function, and the expression is as follows:

[0052] f = σ(W * [h t-1 , x t + b), where σ is the activation function, W and b represent the weight and the bias term respectively, [h t-1 , x t represents the concatenation of the cell state of the (t - 1)-th layer and the input data of the t-th layer, and h t-1 represents the cell state of the (t - 1)-th layer, that is, the hidden state of the (t - 1)-th layer.

[0053] The expression of the input gate is as follows:

[0054] i t = σ(W xi x t + Whi h t-1 +b i

[0055]

[0056] x t is the input data, W hi 、W xi 、W xg 、W hg represent weights respectively, b i 、b g represent bias values respectively, tanh is the activation function, i t is the information passed into the cell state, is the updated content. The output gate of the LSTM model outputs the preliminary predicted value of the slope displacement The expression is as follows:

[0057] where W OUT is the weight of the output layer, b OUT is the bias term, h t represents the hidden state of the t-th layer. Update the memory cell, and the memory cell is updated by combining the forget gate f t and the input gate i t ; the new memory cell state combines the historical information controlled by the forget gate and the current information controlled by the input gate, and finally affects the prediction of the time series of the slope radar displacement volume, C t =f t *C t-1 +i t *C t , the output gate determines that the final information extracted from the memory cell will be output as the current hidden state h t output

[0058] o t =σ(W o *[h t-1 , F t +b o )

[0059] where, o t is the activation of the input gate, W o is the weight matrix of the input gate, b o is the bias term.

[0060] Calculate the hidden state, and the final hidden state h t is jointly determined by the state of the memory cell and the control of the output gate, h t =o t *tanh(C t ).

[0061] The LSTM model first uses the slope deformation historical sample set for model training to obtain the trained LSTM model. The trained LSTM model can obtain the eigenvector prediction output of the preliminary prediction value of the slope displacement (in time series) based on the environmental characteristic factors in the future of the study area in time series. In the case of selecting an open-pit coal mine in Xinjiang in the study area, the slope deformation historical sample set (including 300 hourly data, of which 250 are used as the training set and 50 are used as the prediction verification set) is used to predict the slope displacement prediction data for the next 200 hours, and the preliminary prediction data of the slope displacement including future time series and in time series is obtained, such as Figure 4 shown.

[0062] S5. Construct a linear regression model with time as the independent variable and slope deformation displacement as the dependent variable for regression fitting to obtain the fitted slope deformation displacement y t , in some embodiments, the linear regression fitting expression of the linear regression model is as follows:

[0063] y t =β0 + β1t + ε t , where y t is the fitted slope deformation displacement of the linear regression, t is the continuous time including history and future in the study area, β0 and β1 are regression coefficients, and ε t is the residual term.

[0064] Subtract the fitted slope deformation displacement y t from the preliminary prediction value of the slope displacement to obtain the residual sequence. The residual expression of time t in the residual sequence is as follows: Calculate the deviation degree Z t of the residual sequence. In some embodiments, the deviation degree Z t expression is as follows:

[0065] where Z t represents the deviation of time t, μ represents the mean of the residual sequence, σ represents the standard deviation of the residual sequence, and r t represents the residual of time t in the residual sequence.

[0066] Remove the preliminary prediction value of the slope displacement whose absolute value of the deviation degree Z t is greater than 2 (the absolute value of the deviation degree Z t greater than 2 is the deviation degree Z tIf it is greater than 2 or less than -2, the preliminary prediction value of slope displacement with an absolute deviation degree greater than 2 is determined as an outlier. First, the outlier is removed, and the removed data is considered as a missing value. The average value of the preliminary prediction values of slope displacement before and after is taken using a sliding window to fill the removed data. In some embodiments, the method for filling the preliminary prediction value of slope displacement after removal is as follows: The preliminary prediction values of slope displacement before and after the position of the removed preliminary prediction value of slope displacement are taken using a sliding window and averaged, and the average value is filled into the position of the removed preliminary prediction value of slope displacement (i.e., filling the missing value). In the case of selecting a certain open-pit coal mine in Xinjiang in the study area, using the slope deformation historical sample set (including 300 hourly data, 250 of which are used as the training set and 50 as the prediction verification set) to predict the slope displacement prediction data for the next 200 hours, the optimized prediction data of slope displacement after outlier removal and sliding filling including future time series is obtained, as Figure 5 shown.

[0067] S6. The SG filter is used to optimize the curve fitting of the preliminary prediction value of slope displacement after filling in time series, and the optimized prediction curve and prediction data of slope displacement are obtained. In the case of selecting a certain open-pit coal mine in Xinjiang in the study area, using the slope deformation historical sample set (including 300 hourly data, 250 of which are used as the training set and 50 as the prediction verification set) to predict the slope displacement prediction data for the next 200 hours, the final slope displacement prediction data including future time series and in time series is obtained, as Figure 6 shown.

[0068] In some embodiments, the SG filter of the present invention uses a sliding window to optimize the fitting using the polynomial least squares method, and the expression is as follows:

[0069] where y 最终 represents the predicted slope displacement data after SG filtering (the weighted sum of the data points within the sliding window with weighted coefficients), represents the preliminary predicted value of slope displacement after filling, and c λdenotes the coefficients obtained by least squares fitting of the polynomial (i.e., the weights for smoothing the target points). λ1 represents the data points on one side of the sliding window boundary centered on the sliding window, and λ2 represents the data points on the other side of the sliding window boundary centered on the sliding window. λ1 and λ2 are the positions of the left and right endpoints of the data within the sliding window. In the present invention, a convolutional neural network model is used to extract and identify features from slope environment monitoring data (extract and identify environmental impact factors), and the long short-term memory network is used to fuse the features of long-term dependence of time series data. The difference between the data results of the LSTM model and the linear regression model is further used to obtain the deviation degree, and the outliers are judged through the deviation degree to realize the elimination and filling of outliers, which is beneficial to eliminating the interference of outliers on the prediction results. Finally, the preliminary predicted value of the slope displacement after filling is optimized and fitted by the SG filter according to the time series, and high-precision slope displacement prediction data can be obtained, which not only effectively enhances the stability and accuracy of the model, but also provides a solid technical guarantee for the accurate identification of the change trend of the slope deformation displacement amount, and has important practical value and forward-looking significance in the field of slope monitoring.

[0070] A slope deformation displacement model prediction system considering environmental characteristic factors includes a slope deformation historical sample set, a data acquisition module, a convolutional neural network model, an LSTM model, and a linear regression model. The slope deformation historical sample set includes slope environment monitoring data and slope deformation displacement amounts associated in time series. The convolutional neural network model extracts the feature map of environmental characteristic factors from the slope environment monitoring data in the slope deformation historical sample set and flattens it into a feature vector through a fully connected layer. The LSTM model performs model learning and training on the feature vector and the slope deformation displacement amount according to the time series. The data acquisition module is used to collect the future slope environment monitoring data in the study area in time series and input it into the convolutional neural network model to obtain a feature vector, and the LSTM model outputs the preliminary predicted value of the slope displacement in time series based on the feature vector. The linear regression model constructs and performs regression fitting with time as the independent variable and slope deformation displacement amount as the dependent variable to obtain the fitted slope deformation displacement amount y. t , the fitted slope deformation displacement amount y t and the preliminary predicted value of the slope displacement are subtracted to obtain a residual sequence, and the deviation degree Z of the residual sequence is calculated. t , the preliminary predicted value of the slope displacement with the absolute value of the deviation degree Z t greater than 2 is eliminated, and the average value of the preliminary predicted values of the slope displacement before and after is taken by using a sliding window to fill the eliminated data. The linear regression model includes an SG filter, and the linear regression model uses the SG filter to optimize and fit the curve of the preliminary predicted value of the slope displacement after filling according to the time series to obtain the optimized fitted slope displacement prediction curve and slope displacement prediction data.

[0071] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A slope deformation displacement model prediction method considering environmental characteristic factors, characterized in that: The methods include: S1. Construct a slope deformation history sample set in the study area that includes slope environmental monitoring data, slope deformation and displacement, and is associated in time series; S2, the convolutional neural network model extracts the feature map of environmental characteristic factors from the slope environmental monitoring data of the slope deformation history sample set and flattens it into a feature vector through a fully connected layer; S3, LSTM model performs model learning and training on the feature vector and slope deformation displacement according to the time sequence; S4. Collect the slope environmental monitoring data of the study area in the future according to the time sequence and input it into the convolutional neural network model to obtain the feature vector. The LSTM model outputs the preliminary predicted value of the slope displacement according to the time sequence based on the feature vector S5. Construct a linear regression model with time as the independent variable and the slope deformation displacement as the dependent variable for regression fitting to obtain the fitted slope deformation displacement y t , subtract the fitted slope deformation displacement y t from the preliminary prediction value of the slope displacement to obtain the residual sequence, and calculate the deviation degree Z of the residual sequence t . Exclude the preliminary prediction value of the slope displacement whose absolute value of the deviation degree Z t is greater than 2, and use a sliding window to take the average value of the preliminary prediction values of the slope displacement before and after to fill the excluded data; S6. Using the SG filter to perform curve optimization fitting on the preliminary predicted value of the slope displacement after filling in a time series manner, to obtain the slope displacement prediction curve and slope displacement prediction data after optimization fitting.

2. The method for predicting the slope deformation displacement amount model considering environmental characteristic factors according to claim 1, wherein: The linear regression fitting expression of the linear regression model is as follows: y t = β0 + β1t + ε t , where y t is the fitted slope deformation displacement of the linear regression, t is the continuous time including history and future in the study area, β0 and β1 are regression coefficients, and ε t is the residual term; The deviation degree Z t The expression is as follows: where Z t represents the deviation of time t, μ represents the mean of the residual sequence, σ represents the standard deviation of the residual sequence, and r t represents the residual of time t in the residual sequence.

3. The method for predicting the slope deformation displacement amount model considering environmental characteristic factors according to claim 1, wherein: The LSTM model includes a forget gate, an input gate, a cell state, and an output gate. The forget gate outputs a real number vector between 0 and 1 through a sigmoid function. The expression is as follows: f = σ(W * [h t-1 , x t + b), where σ is the activation function, and W and b represent the weight and bias term respectively, [h t-1 , x t represents the concatenation of the unit state of the (t - 1)-th layer and the input data of the t-th layer, and h t-1 represents the unit state of the (t - 1)-th layer, that is, the hidden state of the (t - 1)-th layer; The input gate expression is as follows: i t = σ(W xi x t + W hi h t-1 + b i x t is the input data, W hi , W xi , W xg , W hg represent weights respectively, b i , b g represent bias values respectively, tanh is the activation function, i t is the information passed into the cell state, is the updated content; the output gate of the LSTM model outputs the preliminary predicted value of the slope displacement The expression is as follows: where W OUT is the weight of the output layer, b OUT is the bias term, and h t represents the hidden state of the t-th layer.

4. The method for predicting the slope deformation displacement amount model considering environmental characteristic factors according to claim 1, wherein: The slope environmental monitoring data includes surface environmental temperature data, rainfall data, atmospheric refractive index data, atmospheric pressure data, and atmospheric relative humidity data. The environmental characteristic factors include surface environmental temperature factor, rainfall factor, atmospheric refractive index factor, atmospheric pressure factor, and atmospheric relative humidity factor. The slope environmental monitoring data of the slope deformation history sample set and the future slope environmental monitoring data of the study area are both normalized as follows: P is the characteristic data of the slope environmental monitoring data, P min is the minimum value of the characteristic data of the slope environmental monitoring data P max is the maximum value of the characteristic data of the slope environmental monitoring data. P' is the normalized value of the characteristic data of the slope environmental monitoring data.

5. The slope deformation displacement model prediction method considering environmental characteristic factors according to claim 1, characterized in that: The method for filling the preliminary predicted value of slope displacement after elimination is as follows: use a sliding window to take out the preliminary predicted values of slope displacement before and after the position of the preliminary predicted value of slope displacement eliminated and calculate the average value, and fill the average value to the position of the preliminary predicted value of slope displacement eliminated.

6. The method for predicting the slope deformation displacement model considering environmental characteristic factors according to claim 1, characterized in that: In step S2, the convolutional neural network model outputs a feature map Z through the convolution layer, Z = W*A+B, where W represents the convolution kernel, which is used to slide the input data A to extract the features of all environmental feature factors; In the convolution operation, the convolution kernel W performs an element-wise multiplication and summation operation with the input data A, and B is the bias term; the input data A is slope environment monitoring data, and the sequence matrix expression of the input data A is as follows: where A n，m represents the m-th feature of the n-th sample, N is the total number of samples, and M is the total number of features; the feature map Z is processed by pooling to obtain the feature map Z′, and the feature map Z′ is the feature map of the environmental feature factor.

7. The method for predicting the slope deformation displacement model considering environmental characteristic factors according to claim 1, characterized in that: The SG filter uses a sliding window to optimize the fitting using the polynomial least squares method, and the expression is as follows: where y 最终 represents the slope displacement prediction data after SG filtering, represents the preliminary prediction value of the slope displacement after filling, c λ represents the coefficient obtained by least squares fitting of the polynomial, λ1 represents the data points on one side of the sliding window boundary centered on the sliding window, and λ2 represents the data points on the other side of the sliding window boundary centered on the sliding window.

8. The method for predicting the slope deformation displacement amount model considering environmental characteristic factors according to claim 2, wherein: The residual expression at time t in the residual sequence is as follows:

9. A slope deformation displacement model prediction system considering environmental characteristic factors, characterized in that: It includes a slope deformation history sample set, a data acquisition module, a convolutional neural network model, an LSTM model, and a linear regression model. The slope deformation history sample set includes slope environmental monitoring data and slope deformation displacement amounts associated in time series. The convolutional neural network model extracts the feature map of environmental feature factors from the slope environmental monitoring data in the slope deformation history sample set and flattens it into a feature vector through a fully connected layer. The LSTM model performs model learning and training on the feature vector and the slope deformation displacement amounts according to the time series. The data acquisition module is used to collect the future slope environmental monitoring data in the study area in time series and input it into the convolutional neural network model to obtain a feature vector. The LSTM model outputs the preliminary slope displacement prediction values in time series based on the feature vector. The linear regression model is constructed with time as the independent variable and the slope deformation displacement amount as the dependent variable and performs regression fitting to obtain the fitted slope deformation displacement amount y. t , the fitted slope deformation displacement amount y t and the preliminary slope displacement prediction value are subtracted to obtain a residual sequence, and the deviation degree Z of the residual sequence is calculated. t , the preliminary slope displacement prediction values with the absolute value of the deviation degree Z t greater than 2 are removed, and the average values of the preliminary slope displacement prediction values before and after are taken using a sliding window to fill the removed data. The linear regression model includes an SG filter, and the linear regression model uses the SG filter to perform curve optimization fitting on the filled preliminary slope displacement prediction values in time series to obtain the optimized fitted slope displacement prediction curve and slope displacement prediction data.

Images