Slope Deformation Prediction and Explanation Method Based on Fuzzy Echo State Network and SHAP

Abstract

The embodiment of this specification provides a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP. The method includes: determining the key point cloud data of the slope area to be detected before the target moment, and extracting features from the key point cloud data to obtain a feature sequence set representing the change of feature values of at least one deformation influencing factor in time series; determining the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network; determining the deformation components corresponding to the at least one deformation influencing factor by using the SHAP formula according to the feature sequence set and the slope deformation prediction value, wherein the deformation components are used to explain the influence degree of each deformation influencing factor on the slope deformation. By the above method, the deformation components of each feature on the slope deformation prediction value are quantified, so as to determine the influence of each feature on the slope deformation and clarify the main factors causing the slope deformation.

Description

Technical Field

[0001] The embodiments of this specification relate to the technical field of geological disaster engineering, and particularly to a method for slope deformation prediction and explanation based on a fuzzy echo state network and SHAP. Background Art

[0002] Slope deformation monitoring is an important part of geological disaster prevention and control and construction project safety. With the development of computer technology and slope radar monitoring technology, high-precision real-time monitoring of small displacements of slopes has been achieved. Radar monitoring is a non-contact and continuous dynamic monitoring method that uses radar waves to detect small deformations of the surface and internal structures, and realizes real-time, efficient, and accurate monitoring of slope stability. With the deep integration of computer technology, signal processing algorithms, and Internet of Things technology, slope radar monitoring has entered the intelligent stage.

[0003] However, the prediction models in the existing technologies can only predict the deformation prediction values of slopes and do not have interpretability. Therefore, there is an urgent need for a method for slope deformation prediction and explanation based on a fuzzy echo state network and SHAP that can not only predict the deformation prediction values of slopes but also clearly explain the reasons for the deformation. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide a method for slope deformation prediction and explanation based on a fuzzy echo state network and SHAP. One or more embodiments of this specification also relate to a device for slope deformation prediction and explanation based on a fuzzy echo state network and SHAP, a computing device, a computer-readable storage medium, and a computer program to solve the technical defects existing in the existing technologies.

[0005] According to the first aspect of the embodiments of this specification, a method for slope deformation prediction and explanation based on a fuzzy echo state network and SHAP is provided, including:

[0006] Determine the key point cloud data of the slope area to be detected before the target moment, and extract features from the key point cloud data to obtain a feature sequence set, where the feature sequence set is used to characterize the change of feature values of at least one deformation influencing factor in time series;

[0007] Determine the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network;

[0008] Determine the deformation components corresponding to the at least one deformation influencing factor according to the feature sequence set, the slope deformation prediction value, and by using the SHAP formula, where the deformation components characterize the influence degree of each deformation influencing factor on the slope.

[0009] In one embodiment of the present disclosure, determining the key point cloud data of the slope area to be detected before the target moment includes:

[0010] Collecting the original point cloud data of the slope area to be detected before the target moment;

[0011] Performing radius filtering on the original point cloud data to obtain the preprocessed point cloud data set at the target moment;

[0012] Performing key point extraction on the preprocessed point cloud data set to obtain the key point cloud data.

[0013] In one embodiment of the present disclosure, performing key point extraction on the preprocessed point cloud data set to obtain the key point cloud data includes:

[0014] Determining a target data point, where the target data point is any point in the preprocessed point cloud data set;

[0015] Determining a neighborhood point set corresponding to the target data point based on a preset quota filtering radius;

[0016] Calculating the average position information of the neighborhood point set;

[0017] Calculating a covariance matrix corresponding to the target data point based on the average position information;

[0018] Calculating a response value corresponding to the target data point according to the covariance matrix;

[0019] If the response value is greater than or equal to a preset threshold, determining the first point cloud data as a key point.

[0020] In one embodiment of the present disclosure, performing feature extraction on the key point cloud data to obtain a feature sequence set includes:

[0021] Extracting at least one terrain feature information in the position dimension of the key point cloud data;

[0022] Extracting at least one environmental feature information in the feature dimension of the key point cloud data;

[0023] Determining a feature sequence set according to the terrain feature information and the environmental feature information.

[0024] In one embodiment of the present disclosure, the feature sequence set includes multi-source feature sequences at each moment before the target moment, and the fuzzy echo state network includes an input layer, a fuzzy layer, a reservoir, and an output layer;

[0025] Determining the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network includes:

[0026] Receive the set of feature sequences through the input layer and input them into the fuzzy layer;

[0027] Process each multi-source feature sequence in the set of feature sequences through at least one fuzzy rule in the fuzzy layer to obtain a clustering center vector and an influence radius vector corresponding to each moment; determine a weight factor corresponding to each fuzzy rule according to the clustering center vector and the influence radius vector;

[0028] Process the set of feature sequences through the reservoir based on the fuzzy rule to output a weight vector;

[0029] Predict the slope deformation prediction value at the target moment through the output layer based on the weight vector and the weight factor.

[0030] In an embodiment of the present disclosure, the fuzzy echo state network is trained in the following manner:

[0031] Obtain training samples and construct training sample pairs, where the training sample pairs include sample labels, and the sample labels include slope deformation amounts;

[0032] Input the training samples into the initial model to obtain slope deformation prediction values;

[0033] Determine a loss value based on the slope deformation prediction value and the slope deformation amount, and train the initial model based on the loss value until the training stop condition is reached to obtain the fuzzy echo state network.

[0034] In an embodiment of the present disclosure, the determining the deformation component corresponding to the at least one deformation influencing factor according to the set of feature sequences and the slope deformation prediction value by using the SHAP formula includes:

[0035] Wherein, the deformation component corresponding to each deformation influencing factor is specifically obtained in the following manner:

[0036] Determine a first set of feature sequence subsets and a second set of feature sequence subsets according to the set of feature sequences and the deformation influencing factor, where the first set of feature sequence subsets is a set of subsets of each deformation influencing factor of the deformation feature factors except the corresponding deformation influencing factor in the set of feature sequences, and the second set of feature sequence subsets is a set of feature subsets obtained by combining each deformation influencing factor;

[0037] Determine a third set of feature sequence subsets according to the second set of feature sequence subsets, where the third set of feature sequence subsets is a set of feature subsets including the corresponding deformation influencing factor;

[0038] Determine the deformation component corresponding to the deformation influence factor according to the first set of characteristic sequence subsets, the second set of characteristic sequence subsets, and the third set of characteristic sequence subsets.

[0039] In one embodiment of the present disclosure, according to the slope deformation prediction value and the deformation component, an analysis summary diagram of the deformation influence factor is generated.

[0040] According to the second aspect of the embodiments of the present specification, a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP is provided, including:

[0041] A data determination module configured to determine key point cloud data of the slope area to be detected before the target moment, and perform feature extraction on the key point cloud data to obtain a set of feature sequences, where the set of feature sequences is used to characterize the eigenvalue changes of at least one deformation influence factor in time series;

[0042] A slope deformation prediction value determination module configured to determine the slope deformation prediction value at the target moment according to the set of feature sequences and the fuzzy echo state network;

[0043] A deformation component determination module configured to determine the deformation component corresponding to the at least one deformation influence factor according to the set of feature sequences and the slope deformation prediction value by using the SHAP formula, where the deformation component is used to explain the influence degree of each deformation influence factor on the slope.

[0044] According to the third aspect of the embodiments of the present specification, a computing device is provided, including:

[0045] A memory and a processor;

[0046] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP are implemented.

[0047] According to the fourth aspect of the embodiments of the present specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP are implemented.

[0048] According to the fifth aspect of the embodiments of the present specification, a computer program is provided, where when the computer program is executed on a computer, the computer is made to execute the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP.

[0049] One embodiment of this specification realizes determining the key point cloud data of the slope area to be detected before the target moment, extracting features from the key point cloud data to obtain a feature sequence set representing the eigenvalue changes of at least one deformation influencing factor over time; determining the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network; and determining the deformation components corresponding to the at least one deformation influencing factor by using the SHAP formula according to the feature sequence set and the slope deformation prediction value, quantifying the deformation components of each feature on the slope deformation prediction value, thereby determining the influence of each feature on the slope deformation and clarifying the main factors causing the slope deformation. Description of the Drawings

[0050] Figure 1 is a flowchart of a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP provided by one embodiment of this specification;

[0051] Figure 2 is a structural diagram of a fuzzy echo state network provided by one embodiment of this specification;

[0052] Figure 3 is a schematic diagram of an analysis summary of deformation influencing factors generated by a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP provided by one embodiment of this specification;

[0053] Figure 4 is a flowchart of the processing process of a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP provided by one embodiment of this specification;

[0054] Figure 5 is a schematic structural diagram of a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP provided by one embodiment of this specification;

[0055] Figure 6 is a structural block diagram of a computing device provided by one embodiment of this specification. Detailed Embodiments

[0056] Many specific details are set forth in the following description in order to provide a thorough understanding of this specification. However, this specification can be implemented in many other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of this specification. Therefore, this specification is not limited by the specific embodiments disclosed below.

[0057] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a", "the", and "said" used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and encompasses any or all possible combinations of one or more of the associated listed items.

[0058] It should be understood that although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0059] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of the relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entrances are provided for the user to select authorization or rejection.

[0060] In this specification, a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP is provided. This specification also relates to a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP, a computing device, and a computer-readable storage medium, which will be described in detail one by one in the following embodiments.

[0061] See Figure 1 , Figure 1 which shows a flowchart of a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP according to an embodiment of this specification, specifically including the following steps.

[0062] Step 102: Determine the key point cloud data of the slope area to be detected before the target moment, and extract features from the key point cloud data to obtain a feature sequence set, where the feature sequence set is used to characterize the eigenvalue changes of at least one deformation influencing factor in time series.

[0063] Specifically, the slope area to be detected refers to any hillside or slope area that requires safety assessment, stability monitoring, and maintenance. Such areas may be formed by natural factors (such as natural slopes formed by mountains and river erosion) or human factors (such as artificial slopes like cut slopes, dams, and mine slopes formed by excavation and filling). The target moment specifically refers to a certain moment to be predicted. For example, if the slope deformation prediction values at two points are to be predicted, the key point cloud data before the two points needs to be determined. The key point cloud data is time-series data. The slope radar collects slope data to obtain a series of point cloud data, which is recorded in chronological order, such as the monitoring results every day, every hour, or even in real-time, forming a series of data points with time tags. Each set of data represents the monitoring index value at a certain moment, such as the position information of the data point.

[0064] Feature extraction specifically refers to the process of selecting, transforming, or constructing attributes or variables that are helpful for model training and prediction from the key point cloud data. Simply put, the purpose of feature extraction is to convert complex data forms into more meaningful forms that are conducive to model processing. The deformation of the slope is affected by various internal and external factors. The deformation influence factors specifically refer to the factors that affect the slope. The feature sequence set obtained by feature extraction from the key point cloud data contains key attributes or features that are helpful for prediction. Specifically, since the key point cloud data is a series of data recorded in chronological order, the feature sequence set obtained by feature extraction also includes multi-source feature sequences at each moment before the target moment recorded in chronological order. Each feature sequence contains the feature values of one deformation influence factor or multiple deformation influence factors.

[0065] Based on this, when predicting the slope deformation at a certain moment, the point cloud data before that moment is collected, the key point cloud data is determined, and the key feature data that is helpful for prediction in the key point cloud data is extracted. It should be noted that when the target moment is a certain moment, the point cloud data before the target moment can be data with a certain time interval from the target moment, which can be one minute, ten minutes, one hour, or even one day.

[0066] After collecting the point cloud data, the original point cloud data can be directly used as the key point cloud data for feature extraction. However, considering that the collected original point cloud data is large and complex, and limited by the software and hardware resources of the computing device, it may be infeasible on the computing device and may also result in low efficiency when predicting slope deformation subsequently. Therefore, when determining the key point cloud data, the original point cloud data collected by the slope radar needs to be denoised and then key points are extracted. In the embodiments of the present disclosure, the determination of the key point cloud data of the slope area to be detected before the target moment includes:

[0067] Collect the original point cloud data of the slope area to be detected before the target moment;

[0068] Perform radius filtering on the original point cloud data to obtain a preprocessed point cloud data set before the target moment;

[0069] Extract key points from the preprocessed point cloud data set to obtain the key point cloud data.

[0070] Specifically, after the acquisition device collects the data of the slope area to be detected, a series of data reflecting the change of the slope over time are obtained, which are the original point cloud data. The original point cloud data includes at least position information. The acquisition device can be one or several of a slope radar, a drone or a satellite, a sensor installed on the slope, and a 3D laser scanner. This application does not make any restrictions.

[0071] The radius filtering process specifically refers to a denoising process for point cloud data. By performing radius filtering, outlier points can be filtered out. Specifically, when implementing, a certain point in the original point cloud data is determined as the target data point, and the target data range is determined according to the target data point and the preset filtering radius; determine whether the number of data points within the target data range reaches the preset threshold; if it does not reach the preset threshold, then remove the target data point; if it reaches the preset threshold, then no processing is performed. It should be noted that the selection of the filtering radius should be cautious. If it is too large, important detail information may be lost, and if it is too small, noise may not be effectively suppressed. Therefore, in practical applications, it is often necessary to flexibly adjust the filtering parameters and optimize the filtering effect in combination with the actual situation and data characteristics.

[0072] Specifically, when implementing, taking the slope radar as an example, use the slope radar to collect the original point cloud data of any slope at M moments before the target moment and perform radius filtering to obtain M preprocessed point cloud data sets P = {P b (m)|m = 1, 2,..., M; b = 1, 2,..., B}, where P b (m) represents the b-th data point at the m-th moment after preprocessing, and the b-th data point at the m-th moment contains denotes P b (m)'s position information, and x m,b , y m,b , z m,b denotes the three-dimensional coordinate values of P b (m), denotes P b (m)'s feature information; B represents the total number of data points at the m-th moment after preprocessing.

[0073] Key point extraction specifically refers to identifying and selecting feature points that are crucial for analysis and decision-making from the monitoring data. Considering that key point extraction has a crucial impact on the subsequent prediction results, therefore, to ensure the effectiveness of the point cloud data, the average curvature of the target data points is used as a measurement index to calculate the response value of the data points and obtain the key points. Specifically, in the embodiments of the present disclosure, the key point extraction of the preprocessed point cloud data set to obtain the key point cloud data includes:

[0074] Determine the target data point, where the target data point is any point in the preprocessed point cloud data set;

[0075] Based on a preset quota filtering radius, determine the set of neighborhood points corresponding to the target data point;

[0076] Calculate the average position information of the set of neighborhood points;

[0077] Based on the average position information, calculate the covariance matrix corresponding to the target data point;

[0078] Calculate the response value corresponding to the target data point according to the covariance matrix;

[0079] If the response value is greater than or equal to the preset threshold, determine the first point cloud data as the key point.

[0080] Specifically, the target data point is any point in the preprocessed point cloud data set, and any point in the preprocessed point cloud data set can be used as the target data point for calculation, and the target average curvature is used as the response value to extract the key points.

[0081] When specifically implemented, the key points can be extracted through the following steps:

[0082] Step 1.1: According to the set quota filtering radius r, use KD-Tree to search for the set of neighborhood points {P b (m)|1 ≤ τ ≤ μ} of the b-th data point P b,τ (m) at the m-th moment, where P b,τ (m) represents the τ-th neighborhood point of the b-th data point P b (m), and let represent the position information of P b,τ (m), and x m,b , y m,b , z m,b represent the three-dimensional coordinate values of P b (m); μ is the number of neighborhood points;

[0083] Step 1.2: Use Equation (1) to calculate the average position of the set of neighborhood points of the b-th data point P b (m)

[0084]

[0085] Step 1.3: Calculate the covariance matrix G of the b-th data point P(m) at the m-th moment according to Equation (2). b (m) m,b :

[0086]

[0087] Step 1.3.1: Calculate the three eigenvalues λ m,b of the covariance matrix G m,b,1 , λ m,b,2 , λ m,b,3 , so as to obtain the average curvature of the b-th data point P b (m).

[0088] Step 1.4: Calculate the response value V b of the b-th data point P m,b (m) according to Equation (3):

[0089]

[0090] Step 1.5: Set a threshold V T for the response value of the key point, and only keep the b-th data point P m,b whose V t is greater than V b as the key point, so as to obtain the key point cloud data P′(m) at the m-th moment, and further obtain the key point cloud data P′ of M moments. It should be noted that if the calculated response value V

[0091] is less than or equal to V m,b , then this point is excluded. T

[0092] In summary, the above embodiments improve the estimation accuracy of the real surface structure by considering neighborhood information instead of individual isolated points, and can effectively extract the key geometric features in the point cloud data, only keep the key points, reduce the amount of data required for subsequent processing, improve the calculation efficiency, and at the same time ensure the relative stability of the key points in different time periods, which is beneficial to the analysis and tracking of dynamic scenes.

[0093] Considering that blindly performing feature extraction cannot meet the subsequent analysis of the slope deformation prediction value, the key point cloud data can be extracted for features purposefully, and the key point cloud data is extracted for features from at least one dimension. Specifically, in the embodiments of the present disclosure, the extraction of features from the key point cloud data to obtain a feature sequence set includes: ​

[0094] Extract at least one type of terrain feature information in the position dimension of the key point cloud data;

[0095] Extract at least one type of environmental feature information in the feature dimension of the key point cloud data;

[0096] Determine a feature sequence set according to the terrain feature information and the environmental feature information.

[0097] Specifically, feature extraction can be performed on the key point cloud data from two dimensions, namely the position dimension and the feature dimension. Further, there can be multiple sub-dimensions under each dimension. For example, the position dimension can be further divided into sub-dimensions such as terrain and landform sub-dimension, human activity sub-dimension, and the feature dimension can be divided into sub-dimensions such as environmental condition sub-dimension, hydrological condition sub-dimension, etc. The specific dimensions can be determined according to the actual situation. The terrain feature information represents the terrain and landform of the slope, such as slope displacement, and the environmental feature information represents the environment of the slope, such as rainfall, groundwater and other information. Specifically, reference can be made to Table 1.

[0098] During specific implementation, feature extraction can be performed through the following steps:

[0099] Step 2.1: Extract n1 types of terrain features that affect landslides from the position information {P′ a (m)|m = 1, 2, …, M} of the slope radar key point cloud data P′(m).

[0100] Step 2.2: Extract n2 types of environmental features that affect landslides from the feature information {P′ f (m)|m = 1, 2, …, M} of the slope radar key point cloud data P′(m).

[0101] Step 2.3: After combining the n1 types of terrain features and the n2 types of environmental features, obtain the slope multi-source time series C at M moments containing N features that affect the slope deformation amount.

[0102] Among them, the position information and the feature information are the position dimension and the feature dimension respectively, the terrain feature is the terrain feature information, the environmental feature is the environmental feature information, and the obtained slope multi-source time series C is denoted as the feature sequence set.

[0103] Table 1 Features Affecting the Slope Deformation Amount

[0104]

[0105]

[0106] In summary, through feature extraction of the original point cloud data in the embodiments of this specification, while reducing the subsequent data processing volume, the most representative points and effective data are extracted, thereby reducing the prediction time and prediction difficulty of the subsequent model.

[0107] Step 104: Determine the predicted value of the slope deformation at the target moment according to the feature sequence set and the fuzzy echo state network.

[0108] Specifically, the problem of predicting the radar deformation of a slope is to predict the slope deformation in a future time period based on historical data, providing a basis for subsequent explanation of the causes of landslides. Since the preprocessed slope radar point cloud data contains complex non-linear relationships and factors such as its distribution, feature extraction, and preprocessing may all affect the modeling accuracy of the ESN, in order to reduce the network training calculation amount, improve the non-linear mapping ability and modeling accuracy of the model, a method combining the TS fuzzy model and the ESN is used to predict the radar deformation value of the slope, and a fuzzy echo state network is constructed. The predicted value of the slope deformation specifically refers to the deformation amount of the slope.

[0109] Based on this, input the feature sequence set into the fuzzy echo state network, and the trained fuzzy echo state network predicts the predicted value of the slope deformation.

[0110] Considering that the data in the feature sequence set contains complex non-linear relationships and its data distribution and feature extraction and other processes will all affect the data prediction results, therefore, in this embodiment, by inputting the feature sequence set into the fuzzy echo state network, the feature sequence set is processed under fuzzy rules to output a weight vector, and then the predicted value of the slope deformation at the target moment is predicted according to the output weight vector. The specific implementation steps are as follows: The feature sequence set includes multi-source feature sequences at each moment before the target moment, and the model structure diagram of the fuzzy echo state network is as Figure 2 shown including an input layer, a fuzzy layer, a reservoir, and an output layer;

[0111] The determining the predicted value of the slope deformation at the target moment according to the feature sequence set and the fuzzy echo state network includes:

[0112] Receive the feature sequence set through the input layer and input it into the fuzzy layer;

[0113] Process each multi-source feature sequence in the feature sequence set through at least one fuzzy rule in the fuzzy layer to obtain the clustering center vector and the influence radius vector corresponding to each moment; determine the weight factor corresponding to each fuzzy rule according to the clustering center vector and the influence radius vector;

[0114] Process the feature sequence set through the reservoir based on the fuzzy rule to output a weight vector;

[0115] Predict the predicted value of the slope deformation at the target moment through the output layer based on the weight vector and the weight factor.

[0116] Specifically, each multi-source feature sequence specifically refers to a feature sequence containing multiple different dimensions of information.

[0117] In specific implementation, in the embodiments of the present disclosure, considering that the distribution of the data itself and the processing of point cloud data and other situations will affect the accuracy of the slope prediction result, a fuzzy layer is integrated into the echo state network to obtain a fuzzy echo state network, which enhances the ability to model and solve uncertainty problems when processing data.

[0118] Considering the prediction accuracy of the model, training sample data can be pre-constructed to pre-train the fuzzy echo state network. In the embodiments of the present disclosure, the fuzzy echo state network can be pre-trained in the following manner:

[0119] Obtain training samples and construct training sample pairs, and the training sample pairs include sample labels, where the sample labels include slope deformation amounts;

[0120] Input the training samples into the initial model to obtain slope deformation prediction values;

[0121] Determine the loss value based on the slope deformation prediction value and the slope deformation amount, and train the initial model based on the loss value until the training stop condition is reached to obtain the fuzzy echo state network.

[0122] Among them, the slope deformation amount refers to the deformation amount of the slope obtained after a preset time interval, and the slope deformation prediction value refers to the slope deformation value at a certain moment after the training sample is input and the initial model outputs the training sample.

[0123] Specifically, the cross-entropy loss function can be calculated based on the slope deformation prediction value and the slope deformation amount included in the sample label to generate the loss value. Among them, the sample label refers to the true slope deformation amount, that is, the slope deformation amount included in the sample label is the true result, and when the training sample is input into the initial model, the output slope deformation prediction value is the prediction result. When the difference between the prediction result and the true result is small enough, it means that the prediction result is close enough to the true result. At this time, the initial model training is completed to obtain the fuzzy echo state network.

[0124] Before pre-training the initial model as shown in Figure 2 , the obtained training samples can be subjected to radius filtering processing, key point extraction, and feature extraction. The specific processing methods can be seen in the above processing and will not be elaborated here.

[0125] After feature extraction, the slope multi-source feature sequences C = {c(1), c(2),..., c(m),..., c(M)} of M moments containing N features affecting the slope deformation amount and the slope deformation amount S = {s1,..., s m ,..., s M}; where, c(m) represents the multi-source feature sequence of the slope at the m-th moment, and

[0126] c(m) = (c m,1 , c m,2 , …, c m,n , …, c m,N ) T ,

[0127] c m,n represents the eigenvalue of the n-th feature affecting the deformation of the slope at the m-th moment, s m represents the m-th slope deformation, N = n1 + n2, 1 ≤ n ≤ N.

[0128] Furthermore, the slope deformation s m of the point cloud data P′(m) at the m-th moment compared with the point cloud data P′(m - 1) at the (m - 1)-th moment is calculated using Equation (4), so as to obtain the slope deformations S = {s1, …, s m , …, s M} at M moments;

[0129]

[0130] In Equation (4): Δx m , Δy m , Δz m are the displacement change values of the point cloud data P′(m) at the m-th moment compared with the point cloud data P′(m - 1) at the (m - 1)-th moment in the X, Y, and Z directions, respectively.

[0131] The slope multi-source time series C and the slope deformations S are input into the input layer to construct training sample pairs. Among them, the extraction method of the multi-source time series is the same as the extraction of the feature sequence set described above, which is not elaborated in this application.

[0132] After obtaining the training sample pairs, they are input into the fuzzy layer, and then the fuzzy layer further processes the training sample pairs. The following Step 3 - 3.2.2 is the processing of the training sample pairs by the fuzzy layer.

[0133] Step 3: The fuzzy layer designs L fuzzy rules and uses them to calculate L normalized weight factors {β l (c(m))|1 ≤ l ≤ L} of c(m), where β l (c(m)) represents the normalized weight factor of c(m) in the l-th rule R l ;

[0134] Step 3.1: Use Equation (5) to design the l-th fuzzy rule R l , so as to obtain L fuzzification rules:

[0135]

[0136] In formula (5): represents c m,n on the l-th fuzzy rule R l fuzzy set, where 1 ≤ l ≤ L; (ESN) l represents the l-th echo state network;

[0137] Step 3.2: Through the subtractive clustering algorithm, calculate the l corresponding normalized weight:

[0138] Step 3.2.1: Process c(m) on R l using the subtractive clustering algorithm to obtain the l-th cluster center vector at the m-th moment and the influence radius vector at the m-th moment, where represents c m,n l cluster centers, represents c m,n influence radii of l cluster centers of c, l = 1, 2, …, L, n = 1, 2, …, N;

[0139] Step 3.2.2: Determine the Gaussian form membership function of the fuzzy set

[0140]

[0141] Step 3.2.3: To improve the prediction accuracy and reduce the influence of noise and uncertainty in the data on the network, introduce the normalized weight factor β1(c(m)), and calculate the normalized weight factor β 1 of c(m) on the rule R l (c(m)) using formula (7):

[0142]

[0143] Furthermore, the reservoir processes the training sample pairs, and the specific steps are as follows:

[0144] Step 4: The reservoir under the fuzzy rule processes C to obtain the output weight vector of the echo state network where represents the output weight of the l-th echo state network (ESN) l ;

[0145] Step 4.1: Initialize the network parameters of the echo state network ESN, including: the connection weight matrix W from the input layer to the reservoir in, the connection weight matrix W of the internal units of the reservoir and the connection weight matrix W from the output layer back to the internal units of the reservoir back , where W = (W1,…,W1,…,W L ) T 、 where represents the connection weight matrix from the input layer to the reservoir of the l-th echo state network (ESN) l , W l represents the connection weight matrix of the internal units of the reservoir of the l-th echo state network (ESN) l , represents the connection weight matrix from the output layer back to the internal units of the reservoir of the l-th echo state network (ESN) l ;

[0146] Step 4.2: Randomly initialize the output weight matrix of the reservoir in the l-th echo state network (ESN) l Initialize m = 0; Define and initialize the internal state e of the l-th echo state network (ESN) at the m-th moment l e(m)=0; l

[0147] Step 4.3: Perform the forward propagation of the l-th echo state network (ESN) l , and use Equations (8) and (9) to calculate the internal state e l of the l-th echo state network (ESN) corresponding to rule R l at the (m + 1)-th moment and the output value h l (m) at the m-th moment: l

[0148]

[0149]

[0150] In Equations (8) and (9): g l (m) represents the state vector of the l-th echo state network (ESN) at the m-th moment, and l represents the output value of the l-th echo state network (ESN) at the (m - 1)-th moment, and T represents the transpose; l

[0151] Step 4.4: Start collecting the internal state and its output value of the (ESN) l from the m-th moment, and correspondingly obtain the internal state matrix E with dimensions (M - m + 1) × Nl = [e l (m), e l (m + 1), …, e l (M)] T and the output value matrix H with dimension (M - m + 1) × 1 l = [h l (m), h l (m + 1), …, h l (M)] T ;

[0152] Step 4.5: Calculate the output weight of the l-th echo state network (ESN) using Equation (10) l of the output weight so as to obtain the output weight vector (output weight) of the echo state network

[0153]

[0154] The output layer performs the next processing on the training sample pairs:

[0155] Step 4.6: When using ESN to solve the slope radar deformation prediction value, there is a defect that the robustness is relatively weak. Especially when facing large noise and uncertainties, the noise and interference will affect the stability and prediction performance of the network, which may lead to inaccurate prediction results of landslide deformation values. To improve the prediction accuracy and reduce the influence of noise and uncertainties in the data on the network, a normalized weight factor β l (c(m)) is introduced, and each linear relationship is "stuck" together in the way of "weighted average". The output layer predicts the slope deformation prediction value at the (M + 1)-th moment using Equation (11)

[0156]

[0157] In Equation (11): c(M + 1) represents the slope multi-source feature sequence at the (M + 1)-th moment, and c(M + 1) = h l (M), that is, the output value at the M-th moment is used as the slope multi-source feature sequence input at the (M + 1)-th moment, represents the state vector of the l-th echo state network (ESN) at the (M + 1)-th moment l of the state vector.

[0158] In this specification, by calculating the loss value, the difference between the prediction result and the true result of the model can be intuitively shown, and then the initial model is trained specifically, and the network parameters, that is, the output weight matrix parameters, can effectively improve the training rate and training effect of the model.

[0159] Among them, training the initial model based on the loss value until the training stop condition is reached may include:

[0160] Determine whether the loss value is less than a preset threshold;

[0161] If not, return to execute the above step of obtaining training samples and continue training;

[0162] If so, determine that the training stop condition is reached.

[0163] Among them, the preset threshold is the critical value of the loss value. When the loss value is greater than or equal to the preset threshold, it indicates that there is still a certain deviation between the prediction result of the initial model and the true result, and the network parameters of the initial model still need to be adjusted, and the training samples of this category need to be obtained to continue training the model; when the loss value is less than the preset threshold, it indicates that the closeness between the prediction result of the initial model and the true result is sufficient, and the training can be stopped. The value of the preset threshold can be determined according to the actual situation, and this specification does not limit it.

[0164] In this specification, the specific training situation of the initial model can be judged according to the loss value, and when the training is not qualified, the parameters of the initial model can be adjusted backward according to the loss value to improve the analysis ability of the model, with high training speed and good training effect.

[0165] Step 106: According to the feature sequence set and the slope deformation prediction value, use the SHAP formula to determine the deformation components corresponding to the at least one deformation influencing factor, where the deformation components are used to explain the influence degree of each deformation influencing factor on the slope deformation.

[0166] Specifically, the degrees of influence of different deformation influencing factors on the slope may be the same or different. They may have a negative impact, that is, the slope stability decreases and there may be a landslide risk, or they may have a positive impact, that is, the slope stability increases. The deformation component characterizes the influence degree of each deformation influencing factor on the slope. For example: a total of 10 deformation influencing factors are included, the deformation component corresponding to deformation influencing factor A is +2, then it is determined that deformation influencing factor A has a positive impact on the slope stability; the deformation component corresponding to deformation influencing factor B is -3, then it is determined that deformation influencing factor B has a negative impact on the slope stability; the deformation component corresponding to deformation influencing factor C is 0, then it is determined that deformation influencing factor C has no impact on the slope stability.

[0167] It should be noted that the representation methods of the deformation components corresponding to the deformation influencing factors may be different. This application only gives examples here, and those skilled in the art can determine the specific representation method according to their actual needs.

[0168] In specific implementation, after obtaining the predicted value of slope deformation, deformation components corresponding to various deformation influence factors are calculated based on the predicted value of slope deformation and the characteristic sequence set.

[0169] Considering that the deformation components accurately represent the influence of each deformation influence factor on the slope, therefore, the corresponding deformation components of the corresponding deformation influence factors can be calculated according to the characteristic sequence subset obtained by removing the corresponding deformation influence factor from the characteristic sequence set, the characteristic sequence subset containing the corresponding deformation influence characteristics, and the characteristic sequence set.

[0170] In specific implementation, in an embodiment of the present disclosure, the deformation components corresponding to each deformation influence factor are specifically obtained by the following method:

[0171] Determine a first set of characteristic sequence subsets and a second set of characteristic sequence subsets according to the characteristic sequence set and the deformation influence factor, where the first set of characteristic sequence subsets is a set of subsets of each deformation influence factor of the deformation characteristic factors except the corresponding deformation influence factor in the characteristic sequence set, and the second set of characteristic sequence subsets is a set obtained from the characteristic subsets combined by each deformation influence factor;

[0172] Determine a third set of characteristic sequence subsets according to the second set of characteristic sequence subsets, where the third set of characteristic sequence subsets is a set composed of characteristic subsets containing the corresponding deformation influence factor;

[0173] Determine the deformation components corresponding to the deformation influence factor according to the first set of characteristic sequence subsets, the second set of characteristic sequence subsets, and the third set of characteristic sequence subsets.

[0174] Specifically, after determining the first set of predicted values of slope deformation, the second set of predicted values of slope deformation, and the third set of characteristic sequence subsets, further determine the first set of predicted values of slope deformation corresponding to the first set of characteristic sequence subsets and the union of the first set of characteristic sequence subsets and the third set of characteristic sequence subsets; and determine the second set of predicted values of slope deformation corresponding to the union, and finally determine the deformation components corresponding to the deformation influence factor.

[0175] Furthermore, it can be obtained by the following formula:

[0176]

[0177] In formula (12): Q represents the set of all possible feature subsets obtained by removing the nth feature from C, which contains N deformation influence factors affecting the slope deformation amount, and then selecting 1 to N - 1 deformation influence factors for combination; F represents the set of all possible features obtained by selecting 1 to N deformation influence factors for combination from C, which contains N deformation influence factors affecting the slope deformation amount; |Q| represents the number of sets in Q; |F| represents the number of sets in F; □ n represents the set of feature subsets in F that contain the nth feature; represents the set of predicted values of the slope deformation amount obtained by processing Q through the fuzzy echo state network, represents Q ∪ ε n The set of predicted values of the slope deformation amount obtained by processing through the fuzzy echo state network. Further, if then it indicates that the nth feature helps to improve the deformation of the slope; if then it indicates that the nth feature inhibits the deformation of the slope. The deformation component corresponding to the deformation influence factor can be obtained by calculating the SHAP value of each deformation influence factor.

[0178] After obtaining the deformation components corresponding to each deformation influence factor, to ensure that users can more clearly understand the influence of each deformation influence factor on the slope deformation, it can be visualized. Specifically, in this embodiment, after determining the deformation components corresponding to the at least one deformation influence factor according to the feature sequence set and the slope deformation prediction value by using the SHAP formula, it includes:

[0179] Generate a summary diagram for analyzing the deformation influence factors according to the slope deformation prediction value and the deformation components.

[0180] Specifically, Figure 3 is a summary diagram for analyzing the deformation influence factors generated by a slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP provided in an embodiment of this specification. As Figure 3 shown, the features affecting the radar deformation prediction value of the slope can be sorted according to the deformation components to generate a summary diagram for analyzing the deformation influence factors. Among them, the x-axis is the deformation component, the y-axis is the name and characteristic value of the deformation influence factor, red indicates that the feature has a positive influence on the prediction, and blue indicates that the feature has a negative influence on the prediction. This diagram describes obtaining the predicted value of the slope deformation at the target moment by starting from the overall benchmark value of the model and adding (red) or subtracting (blue) the deformation components of each deformation influence factor.

[0181] After generating the summary diagram of the deformation influence factor analysis, the user can take corresponding preventive measures according to the influence degree (positive or negative) of each deformation influence factor on the slope deformation, allocate resources more reasonably and formulate relevant risk management strategies. For example, if it is determined that the deformation influence factor A has a negative impact on the slope deformation and the impact is relatively large, then when designing the slope reinforcement measures, the negative impact of the deformation influence factor A on the slope can be considered to be weakened.

[0182] After obtaining the deformation components corresponding to the deformation influence factors, it is possible to more clearly understand the influence degree of each deformation influence factor on the slope deformation prediction value. Similarly, it is also helpful to discover some important features, so that the feature extraction can be more accurate and targeted. At the same time, it can also find the defects of the model during training and make adjustments based on this, so that the model prediction efficiency can be higher and the performance of the model can be significantly improved.

[0183] The following combines the attached Figure 3 and Figure 4 , taking the application of the slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP provided in this specification in a certain county in Sichuan as an example, further illustrate the slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP. Among them, Figure 4 shows the processing procedure flowchart of a slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP provided in an embodiment of this specification, which specifically includes the following steps.

[0184] Step 402: Collect the original point cloud data of the slope area to be detected before the target time.

[0185] Step 404: Perform radius filtering on the original point cloud data to obtain the preprocessed point cloud data set at the target time.

[0186] Step 406: Extract key points from the preprocessed point cloud data set to obtain the key point cloud data.

[0187] Step 408: Determine the key point cloud data of the slope area to be detected before the target time, and extract features from the key point cloud data to obtain a feature sequence set, where the feature sequence set is used to characterize the eigenvalue changes of sixteen deformation influence factors in time series.

[0188] Among them, as Figure 3 shown, the sixteen deformation influence factors include: rainfall, normalized vegetation index, slope, distance from the fault, profile curvature, etc.

[0189] Step 410: Determine the slope deformation prediction value at the target time according to the feature sequence set and the fuzzy echo state network.

[0190] Step 412: According to the feature sequence set and the slope deformation prediction value, use the SHAP formula to determine the deformation components corresponding to the sixteen deformation influencing factors.

[0191] It can be seen from Figure 3 that 11 influencing factors such as rainfall, normalized vegetation index, and slope have a significant impact on the model output. Among them, as the rainfall, slope, distance from the fault, profile curvature, slope aspect, elevation, and peak ground motion characteristic value increase, the deformation component also increases, which has a positive impact on the slope deformation. The normalized vegetation index, engineering rock group, and land use have a negative impact on the slope deformation. For categorical features such as undulation and distance from the river, the feature value represents its category, and the impact on the model output cannot be judged by the change of the feature value, but it can be seen that the distribution has aggregation, indicating that specific types of values have a greater impact on the model.

[0192] Corresponding to the above method embodiments, this specification also provides an embodiment of a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP. Figure 5 FIG. shows a schematic structural diagram of a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP provided by an embodiment of this specification. As Figure 5 shown, the device includes:

[0193] A data determination module 502, configured to determine the key point cloud data of the slope area to be detected before the target moment, and perform feature extraction on the key point cloud data to obtain a feature sequence set, where the feature sequence set is used to characterize the eigenvalue changes of at least one deformation influencing factor in time series;

[0194] A slope deformation prediction value determination module 504, configured to determine the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network;

[0195] A deformation component determination module 506, configured to determine the deformation components corresponding to the at least one deformation influencing factor according to the feature sequence set and the slope deformation prediction value by using the SHAP formula.

[0196] The above is a schematic solution of a slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP according to this embodiment. It should be noted that the technical solution of the slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP belongs to the same concept as the above technical solution of the slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP. For the details not described in the technical solution of the slope deformation prediction and interpretation device based on a fuzzy echo state network and SHAP, reference can be made to the description of the technical solution of the slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP above.

[0197] Figure 6 A structural block diagram of a computing device 600 according to an embodiment of this specification is shown. The components of the computing device 600 include, but are not limited to, a memory 610 and a processor 620. The processor 620 is connected to the memory 610 through a bus 630, and a database 650 is used to store data.

[0198] The computing device 600 further includes an access device 640, which enables the computing device 600 to communicate via one or more networks 660. Examples of these networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 640 may include one or more of any type of wired or wireless network interface (for example, a network interface card (NIC)), such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC).

[0199] In an embodiment of this specification, the above components of the computing device 600 and Figure 6 other components not shown therein may also be connected to each other, for example, through a bus. It should be understood that Figure 6The block diagram of the computing device shown is for illustrative purposes only and is not a limitation on the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0200] The computing device 600 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs, Personal Computers). The computing device 600 can also be a mobile or stationary server.

[0201] Among them, the processor 620 is used to execute the following computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP are implemented.

[0202] The above is a schematic solution of a computing device in this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP belong to the same concept. For the details not described in detail in the technical solution of the computing device, reference can be made to the description of the technical solution of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP.

[0203] An embodiment of this specification also provides a computer-readable storage medium, which stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP are implemented.

[0204] The above is a schematic solution of a computer-readable storage medium in this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP belong to the same concept. For the details not described in detail in the technical solution of the storage medium, reference can be made to the description of the technical solution of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP.

[0205] An embodiment of this specification also provides a computer program, wherein when the computer program is executed on a computer, the computer is made to execute the steps of the above-mentioned slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP.

[0206] The above is a schematic solution of a computer program according to this embodiment. It should be noted that the technical solution of this computer program and the technical solution of the above slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP belong to the same concept. For the details not described in the technical solution of the computer program, reference can be made to the description of the technical solution of the above slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP.

[0207] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0208] The computer instructions include computer program code, which may be in the form of source code, object code, executable files, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0209] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of this specification are not limited by the described order of actions, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0210] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0211] The preferred embodiments of the present specification disclosed above are only used to help explain the present specification. The alternative embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of the embodiments of the present specification. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present specification, so that those skilled in the art can well understand and utilize the present specification. The present specification is only limited by the claims and their full scope and equivalents.

Claims

1. A slope deformation prediction and interpretation method based on a fuzzy echo state network and SHAP, characterized in that Including: Determine the key point cloud data of the slope area to be detected before the target moment, and extract features from the key point cloud data to obtain a feature sequence set, where the feature sequence set is used to characterize the eigenvalue changes of at least one deformation influencing factor in time series. The extracting features from the key point cloud data to obtain a feature sequence set includes: extracting at least one terrain feature information in the position dimension of the key point cloud data; extracting at least one environmental feature information in the feature dimension of the key point cloud data; determining the feature sequence set according to the terrain feature information and the environmental feature information; According to the feature sequence set and the fuzzy echo state network, determine the slope deformation prediction value at the target moment. The feature sequence set includes multi-source feature sequences at each moment before the target moment, and the multi-source feature sequences refer to feature sequences containing information in multiple different dimensions. The fuzzy echo state network includes an input layer, a fuzzy layer, a reservoir, and an output layer. The determining the slope deformation prediction value at the target moment according to the feature sequence set and the fuzzy echo state network includes: receiving the feature sequence set through the input layer and inputting it into the fuzzy layer; processing each multi-source feature sequence in the feature sequence set through at least one fuzzy rule in the fuzzy layer to obtain a cluster center vector and an influence radius vector corresponding to each moment; determining a weight factor corresponding to each fuzzy rule according to the cluster center vector and the influence radius vector; processing the feature sequence set based on the fuzzy rule through the reservoir to output a weight vector; predicting the slope deformation prediction value at the target moment through the output layer based on the weight vector and the weight factor; According to the feature sequence set and the slope deformation prediction value, use the SHAP formula to determine the deformation component corresponding to the at least one deformation influencing factor, where the deformation component is used to explain the influence degree of each deformation influencing factor on the slope deformation.

2. The slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to claim 1, wherein The determining the key point cloud data of the slope area to be detected before the target moment includes: Collect the original point cloud data of the slope area to be detected before the target moment; Perform radius filtering processing on the original point cloud data to obtain a preprocessed point cloud data set before the target moment; Perform key point extraction on the preprocessed point cloud data set to obtain the key point cloud data.

3. The slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to claim 2, characterized in that The performing key point extraction on the preprocessed point cloud data set to obtain the key point cloud data includes: Determine a target data point, where the target data point is any point in the preprocessed point cloud data set; Determine a neighborhood point set corresponding to the target data point based on a preset quota filtering radius; Calculate the average position information of the neighborhood point set; Calculate the covariance matrix corresponding to the target data point based on the average position information; Calculate the response value corresponding to the target data point according to the covariance matrix; If the response value is greater than a preset threshold, determine that the target data point is a key point.

4. The slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to claim 1, characterized in that The fuzzy echo state network is trained in the following way: Obtain training samples and construct training sample pairs, where the training sample pairs include sample labels, and the sample labels include slope deformation amounts; Input the training samples into the initial model to obtain slope deformation prediction values; Determine the loss value based on the slope deformation prediction values and the slope deformation amounts, and train the initial model based on the loss value until the training stop condition is reached to obtain the fuzzy echo state network.

5. The slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to claim 1, characterized in that According to the feature sequence set and the slope deformation prediction value, use the SHAP formula to determine the deformation components corresponding to the at least one deformation influencing factor, including: wherein the deformation components corresponding to each deformation influencing factor are specifically obtained by the following method: Determine the first feature sequence subset set and the second feature sequence subset set according to the feature sequence set and the deformation influencing factor. Among them, the first feature sequence subset set is the set of each deformation influencing factor subset of the deformation feature factors except the corresponding deformation influencing factor in the feature sequence set, and the second feature sequence subset set is the set of feature subsets combined by each deformation influencing factor; Determine the third feature sequence subset set according to the second feature sequence subset set, where the third feature sequence subset set is the set of feature subsets including the corresponding deformation influencing factor; Determine the deformation components corresponding to the deformation influencing factor according to the first feature sequence subset set, the second feature sequence subset set, and the third feature sequence subset set.

6. The slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to claim 1, characterized in that After determining the deformation components corresponding to the at least one deformation influencing factor according to the feature sequence set, the slope deformation prediction value, and using the SHAP formula, it includes: Generate a deformation influencing factor analysis summary diagram according to the slope deformation prediction value and the deformation components, where the deformation influencing factor analysis summary diagram is used to analyze the influence of the deformation influencing factor on the slope deformation.

7. A computing device, characterized by including: A memory and a processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the slope deformation prediction and interpretation method based on the fuzzy echo state network and SHAP according to any one of claims 1 to 6 are implemented.

Images