Slope stability analysis method and system based on improved TimesNet model

Abstract

The invention discloses a slope stability analysis method and system based on an improved TimesNet model, and the method comprises the steps: obtaining the original data of a slope, carrying out the dynamic analysis of the original data of the slope based on a numerical simulation technology, and obtaining a simulation displacement curve sequence; performing sliding window processing on the simulated displacement curve sequence to obtain a slope variable-length displacement sequence; an improved TimesNet model is constructed, the slope variable-length displacement sequence is input into the improved TimesNet model for model training, and a trained TimesNet model is obtained; carrying out regression based on the trained TimesNet model, and predicting to obtain a soil property parameter; according to the soil property parameters, slope stability analysis is carried out based on a limit equilibrium model, and the safety coefficient is obtained. The slope instability risk prediction method can better adapt to various geological conditions and environmental changes, predicts the slope instability risk in real time under the dynamic condition, achieves the effect of early warning in advance, and is high in accuracy and good in effect.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring and early warning, and specifically relates to a slope stability analysis method and system based on an improved TimesNet model. Background Art

[0002] Slope stability analysis has always been an important topic in the field of geotechnical engineering. With the acceleration of the urbanization process, more and more road, building, and other infrastructure construction projects need to rely on slopes or inclined terrains. The instability of slopes not only may lead to disastrous consequences but even endanger life and property safety. Therefore, timely and accurate prediction of slope instability is of crucial significance for disaster prevention. Traditional slope stability early warning methods usually rely on on-site investigation and static analysis. For example, stability assessment is carried out through traditional soil mechanics parameters (such as internal friction angle, cohesion coefficient, etc.) and the geometric shape of the slope. However, these methods are often limited by factors such as the lack of real-time monitoring data and untimely data collection, resulting in difficulties in ensuring the accuracy and timeliness of early warning, and poor prediction effects for slopes in complex terrains and changing environments.

[0003] With the development of computer technology and artificial intelligence technology, deep learning provides new possibilities for efficient and automated slope stability analysis. Currently, some slope stability prediction models based on deep learning have gradually emerged. For example, convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) have been applied to the prediction of time series data. However, these models usually have problems such as insufficient ability to process variable-length sequences and inability to handle regression tasks simultaneously. In addition, although existing time series prediction models such as TimesNet have certain advantages in sequence modeling, they still face certain challenges in processing variable-length sequences and regression tasks. And how to introduce improved mechanisms into time series prediction models to enable them to handle more complex slope stability analysis tasks remains an urgent problem to be solved.

[0004] In view of this, the present application is specifically proposed. Summary of the Invention

[0005] The technical problems to be solved by the present invention are that traditional static analysis methods make it difficult to ensure the accuracy and timeliness of early warning, and have poor prediction effects for slope risks in complex terrains and changing environments; and existing models have insufficient ability to process variable-length sequences and regression tasks and cannot handle more complex slope stability analysis tasks, etc.

[0006] The object of the present invention is to provide a slope stability analysis method and system based on an improved TimesNet model. The method of the present invention combines deep learning technology with the actual requirements of slope stability analysis. By regressing the collected displacement data (horizontal displacement and vertical displacement data), parameters such as the internal friction angle, cohesion coefficient, and reduction coefficient of the soil mass are predicted, so as to perform slope stability analysis to achieve the purpose of early warning. Through numerical simulation technology, the displacement data of the soil or rock layer at multiple monitoring points is combined with the internal friction angle, cohesion coefficient, and reduction coefficient for dynamic analysis, and simulated displacement curves corresponding to multiple relationships can be obtained in real time. These displacement curves are processed through a sliding window to obtain data sets with different sequence lengths; and aiming at the problem that the existing time series prediction model TimesNet cannot handle variable-length sequences and regression tasks, an improved TimesNet model is proposed, which can input displacement data with different sequence lengths, obtain initial soil parameters, and perform slope stability analysis. The present invention not only overcomes the deficiencies of traditional static analysis, but also can better adapt to various geological conditions and environmental changes, and can predict the instability risk of slopes in real time under dynamic conditions, so as to achieve the effect of early warning, with high accuracy and good results.

[0007] The present invention is realized through the following technical solutions:

[0008] In the first aspect, the present invention provides a slope stability analysis method based on an improved TimesNet model, and the method includes:

[0009] Obtain the original slope data, and perform dynamic analysis on the original slope data based on numerical simulation technology to obtain a sequence of simulated displacement curves; the original slope data is the displacement data of the soil or rock layer;

[0010] Perform sliding window processing on the sequence of simulated displacement curves to obtain a variable-length displacement sequence of the slope;

[0011] Construct an improved TimesNet model, and input the variable-length displacement sequence of the slope into the improved TimesNet model for model training to obtain a trained TimesNet model;

[0012] Perform regression based on the trained TimesNet model to predict soil parameters; based on the soil parameters, perform slope stability analysis based on the limit equilibrium model to obtain a safety factor.

[0013] Furthermore, the method further includes:

[0014] Perform slope instability warning according to the safety factor, including:

[0015] When the safety factor is less than the set threshold, a slope instability warning is issued;

[0016] When the safety factor is greater than or equal to the set threshold, no slope instability warning is issued.

[0017] Further, obtain the original slope data, and conduct dynamic analysis on the original slope data based on numerical simulation technology to obtain a sequence of simulated displacement curves, including:

[0018] Collect the displacement data of the soil or rock stratum, and the displacement data includes horizontal displacement data and vertical displacement data;

[0019] Perform regression on the horizontal displacement data and the vertical displacement data to predict the soil parameters;

[0020] Based on numerical simulation technology, combine the displacement data of multiple monitoring points with the soil parameters for dynamic analysis to obtain multiple sequences of simulated displacement curves corresponding to the slope.

[0021] Further, the soil parameters include the internal friction angle of the soil, the cohesion coefficient, and the reduction coefficient.

[0022] Further, perform a sliding window process on the sequence of simulated displacement curves to obtain a sequence of variable-length slope displacements, including:

[0023] Step A: Take the 8 displacement data of multiple monitoring points and the soil parameters corresponding to the first-row displacement data as a piece of training data. The length of each displacement data is 16, that is, displacement data with a size of (16, 8), and label data with a size of (1, 3);

[0024] Step B: Based on the sliding window technology, move the window downward in the form of a step size of 1, and at the same time take the soil parameters corresponding to the first-row displacement data as a piece of training data, that is, displacement data with a size of (15, 8), and label data with a size of (1, 3);

[0025] Step C: Repeat the above steps A and B to obtain the maximum length Max_L of all sequences of variable-length slope displacements; according to the maximum length Max_L, perform data filling on each sequence of simulated displacement curves to obtain N sequences with a size of (Max_L, 8), that is, the sequence of variable-length slope displacements.

[0026] Further, construct an improved TimesNet model, including:

[0027] Define the input structure of the model, and its size is the size of the sequence of variable-length slope displacements;

[0028] Construct a data embedding layer Embedding for converting the sequence of variable-length slope displacements into an embedded representation;

[0029] Construct the first TimesBlock module and the second TimesBlock module, which are used to capture the periodicity of the variable-length displacement sequence of the slope through frequency-domain analysis and multi-scale feature extraction;

[0030] Construct a backbone network, which is used to stack the first TimesBlock module and the second TimesBlock module in a residual manner;

[0031] Define an additional layer, including a GELU activation function and a Dropout regularization network structure;

[0032] Define a Mark module, which is used to set the sequence indices filled with 0 to 0 and the indices of other valid values to 1, perform matrix multiplication with the output result of the GELU activation function, and the positions of the sequences filled with 0 do not participate in the gradient calculation;

[0033] Define an output layer, which is used to output the result of the Mark module through a fully connected layer Linear.

[0034] Furthermore, according to the soil parameters, perform slope stability analysis based on the limit equilibrium model to obtain a safety factor, including:

[0035] Determine the slip surface according to experience or numerical methods;

[0036] Calculate the sliding force and the shear resistance according to the soil parameters;

[0037] Calculate the force balance between the sliding body and the stable body on the slip surface according to the sliding force and the shear resistance to obtain the safety factor of the slope.

[0038] In a second aspect, the present invention further provides a slope stability analysis system based on an improved TimesNet model, and the system includes:

[0039] An acquisition unit, which is used to acquire the original slope data, and the original slope data is the displacement data of soil or rock strata;

[0040] A numerical simulation unit, which is used to perform dynamic analysis on the original slope data based on numerical simulation technology to obtain a simulated displacement curve sequence;

[0041] A sliding window processing unit, which is used to perform sliding window processing on the simulated displacement curve sequence to obtain a variable-length displacement sequence of the slope;

[0042] A model construction and training unit, which is used to construct an improved TimesNet model, input the variable-length displacement sequence of the slope into the improved TimesNet model for model training, and obtain a trained TimesNet model;

[0043] A slope stability analysis unit is used to perform regression based on the trained TimesNet model to predict soil parameters; and based on the soil parameters, slope stability analysis is performed based on the limit equilibrium model to obtain a safety factor.

[0044] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the slope stability analysis method based on the improved TimesNet model described above is implemented.

[0045] In a fourth aspect, the present invention further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the slope stability analysis method based on the improved TimesNet model described above is implemented.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] 1. The slope stability analysis method and system based on the improved TimesNet model of the present invention combine deep learning technology with the actual needs of slope stability analysis. Through numerical simulation technology, the displacement data of soil or rock layers at multiple monitoring points is combined with the internal friction angle, cohesion coefficient, and reduction coefficient for dynamic analysis, and multiple simulated displacement curves corresponding to relationships can be obtained in real time. These displacement curves are processed through a sliding window to obtain datasets with different sequence lengths; and aiming at the problem that the existing time series prediction model TimesNet cannot handle variable-length sequences and regression tasks, an improved TimesNet model is proposed, which can input displacement data with different sequence lengths, obtain initial soil parameters, and perform slope stability analysis. The present invention not only overcomes the deficiencies of traditional static analysis, but also can better adapt to various geological conditions and environmental changes, and can predict the instability risk of slopes in real time under dynamic conditions, so as to achieve the effect of early warning, with high accuracy and good results.

[0048] 2. The slope stability analysis method and system based on the improved TimesNet model of the present invention can make full use of the advantages of time series data and deep learning through the improved TimesNet-Slope model, improve the accuracy and real-time performance of slope stability analysis, show higher robustness in the face of various slope state changes, have real-time warning capabilities and good adaptability, and provide strong technical support for the prevention and mitigation of slope disasters, with broad application prospects. Description of the Drawings

[0049] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0050] Figure 1 This is the flow chart of the slope stability analysis method based on the improved TimesNet model of the present invention;

[0051] Figure 2 This is the detailed flow chart of the slope stability analysis method based on the improved TimesNet model of the present invention;

[0052] Figure 3 This is the schematic diagram of data filtering of the present invention;

[0053] Figure 4 This is the structural diagram of the TimesNet-Slope model of the present invention;

[0054] Figure 5 This is the schematic diagram of the visualization of the fitting of the model prediction parameters of the present invention;

[0055] Figure 6 This is the structural block diagram of the slope stability analysis system based on the improved TimesNet model of the present invention. Detailed implementation manners

[0056] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments and drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and do not limit the present invention.

[0057] In the training stage of the present invention, it is necessary to obtain the corresponding data of the displacement curve (i.e., the displacement data of soil or rock stratum) and the internal friction angle, cohesion coefficient, and reduction coefficient. First, it is necessary to preprocess the displacement data, then obtain the variable-length displacement sequence of the slope through a sliding window, and finally input it into the TimesNet-Slope model (i.e., the improved TimesNet model) to obtain the output of the initial internal friction angle, cohesion coefficient, and reduction coefficient. The flow chart of the slope stability analysis method of the TimesNet-Slope model is as Figure 1 shown.

[0058] At present, traditional methods for slope stability analysis mostly rely on on-site investigation and static analysis. For example, stability assessment is carried out through traditional soil mechanics parameters (such as the angle of internal friction, cohesion coefficient, etc.) and the geometric shape of the slope. However, these methods have low real-time performance. With the continuous development of computer technology, artificial intelligence, and machine learning, some slope stability prediction models based on deep learning have gradually emerged. For example, convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) have been applied to the prediction of time series data. However, these models usually have problems such as insufficient ability to process variable-length sequences and inability to handle regression tasks simultaneously. In addition, although existing time series prediction models such as TimesNet have certain advantages in sequence modeling, they still face certain challenges when dealing with variable-length sequences and regression tasks. Therefore, how to introduce improved mechanisms into time series prediction models to enable them to handle more complex slope stability analysis tasks remains an urgent problem to be solved.

[0059] Therefore, to solve the above problems, a slope stability analysis method and system based on an improved TimesNet model are proposed. The method of the present invention combines deep learning technology with the actual needs of slope stability analysis. By performing regression on the collected horizontal displacement and vertical displacement data, parameters such as the angle of internal friction, cohesion coefficient, and reduction coefficient of the soil mass are predicted, thereby realizing the stability analysis of the slope. Specifically, through numerical simulation technology, the displacement data of the soil or rock layer at multiple monitoring points is combined with the angle of internal friction, cohesion coefficient, and reduction coefficient for dynamic analysis, and simulated displacement curves corresponding to multiple relationships can be obtained in real time. These displacement curves are processed through a sliding window to obtain data sets with different sequence lengths; and in view of the problem that the existing time series prediction model TimesNet cannot handle variable-length sequences and regression tasks, an improved TimesNet model is proposed, which can input displacement data with different sequence lengths, obtain initial soil mass parameters, and perform slope stability analysis. The present invention not only overcomes the deficiencies of traditional static analysis but also can better adapt to various geological conditions and environmental changes, and can predict the instability risk of the slope in real time under dynamic conditions, thus achieving the effect of early warning, with high accuracy and good results.

[0060] The improved TimesNet-Slope model of the present invention has stronger flexibility and multi-task regression ability compared with the traditional TimesNet model. It can process variable-length time series data and perform dynamic analysis by combining data from multiple monitoring points. Through the improved model structure, it can predict soil mass parameters in real time and accurately, thereby realizing slope stability analysis and intelligent early warning. These improvements enable the TimesNet-Slope model to adapt to more complex and dynamic environments, improving the accuracy and timeliness of slope instability early warning. The improvements are as follows:

[0061] (1) Ability to process variable-length time series data

[0062] Limitations of the TimesNet model: Existing TimesNet models (e.g., in time series data prediction) usually require the input sequence length to be fixed, that is, the time series data input each time must have the same time step. For time series data in practical applications such as slope monitoring data, the input lengths are often inconsistent, which makes it difficult for the standard TimesNet model to handle.

[0063] TimesNet-Slope model: The TimesNet-Slope model can handle variable-length sequence data by adopting a more flexible model architecture. By introducing a variable-length sliding window or by dynamically padding and truncating the input data, and adding the Mark module, the TimesNet-Slope model can adapt to different input lengths, making it more suitable for data in practical applications (such as different monitoring time periods or different numbers of monitoring points).

[0064] (2) Multi-task regression ability

[0065] Limitations of the TimesNet model: Existing TimesNet models are mostly used for single prediction tasks, such as predicting the value at a certain future moment based on historical data. This makes it difficult to directly apply them to complex tasks that require predicting multiple parameters simultaneously.

[0066] TimesNet-Slope model: The TimesNet-Slope model extends the regression ability of the traditional TimesNet model. The improved output layer can simultaneously predict multiple soil parameters, such as the internal friction angle, cohesion coefficient, and reduction coefficient, etc., which are crucial for slope stability analysis. Through multi-task learning, the model can optimize multiple outputs simultaneously to achieve a comprehensive assessment of slope stability.

[0067] (3) Fusion of data from multiple monitoring points

[0068] Limitations of the TimesNet model: Existing TimesNet models may only process data from a single source and cannot fully fuse data from multiple monitoring points.

[0069] Improved TimesNet-Slope model: The TimesNet-Slope model can comprehensively analyze the overall stability of the slope by fusing data from multiple monitoring points, such as horizontal displacement and vertical displacement information from different locations. The fusion of this multi-source data enables the model to obtain more spatial information and improves the accuracy of early warning.

[0070] (4) Improve the generalization ability of the model

[0071] Limitations of the TimesNet model: Facing complex slope geological structures and non-linear relationships, the existing TimesNet model may have problems such as overfitting or insufficient generalization ability of the model.

[0072] TimesNet-Slope model: By introducing more regularization methods, including the Dropout layer and L2 regularization, the improved TimesNet-Slope model can effectively improve the generalization ability of the model and still maintain a high prediction accuracy when dealing with different geological conditions and complex slope change patterns in practical applications.

[0073] Example 1

[0074] As Figure 2 shown, the slope stability analysis method based on the improved TimesNet model of the present invention includes:

[0075] Step 1, obtain the original slope data, and the original slope data is the displacement data of the soil or rock stratum; and perform data preprocessing on the original data.

[0076] In this embodiment, in step 1, the displacement data of the soil or rock stratum is collected, and the displacement data includes horizontal displacement data and vertical displacement data; regression is performed on the horizontal displacement data and the vertical displacement data, and soil parameters are predicted. The soil parameters include the internal friction angle, cohesion coefficient, and reduction coefficient of the soil.

[0077] In this embodiment, the data preprocessing includes denoising, handling missing values, etc., specifically:

[0078] (1) Denoising, when the instrument collects data, obvious noise points may occur in the data due to external factors, and obvious noise needs to be removed before inputting the model for training. Here, an adaptive filtering method is used to denoise the data, so as to ensure the effective elimination of outliers. The relevant steps are as follows:

[0079] a. Initialize the filter, set it to 0 or a small random number;

[0080] b. Select the loss function, and use the minimum mean square error MSE to optimize the coefficients of the filter;

[0081] c. Update the filter coefficients, and continuously update the filter coefficients according to the input data and the expected output to minimize the error;

[0082] For the LMS algorithm:

[0083] w k = w k-1 + u(d k-w k-1 x k )x k

[0084] where w k is the filter coefficient, d k is the desired output, x k is the input signal, and u is the step size.

[0085] After obtaining the updated filter coefficients for d, use these coefficients to filter the input signal.

[0086] As shown above, the effect diagram after denoising the displacement data is as Figure 3 shown, Figure 3 where Nosisy Data is the noise data and Filter Data is the data after denoising; this step is crucial for removing the data noise caused by instrument errors or environmental interference.

[0087] (2) Process missing values. To better train the sequence features of the model, it is necessary to process the missing values. Here, the adjacent mean filling method is adopted, that is, traverse the entire sequence cyclically. If there is a missing value at the current node, use the mean of the previous node and the next node as the value of the current node.

[0088] Step 2: Based on numerical simulation technology, perform dynamic analysis on the original slope data to obtain a sequence of simulated displacement curves;

[0089] In this embodiment, based on numerical simulation technology, the displacement data of multiple monitoring points is combined with soil parameters for dynamic analysis to obtain multiple sequences of simulated displacement curves corresponding to the slope.

[0090] Step 3: Perform a sliding window process on the sequence of simulated displacement curves to obtain a variable-length displacement sequence of the slope;

[0091] In practical applications, since it is necessary to analyze the stability of the slope, the length of the slope displacement curve sequence input into the model is not fixed. Therefore, it is necessary to perform a sliding window operation on the current data.

[0092] The simulated data contains multiple sequences of simulated displacement curves corresponding to the slope. Each piece of data includes the process of the slope from stable - gradual change - transient change. Therefore, each curve needs to be extracted separately as follows:

[0093] Step A: Taking the data of a single slope sequence as an example, its length is 16. First, take the 8 displacement curves of 4 monitoring points and the internal friction angle, cohesion coefficient, and reduction coefficient corresponding to the displacement data in the first row as a piece of training data, that is, the displacement data with a size of (16, 8) and the label data with a size of (1, 3);

[0094] Step B: Based on the sliding window technique, move the window downward with a step size of 1, and use the soil parameters corresponding to the displacement data in the first row as a piece of training data, that is, displacement data with a size of (15, 8) and label data with a size of (1, 3).

[0095] Step C: Repeat the above Steps A and B to obtain the maximum length Max_L of all the slope variable-length displacement sequences. For example, if the length of a certain sequence is L, then a zero-filled array with a size of (Max_L - L, 8) needs to be filled to make its final size become (Max_L, 8); fill other sequences in the above manner, and a total of N sequences with a size of (Max_L, 8) are obtained, that is, the slope variable-length displacement sequences.

[0096] Step 4: Build an improved TimesNet model, and input the slope variable-length displacement sequences into the improved TimesNet model for model training to obtain a trained TimesNet model.

[0097] In this embodiment, a TimesNet-Slope model is built, and its model is as Figure 4 shown as follows:

[0098] (1) The maximum length of the current sequence Max_L = 27, so the input structure shape of the model is defined as (27, 8), indicating that the input of the model is 8-dimensional data for 27 days.

[0099] (2) Build a data embedding layer Embedding, which is used to convert the slope variable-length displacement sequences into an embedding representation that the model can process; set the embedding dimension to d_model = 8, so the data output size after passing through Embedding is (27, 8). It includes value embedding (ValueEmbedding), positional embedding (PositionalEmbedding), and temporal embedding (TemporalEmbedding).

[0100] (3) Build TimesBlock modules, including the first TimesBlock module and the second TimesBlock module, which are used to capture the periodic patterns of the slope variable-length displacement sequences through frequency domain analysis and multi-scale feature extraction.

[0101] Taking the first TimesBlock module as an example for illustration, it includes:

[0102] i. Fourier Transform FFT. Perform FFT on the embedded representation input after Embedding to extract its frequency domain information (frequency and amplitude). According to the amplitude magnitude, select the top-k (k = 3) frequency components, which correspond to the most significant periodic patterns in the time series. Convert the selected top-k frequency components back to the time domain to obtain the corresponding periodic signals. After FFM processing, two results are output. The first result is an array with the shape of [top-k], representing the periods of the most important top-k frequency components. The second result is the amplitude information with the shape of [Max_L / / 2+1, top-k], representing the average amplitude of the selected top-k frequency components in all samples.

[0103] ii. Introduce the InceptionBlock in GoogleNet, called "Parameter-Efficient Inception Block", for multi-scale feature extraction and reducing the number of computational parameters. According to the first return value of FFM (the top-k period array), reshape the input data (27, 8) into a three-dimensional array (27 / [i], [i], 8) according to the period length, where [i] represents the i-th period length of the array. Then input the reshaped three-dimensional array into the InceptionBlock and use 2D convolutional kernels for multi-scale feature extraction. Finally, restore the result after the convolutional operation to the original shape (27, 8).

[0104] iii. At the final stage of the TimesBlock module, introduce the Adaptive Aggregation mechanism for fusing features of different periods. Convert the second return value of FFM (the average amplitude of the top-k frequency components) into weights through the Softmax function, and use the calculated weights to perform weighted summation on the output results of each period in step ii to obtain the final output of the TimesBlock, with the output shape of (27, 8), which is the same as the input shape.

[0105] (4) Construct the backbone network, which uses the above two TimesBlock modules (the first TimesBlock module and the constructed second TimesBlock module) to be stacked in a residual manner; it can effectively solve the problems of gradient disappearance or gradient explosion;

[0106] (5) Define the additional layer, including the GELU activation function and the Dropout regularization network structure for preventing overfitting, with the coefficient set to 0.25;

[0107] (6) Define the Mark module, which is used to set the sequence indices filled with 0 to 0 and the indices of other valid values to 1, perform matrix multiplication with the output result of the GELU activation function, and the positions of the sequences filled with 0 do not participate in the gradient operation, thus improving the calculation speed;

[0108] (7) Define the output layer, which is used to output the result of the Mark module through the fully connected layer Linear, and the output is the defined parameter length, with a size of (3, 1).

[0109] In this embodiment, the training of the TimesNet-Slope model is as follows:

[0110] Split all the data into 80% training data + 20% test data, and then construct the data according to the method in step 3, that is, first perform normalization processing on the training data alone, and then use the sliding window method to preprocess the data. The sliding window method defines the window size as 1, which can divide the training set to the maximum extent. Finally, a total of 6884 groups of training data of Max_L * 8 and 1720 groups of test data of Max_L * 8 are obtained.

[0111] During training, the model parameters need to be continuously optimized until the final prediction effect is obtained, and the currently trained model is the final result.

[0112] Finally, the average errors and root mean square errors of the internal friction angle, cohesion coefficient, and reduction coefficient corresponding to each node are calculated to be 0.061, 0.062, 0.067 and 0.0056, 0.0057, 0.0055 respectively. The visualization result of the model test is as Figure 5 shown. Input a piece of original data from the test set, process the original data according to the method in step 3, and input it into the trained model. The internal friction angle, cohesion coefficient, and reduction coefficient corresponding to each displacement data node can be obtained, which are basically fitted with the original internal friction angle, cohesion coefficient, and reduction coefficient, indicating that this method can accurately predict the corresponding soil coefficient. Figure 4 In [the figure], point represents a node, referring to a certain day; K represents the reduction coefficient, Internal friction angle represents the internal friction angle; Cohesion represents the cohesion coefficient; true represents the real geological parameter, and pred represents the geological parameter predicted by the model.

[0113] Step 5: Perform regression based on the trained TimesNet model to predict the soil parameters; based on the soil parameters, perform slope stability analysis based on the limit equilibrium model to obtain the safety factor.

[0114] When conducting slope stability analysis, the limit equilibrium model or numerical simulation model of the slope is often established, combined with the initial internal friction angle, cohesion coefficient, and reduction coefficient, to evaluate the stability and safety factor of the slope.

[0115] Internal friction angle (φ) and cohesion coefficient (c): Both jointly affect the shear strength of the soil mass, and their changes directly affect the slope stability. If these parameters decrease, the possibility of slope sliding increases.

[0116] Reduction coefficient (K): Considering the strength reduction of the soil mass under actual working conditions, the reduction coefficient can help simulate the slope stability changes under complex environments (such as rainfall, construction, etc.). If the reduction coefficient increases, the internal friction angle and cohesion coefficient will decrease proportionally, showing a linear relationship.

[0117] The steps to construct the limit equilibrium model are as follows:

[0118] (1) Determine the slip surface: First, one or more slip surfaces need to be determined (usually the position and shape of the slip surface are selected through experience or numerical methods). The slip surface can be circular, linear, or other irregular shapes. Selecting a suitable slip surface is crucial for accurately calculating the slope stability.

[0119] (2) Force balance:

[0120] 1) Driving force: The driving force refers to the external force that pushes the slope to slide. The driving force is usually caused by factors such as gravity and water pressure.

[0121] F slide =W·sin(θ)

[0122] Where, W is the gravity of the sliding body, and θ is the angle between the slip surface and the horizontal plane.

[0123] 2) Shearing resistance: The shearing resistance refers to the force that resists shear failure during the sliding process of the soil mass or rock mass. The shearing resistance is usually determined by the cohesion (c) and internal friction angle (φ) of the soil mass. It is expressed as:

[0124] T=c + σtan(φ)

[0125] Where, σ is the normal stress (positive stress) acting on the slip surface, which is usually calculated through the thickness, weight of the soil layer, and other external loads.

[0126] (3) Calculate the safety factor (FS): The safety factor refers to the ratio of the shear force resisting sliding to the shear force promoting sliding on the slip surface. The limit equilibrium method obtains the safety factor of the slope by calculating the force balance between the sliding body and the stable body on the assumed slip surface. Among them: FS = Shearing resistance / Driving force.

[0127] (4) Safety Factor (FS) Warning: When the safety factor is less than the set threshold, a slope instability warning is issued; for example, a first-level warning is issued when FS < 1.2, and a second-level warning is issued when FS < 1.0; when the safety factor is greater than or equal to the set threshold, no slope instability warning is issued.

[0128] Embodiment 2

[0129] As Figure 6 shown, the difference between this embodiment and Embodiment 1 is that this embodiment provides a slope stability analysis system based on the improved TimesNet model, which corresponds one-to-one with the slope stability analysis method based on the improved TimesNet model in Embodiment 1; the system includes:

[0130] An acquisition unit for acquiring original slope data, where the original slope data is displacement data of soil or rock strata;

[0131] A numerical simulation unit for dynamically analyzing the original slope data based on numerical simulation technology to obtain a sequence of simulated displacement curves;

[0132] A sliding window processing unit for performing sliding window processing on the sequence of simulated displacement curves to obtain a sequence of variable-length displacements of the slope;

[0133] A model construction and training unit for constructing an improved TimesNet model and inputting the sequence of variable-length displacements of the slope into the improved TimesNet model for model training to obtain a trained TimesNet model;

[0134] A slope stability analysis unit for performing regression based on the trained TimesNet model to predict soil parameters; and based on the soil parameters, performing slope stability analysis based on the limit equilibrium model to obtain the safety factor.

[0135] As a further implementation, the system further includes:

[0136] A warning unit for performing slope instability warning according to the safety factor, including: when the safety factor is less than the set threshold, a slope instability warning is issued; when the safety factor is greater than or equal to the set threshold, no slope instability warning is issued.

[0137] Wherein, the execution process of each unit is carried out according to the process steps of the slope stability analysis method based on the improved TimesNet model in Embodiment 1, and will not be elaborated one by one in this embodiment.

[0138] Meanwhile, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned slope stability analysis method based on the improved TimesNet model is implemented.

[0139] Meanwhile, the present invention also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned slope stability analysis method based on the improved TimesNet model is implemented.

[0140] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0141] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0142] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide for implementing the functions in one Figure 1 one process or multiple processes and / or blocksFigure 1 Steps of functions specified in one or more boxes.

[0144] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of 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. The slope stability analysis method based on the improved TimesNet model is characterized by: The method includes: Obtaining original slope data, and dynamically analyzing the original slope data based on numerical simulation technology to obtain a simulated displacement curve sequence; the original slope data is displacement data of soil or rock layer; Performing sliding window processing on the simulated displacement curve sequence to obtain a slope variable length displacement sequence; Constructing an improved TimesNet model, and inputting the variable-length displacement sequence of the slope into the improved TimesNet model for model training to obtain a trained TimesNet model; Regression is performed based on the trained TimesNet model to predict soil parameters; based on the soil parameters, slope stability analysis is performed based on the limit equilibrium model to obtain the safety factor.

2. The slope stability analysis method based on the improved TimesNet model according to claim 1 is characterized in that: The method further includes: According to the safety factor, the slope instability warning is carried out, including: When the safety factor is less than a set threshold, a slope instability warning is issued; When the safety factor is greater than or equal to the set threshold, no slope instability warning is issued.

3. The slope stability analysis method based on the improved TimesNet model according to claim 1 is characterized in that: The original slope data is obtained, and the original slope data is dynamically analyzed based on numerical simulation technology to obtain a simulated displacement curve sequence, including: Collecting displacement data of soil or rock formations, wherein the displacement data includes horizontal displacement data and vertical displacement data; Regressing the horizontal displacement data and the vertical displacement data to predict soil parameters; Based on numerical simulation technology, the displacement data of multiple monitoring points are combined with soil parameters for dynamic analysis to obtain a sequence of simulated displacement curves corresponding to multiple slope occurrences.

4. The slope stability analysis method based on the improved TimesNet model according to claim 3 is characterized in that: The soil parameters include the internal friction angle, cohesion coefficient and reduction coefficient of the soil.

5. The slope stability analysis method based on the improved TimesNet model according to claim 3 is characterized in that: The simulated displacement curve sequence is subjected to sliding window processing to obtain a slope variable length displacement sequence, including: Step A, taking 8 displacement data of multiple monitoring points and the soil parameters corresponding to the first row of displacement data as a training data, the length of each displacement data is 16, that is, the displacement data of size (16, 8) and the label data of size (1, 3); Step B, based on the sliding window technology, the window is moved downward in a step size of 1, and the soil parameters corresponding to the first row of displacement data are used as a training data, that is, displacement data with a size of (15, 8) and label data with a size of (1, 3); Step C, repeat the above steps A and B to obtain the maximum length Max_L of all slope variable length displacement sequences; according to the maximum length Max_L, fill the data of each simulated displacement curve sequence to obtain N sequences of size (Max_L, 8), namely, the slope variable length displacement sequence.

6. The slope stability analysis method based on the improved TimesNet model according to claim 1 is characterized in that: Build an improved TimesNet model, including: Define the input structure of the model, whose size is the size of the slope into displacement sequence; Constructing a data embedding layer Embedding, which is used to convert the variable-length displacement sequence of the slope into an embedding representation; Constructing a first TimesBlock module and a second TimesBlock module, for capturing the periodicity of the variable length displacement sequence of the slope through frequency domain analysis and multi-scale feature extraction; Constructing a backbone network for superimposing the first TimesBlock module and the second TimesBlock module in a residual manner; Define additional layers, including GELU activation function and Dropout regularized network structure; Define a Mark module, which is used to set the index of the sequence filled with 0 to 0, and the index of other valid values ​​to 1, and perform matrix multiplication with the output result of the GELU activation function. The sequence position filled with 0 does not participate in the gradient calculation; Define the output layer, which is used to output the results of the Mark module through the fully connected layer Linear.

7. The slope stability analysis method based on the improved TimesNet model according to claim 1 is characterized in that: According to the soil parameters, the slope stability analysis is carried out based on the limit equilibrium model to obtain the safety factor, including: Determine the sliding surface based on empirical or numerical methods; According to the soil parameters, sliding force and shear force are calculated; The force balance between the sliding body and the stabilizing body on the sliding surface is calculated based on the sliding force and the shearing force to obtain the safety factor of the slope.

8. The slope stability analysis system based on the improved TimesNet model is characterized by: The system includes: An acquisition unit, used for acquiring original slope data, wherein the original slope data is displacement data of soil or rock layer; A numerical simulation unit, used for dynamically analyzing the original data of the slope based on numerical simulation technology to obtain a simulated displacement curve sequence; A sliding window processing unit, used for performing sliding window processing on the simulated displacement curve sequence to obtain a slope variable length displacement sequence; A model building and training unit, used for building an improved TimesNet model, and inputting the variable-length displacement sequence of the slope into the improved TimesNet model for model training to obtain a trained TimesNet model; The slope stability analysis unit is used to perform regression based on the trained TimesNet model to predict soil parameters; according to the soil parameters, the slope stability analysis is performed based on the limit equilibrium model to obtain the safety factor.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the slope stability analysis method based on the improved TimesNet model as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the slope stability analysis method based on the improved TimesNet model as described in any one of claims 1 to 7 is implemented.

Images