Road slope stability monitoring and early warning method and system

Abstract

The invention discloses a road slope stability monitoring and early warning method and system, and belongs to the field of slope monitoring and early warning. The method comprises the following steps: acquiring a remote sensing image of a road slope in real time, performing feature extraction on image features, and fusing the road slope image features with time sequence data; the basic problem of data source input of the LSTM model is corrected from the source, the problem that the LSTM model is difficult to capture a long-range dependency relationship and local mutation features is solved, the accuracy of road slope instability prediction performed by the LSTM model is improved, and finally more accurate slope instability risk early warning is realized through a grading early warning mechanism. The problem that in the prior art, due to the fact that an LSTM model is directly used for prediction, extraction of a long-range dependency relation and local mutation features in time series data is not fully considered, and consequently the road slope instability prediction accuracy is low is solved.

Description

Technical Field

[0001] The present invention relates to the field of slope monitoring and early warning, and particularly to a method and system for monitoring and early warning the stability of highway slopes. Background Art

[0002] The monitoring of highway slope stability is an important link to ensure road traffic safety. Traditional monitoring methods mainly rely on manual inspections and static data collection of single sensors (such as displacement gauges or inclinometers). However, these methods have obvious limitations.

[0003] Currently, Internet of Things technology and machine learning algorithms have been gradually introduced into the field of slope monitoring. For example, LSTM (Long Short-Term Memory Network) is used to predict landslides for time series data. However, existing technologies usually directly input raw sensor data into the LSTM model without fully considering the extraction of long-range dependence relationships and local mutation features in time series data. Since slope instability is often caused by the combined action of long-term geological changes and short-term extreme weather (such as heavy rainfall), it is difficult to accurately capture such complex patterns only relying on the time series modeling ability of LSTM, resulting in insufficient prediction accuracy and a high false alarm rate.

[0004] Therefore, the existing technology has the problem that due to directly using the LSTM model for prediction without fully considering the extraction of long-range dependence relationships and local mutation features in time series data, the prediction accuracy of highway slope instability is low. Summary of the Invention

[0005] The present invention provides a method and system for monitoring and early warning the stability of highway slopes. By collecting remote sensing images of highway slopes in real time and extracting image features, the highway slope image features are fused with time series data, fundamentally correcting the fundamental problem of the data source input of the LSTM model, and being able to solve the problem that the existing technology has low prediction accuracy of highway slope instability due to directly using the LSTM model for prediction without fully considering the extraction of long-range dependence relationships and local mutation features in time series data.

[0006] To solve the above technical problems, an embodiment of the present invention provides a method for monitoring and early warning the stability of highway slopes, including: Obtaining the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment within a currently preset time period, and constructing an initial feature dataset; Collecting remote sensing images of the highway slope in real time, dividing the remote sensing images into cells and marking corresponding coordinate values for each cell; calculating weights for each coordinate value according to the initial feature data, so as to update the coordinate values marked for each cell, and obtaining highway slope time series data composed of the corresponding values of each cell; Input the time series data of the highway slope into the trained TCN model, and perform initial feature extraction on the time series data of the highway slope through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope; Input the initial feature sequence data into the trained LSTM model, and perform instability prediction on the highway slope for the initial feature sequence data through the trained LSTM model to obtain the instability prediction probability value of the highway slope; According to the instability prediction probability value of the highway slope, determine the classification early warning result, and then process the highway slope according to the classification early warning result.

[0007] Further, in the step of collecting the remote sensing image of the highway slope in real time, dividing the remote sensing image into cells, and marking the corresponding coordinate values for each cell, it specifically includes: Perform normalization processing on the collected remote sensing image, and at the same time perform preset reference point recognition on the remote sensing image. Set the recognized preset reference point in each frame of the remote sensing image as the alignment point, and perform feature alignment on each frame of the remote sensing image; Through the rectangular coordinate system, divide the coordinate axes with the alignment point of each frame of the remote sensing image as the origin of the rectangular coordinate system, and perform numerical unit analysis on the initial feature dataset. Determine the cell interval size of the coordinate axis division according to the change range of the initial feature dataset within the unit time; Determine the coordinate values corresponding to each cell mark according to the interval size after each cell division.

[0008] Further, in the step of calculating the weight for each coordinate value according to the initial feature dataset to update the coordinate value corresponding to each cell mark, and obtaining the time series data of the highway slope composed of the corresponding values of each cell, it specifically includes: According to the coordinate values corresponding to each cell mark, calculate the spatial distance values between the tilt angle data, soil humidity data, displacement data, and geological data and the alignment point respectively; Perform superposition calculation on each coordinate value in turn according to the spatial distance value, so that the interval size of each cell changes, thereby updating the coordinate value corresponding to each cell mark; Sort the updated data to obtain the time series data of the highway slope composed of the corresponding values of each cell.

[0009] Further, the initial feature extraction of the time series data of the highway slope through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope includes: Extract the temporal features of the highway slope time series data through the causal convolution layer to obtain the temporal features at each moment in the highway slope time series data; Expand the temporal features at each moment in the highway slope time series data through the dilated convolution layer to obtain the dilated features at each moment in the highway slope time series data; Add the temporal features and dilated features at each moment in the highway slope time series data according to the time step through the residual connection layer to obtain the connection features at each moment in the highway slope time series data; Nonlinearly activate the connection features at each moment in the highway slope time series data through the activation layer to obtain the initial feature sequence data of the highway slope.

[0010] Furthermore, the expression of the causal convolution layer is: ; Among them, is the temporal feature at the t-th moment in the highway slope time series data; t is the t-th moment of the highway slope time series data; K is the convolution kernel size of the causal convolution layer; is the convolution kernel weight of the causal convolution layer; is the highway slope data at the (t - k)-th highway slope data of the highway slope time series data; is the bias term of the causal convolution layer; The expression of the dilated convolution layer is: ; Among them, is the dilated feature at the t-th moment in the highway slope time series data; t is the t-th moment of the highway slope time series data; K is the convolution kernel size of the dilated convolution layer; is the temporal feature at the (t - d*k)-th moment in the highway slope time series data, and d is the dilation factor; is the convolution kernel weight of the dilated convolution layer, and ; is the bias term of the dilated convolution layer, and ; The expression of the residual connection layer is: ; Among them, is the connection feature at the t-th moment in the highway slope time series data; is the temporal feature at the t-th moment in the highway slope time series data; is the dilated feature at the t-th moment in the highway slope time series data; The expression of the activation layer is: ; Among them, is the initial feature at the t-th moment in the initial feature sequence data of the highway slope; is the connection feature at the t-th moment in the time series data of the highway slope; is the ReLU activation function.

[0011] Further, after the activation layer, there is also included: an attention mechanism layer; The attention mechanism layer is used to calculate the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculate the attention weight of the initial feature data at each moment according to the attention score of the initial feature data at each moment. Then, multiply the attention weight of the initial feature data at each moment by the initial feature data at each moment to obtain the enhanced initial feature data at each moment.

[0012] Further, the calculating the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculating the attention weight of the initial feature data at each moment according to the attention score of the initial feature data at each moment includes: For the initial feature data at each moment, perform a linear transformation on the initial feature data through the learnable weight matrix and bias vector of the attention mechanism to obtain the transformed initial feature data; Perform a non-linear activation on the transformed initial feature data through the hyperbolic tangent activation function to obtain the activated initial feature data; Perform a scalar score conversion on the activated initial feature data through a preset learnable weight vector to obtain the attention score of the initial feature data; Perform an exponential processing on the attention score of each initial feature data to obtain the attention score exponential value of each initial feature data, and calculate the sum value of the attention score exponents of all initial feature data; Perform a division calculation on the attention score exponential value of each initial feature data and the sum value of the attention score exponents of all initial feature data to obtain the attention weight of each initial feature data.

[0013] Further, the LSTM model includes: an input gate, a forget gate, a candidate memory unit, an update memory unit, an output gate, a hidden state update unit, and a fully connected layer; The inputting the initial feature sequence data into the trained LSTM model, and performing the instability prediction of the highway slope on the initial feature sequence data through the trained LSTM model to obtain the instability prediction probability value of the highway slope includes: For each initial feature sequence at each moment of the initial feature sequence data, perform a linear transformation and non-linear activation of the input gate according to the initial feature data at the current moment and the hidden state at the previous moment to obtain the input gate value at the current moment; Perform a linear transformation and non-linear activation of the forget gate according to the initial feature data at the current moment and the hidden state at the previous moment to obtain the forget gate value at the current moment; Perform a linear transformation and hyperbolic tangent non-linear activation on the candidate memory cell according to the initial feature data at the current moment and the hidden state at the previous moment to obtain the candidate memory data at the current moment; Multiply the input gate value at the current moment by the candidate memory data at the current moment to obtain a new memory increment data, and multiply the forget gate value at the current moment by the memory update data at the previous moment to obtain the retained memory data. Calculate the sum of the new memory increment data and the retained memory data to obtain the memory update data of the update memory cell at the current moment; Perform a linear transformation and non-linear activation of the output gate according to the initial feature data at the current moment and the hidden state at the previous moment to obtain the output gate value at the current moment; Perform a hyperbolic tangent non-linear activation on the memory update data at the current moment to obtain the activated memory update data, and calculate the hidden state of the hidden state update unit at the current moment according to the activated memory update data and the output gate value at the current moment; Input the hidden state at the last moment of the initial feature time series data into the fully connected layer, and perform a linear transformation and non-linear activation on the hidden state at the current moment through the fully connected layer to generate the instability prediction probability value of the highway slope.

[0014] Further, determining a hierarchical early warning result according to the instability prediction probability value of the highway slope, and then processing the highway slope according to the hierarchical early warning result, including: Compare the instability prediction probability value of the highway slope with the upper limit index and the lower limit index of the preset hierarchical early warning interval to determine the early warning level; wherein, the preset hierarchical early warning interval includes: low-risk early warning interval, medium-risk early warning interval, higher-risk early warning interval and high-risk early warning interval; the upper limit index of the low-risk early warning interval is less than the lower limit index of the medium-risk early warning interval; the upper limit index of the medium-risk early warning interval is less than the lower limit index of the higher-risk early warning interval; the upper limit index of the higher-risk early warning interval is less than the lower limit index of the high-risk early warning interval; When the early warning level is low risk, then conduct routine monitoring on the highway slope; When the early warning level is medium risk, then shorten the monitoring period of the highway slope, and notify the technical personnel to conduct on-site investigation and formulate a preliminary emergency response plan; When the early warning level is a relatively high risk, continuous monitoring of the highway slope is carried out, traffic control is implemented in the surrounding area of the slope, and the emergency rescue team is notified; When the early warning level is a high risk, the people within the influence range of the slope are notified to evacuate, and the emergency rescue team is notified to carry out emergency rescue.

[0015] Based on the above method embodiment, the present invention correspondingly provides a system embodiment; An embodiment of the present invention provides a monitoring and early warning system for the stability of a highway slope, including: a data acquisition module, an image acquisition module, a feature extraction module, a slope instability prediction module, and a slope instability early warning module; The data acquisition module is used to acquire the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment within the currently preset time period, and construct an initial feature dataset; The image acquisition module is used to collect the remote sensing image of the highway slope in real time, divide the remote sensing image into cells and mark the corresponding coordinate values for each cell; calculate the weight of each coordinate value according to the initial feature dataset, so as to update the coordinate value marked by each cell, and obtain the highway slope time series data composed of the corresponding values of each cell; The feature extraction module is used to input the highway slope time series data into the trained TCN model, and perform initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model, so as to obtain the initial feature sequence data of the highway slope; The slope instability prediction module is used to input the initial feature sequence data into the trained LSTM model, and perform highway slope instability prediction on the initial feature sequence data through the trained LSTM model, so as to obtain the instability prediction probability value of the highway slope; The slope instability early warning module is used to determine the classification early warning result according to the instability prediction probability value of the highway slope, and then process the highway slope according to the classification early warning result.

[0016] Compared with the prior art, the embodiment of the present invention has the following beneficial effects: First, the present invention constructs an initial feature dataset by using the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment within the current preset time period. Then, it collects the remote sensing images of the highway slope in real time, divides the remote sensing images into cells, and marks the corresponding coordinate values for each cell. According to the initial feature dataset, weight calculation is performed on each coordinate value to update the coordinate value marked for each cell, obtaining the highway slope time series data composed of the corresponding values of each cell. Immediately afterwards, the highway slope time series data is input into the trained TCN model, and the initial features of the highway slope time series data are extracted through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model, obtaining the initial feature sequence data of the highway slope. Then, the initial feature sequence data is input into the trained LSTM model, and the trained LSTM model is used to predict the instability of the highway slope for the initial feature sequence data, obtaining the instability prediction probability value of the highway slope. Finally, based on the instability prediction probability value of the highway slope, a hierarchical early warning result is determined, and then the highway slope is processed according to the hierarchical early warning result. That is, the present invention corrects the fundamental problem of the data source input of the LSTM model from the root by collecting the remote sensing images of the highway slope in real time and extracting the image features, fusing the highway slope image features with the time series data. By constructing the highway slope time series data and using the causal convolution layer, dilated convolution layer, and residual connection layer of the TCN model to extract the features of the highway slope time series data before the prediction stage of the LSTM model, high-accuracy feature data is provided for the subsequent LSTM model to predict the instability of the highway slope, solving the problem that the LSTM model is difficult to capture long-range dependence relationships and local mutation features, avoiding directly using the LSTM model to predict the instability probability of the highway slope, improving the accuracy of the LSTM model in predicting the instability of the highway slope. Finally, based on the hierarchical early warning mechanism of the highway slope instability probability prediction value, more accurate and timely slope risk early warning is realized, solving the problem in the prior art that the direct use of the LSTM model for prediction does not fully consider the extraction of long-range dependence relationships and local mutation features in the time series data, resulting in low accuracy of the highway slope instability prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a method for monitoring and early warning the stability of a highway slope provided by an embodiment of the present invention; Figure 2 It is a module diagram of a system for monitoring and early warning the stability of a highway slope provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0019] Embodiment 1: Refer to Figure 1 , which is a flowchart of a method for monitoring and warning the stability of a highway slope provided by an embodiment of the present invention; in order to solve the problem in the prior art that the prediction accuracy of highway slope instability is low due to directly using the LSTM model for prediction without fully considering the long-range dependence relationship in time-series data and the extraction of local mutation features; the method at least includes the following steps: Step S1: Obtain the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment within the currently preset time period, and construct an initial feature dataset; In this embodiment, for the inclination angle data, high-precision (±0.01°) and low-power MEMS inclination sensors can be selected, and a monitoring point is arranged every 50 meters longitudinally along the slope and every 20 meters transversely to form a grid monitoring network to collect the inclination angle data of the highway slope.

[0020] In this embodiment, for the soil humidity data, sensors based on the TDR (Time Domain Reflectometry) principle can be used to collect the soil humidity data of the highway slope. The arrangement depth is 0.5 meters, 1.0 meters, and 1.5 meters, and a monitoring point is arranged every 10 meters for each layer to capture the soil humidity data of different soil layers.

[0021] In this embodiment, for the displacement data, the combination of GNSS (Global Navigation Satellite System) and InSAR (Interferometric Synthetic Aperture Radar) can be used for collection. The GNSS receiver is arranged at the key parts of the slope, and InSAR regularly obtains the slope deformation data through satellites or drones to achieve millimeter-level displacement monitoring.

[0022] In this embodiment, the geological data can be obtained from the geological database constructed before the construction of the highway slope or during regular geological exploration work; the geological data stored in the geological database is the geological information data such as the rock and soil types (such as clay, sand, rock, etc.), geological structures (such as the location and characteristics of faults and folds), and stratigraphic distributions of the slope determined by means of drilling, geophysical exploration (such as geological radar), etc. during the construction of the highway slope or during regular geological exploration work.

[0023] Exemplarily, if the current preset time period is set to 5 minutes, the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope for each minute within the current 5 minutes are acquired, and an initial feature dataset is constructed based on these data. ; where each includes the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment t. , for example .

[0024] Step S2: Real-time collect remote sensing images of the highway slope, divide the remote sensing images into cells and mark corresponding coordinate values for each cell; calculate the weights for each coordinate value according to the initial feature data, so as to update the coordinate values marked for each cell, and obtain the highway slope time series data composed of the corresponding values of each cell.

[0025] In this embodiment, this step is divided into two aspects. The first aspect: Normalize the collected remote sensing images, and at the same time identify preset reference points in the remote sensing images. Set the identified preset reference points in each frame of the remote sensing image as alignment points, and perform feature alignment on each frame of the remote sensing image; through a rectangular coordinate system, divide the coordinate axes with the alignment point of each frame of the remote sensing image as the origin of the rectangular coordinate system, and perform numerical unit analysis on the initial feature dataset. Determine the cell interval size of the coordinate axis division according to the change range of the initial feature dataset within the unit time; determine the coordinate values corresponding to each cell mark according to the interval size after each cell division. The second aspect: According to the coordinate values corresponding to each cell mark, calculate the spatial distance values between the inclination angle data, soil humidity data, displacement data, and geological data and the alignment point respectively; perform superposition calculations on each coordinate value in sequence according to the spatial distance values, so that the interval size of each cell changes, thereby updating the coordinate values corresponding to each cell mark; sort the updated data to obtain the highway slope time series data composed of the corresponding values of each cell.

[0026] This step corrects the fundamental problem of the data source input of the LSTM model at the root by real-time collecting remote sensing images of the highway slope and extracting image features, and fusing the highway slope image features with the time series data. Before the subsequent prediction stage of the LSTM model, it solves the problem in the prior art that the prediction accuracy of highway slope instability is low because the long-range dependence relationship and local mutation feature extraction in the time series data are not fully considered when directly using the LSTM model for prediction.

[0027] Step S3: Input the highway slope time series data into the trained TCN model, and perform initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope; In this embodiment, the performing initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope includes: Perform temporal feature extraction on the highway slope time series data through the causal convolution layer to obtain the temporal features at each moment in the highway slope time series data; For the temporal feature at each moment in the highway slope time series data, perform feature dilation through the dilated convolution layer to obtain the dilated features at each moment in the highway slope time series data; Add the temporal features and dilated features at each moment in the highway slope time series data according to the time step through the residual connection layer to obtain the connection features at each moment in the highway slope time series data; Perform non-linear activation on the connection features at each moment in the highway slope time series data through the activation layer to obtain the initial feature sequence data of the highway slope.

[0028] In this embodiment, the expression of the causal convolution layer is: ; where, is the temporal feature at the t-th moment in the highway slope time series data; t is the t-th moment of the highway slope time series data; K is the convolution kernel size of the causal convolution layer; is the convolution kernel weight of the causal convolution layer; is the highway slope data at the (t - k)-th of the highway slope time series data; is the bias term of the causal convolution layer; The expression of the dilated convolution layer is: ; where, is the dilated feature at the t-th moment in the highway slope time series data; t is the t-th moment of the highway slope time series data; K is the convolution kernel size of the dilated convolution layer; is the temporal feature at the (t - d*k)-th moment in the highway slope time series data, and d is the dilation factor; is the convolution kernel weight of the dilated convolution layer, and ; is the bias term of the dilated convolution layer, and ; The expression of the residual connection layer is: ; wherein, is the connection feature at the t-th moment in the time series data of the highway slope; is the time series feature at the t-th moment in the time series data of the highway slope; is the expansion feature at the t-th moment in the time series data of the highway slope; The expression of the activation layer is: ; wherein, is the initial feature at the t-th moment in the initial feature sequence data of the highway slope; is the connection feature at the t-th moment in the time series data of the highway slope; is the ReLU activation function.

[0029] In this embodiment, after the activation layer, an attention mechanism layer is further included; The attention mechanism layer is used to calculate the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculate the attention weight of the initial feature data at each moment according to the attention score of the initial feature data at each moment, and then multiply the attention weight of the initial feature data at each moment by the initial feature data at each moment to obtain the enhanced initial feature data at each moment.

[0030] In this embodiment, calculating the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculating the attention weight of the initial feature data at each moment according to the attention score of the initial feature data at each moment includes: For the initial feature data at each moment, perform a linear transformation on the initial feature data through the learnable weight matrix and bias vector of the attention mechanism to obtain the transformed initial feature data; Perform a non-linear activation on the transformed initial feature data through the hyperbolic tangent activation function to obtain the activated initial feature data; Perform a scalar score conversion on the activated initial feature data through a preset learnable weight vector to obtain the attention score of the initial feature data; Perform an exponential processing on the attention score of each initial feature data to obtain the attention score exponential value of each initial feature data, and calculate the sum value of the attention score exponents of all initial feature data; Divide the attention score exponential value of each initial feature data by the sum value of the attention score exponents of all initial feature data to obtain the attention weight of each initial feature data.

[0031] In this embodiment, the calculation formula of the attention score is as follows: ; where, is the attention score of the initial feature data at the t-th moment; is the transpose of the learnable weight vector , The dimension of is , which is the internal hidden dimension of the attention mechanism; is the hyperbolic tangent activation function; is the learnable weight matrix, and the dimension is , is the dimension of; is the learnable bias vector, and the dimension is ; The calculation formula of the attention weight is as follows: ; where, is the attention weight of the initial feature data at the t-th moment; is the attention score of the initial feature data at the t-th moment; is the attention score of the initial feature data at the i-th moment; T is the total number of moments.

[0032] In this embodiment, before the prediction stage of the LSTM model, the causal convolution layer, dilated convolution layer and residual connection layer of the TCN model are used to extract features from the highway slope time series data, providing high-accuracy feature data for the subsequent LSTM model to predict highway slope instability, solving the problem that the LSTM model is difficult to capture long-range dependence relationships and local mutation features, avoiding directly using the LSTM model to predict the instability probability of highway slopes, and improving the accuracy of the LSTM model in predicting highway slope instability.

[0033] Step S4: Input the initial feature sequence data into the trained LSTM model, and use the trained LSTM model to predict the instability of the highway slope for the initial feature sequence data, obtaining the instability prediction probability value of the highway slope; In this embodiment, the LSTM model includes: an input gate, a forget gate, a candidate memory unit, an updated memory unit, an output gate, a hidden state update unit and a fully connected layer.

[0034] In this embodiment, the step of inputting the initial feature sequence data into the trained LSTM model and using the trained LSTM model to predict the instability of the highway slope for the initial feature sequence data, obtaining the instability prediction probability value of the highway slope, includes: For each initial feature sequence at each moment of the initial feature sequence data, a linear transformation and a non-linear activation of the input gate are performed according to the initial feature data at the current moment and the hidden state at the previous moment through the input gate, and the input gate value at the current moment is obtained; A linear transformation and a non-linear activation of the forget gate are performed according to the initial feature data at the current moment and the hidden state at the previous moment through the forget gate, and the forget gate value at the current moment is obtained; A linear transformation and a hyperbolic tangent non-linear activation are performed on the candidate memory cell according to the initial feature data at the current moment and the hidden state at the previous moment, and the candidate memory data at the current moment is obtained; Multiply the input gate value at the current moment by the candidate memory data at the current moment to obtain a new memory increment data, and multiply the forget gate value at the current moment by the memory update data at the previous moment to obtain the retained memory data. Add the new memory increment data and the retained memory data to calculate the memory update data of the update memory cell at the current moment; A linear transformation and a non-linear activation of the output gate are performed according to the initial feature data at the current moment and the hidden state at the previous moment through the output gate, and the output gate value at the current moment is obtained; Perform a hyperbolic tangent non-linear activation on the memory update data at the current moment to obtain the activated memory update data, and calculate the hidden state of the hidden state update unit at the current moment according to the activated memory update data and the output gate value at the current moment; Input the hidden state at the last moment of the initial feature time series data into the fully connected layer, and perform a linear transformation and a non-linear activation on the hidden state at the current moment through the fully connected layer to generate the instability prediction probability value of the highway slope.

[0035] In this embodiment, the calculation expressions of the input gate, the forget gate, the candidate memory cell, the update memory cell, the output gate, the hidden state update unit, and the fully connected layer are as follows: The expression of the input gate is: ; The expression of the forget gate is: ; The expression of the candidate memory cell is: ; The expression of the update memory cell is: ; The expression of the output gate is: ; The expression of the hidden state update unit is: ; The expression of the fully connected layer is: ; Where 、 , , and They are the learnable input gate, forget gate, candidate memory unit, output gate, and weight matrix of the fully connected layer in the LSTM model; , , , and They are the learnable input gate, forget gate, candidate memory unit, output gate, and bias vector of the fully connected layer in the LSTM model; is the initial feature of the initial feature sequence data of the highway slope at the tth moment; is the hidden state at the previous moment; is the hidden state at the current time t; is the Sigmoid function; is the hyperbolic tangent activation function; is the hidden state of the initial feature time series data at the last moment; is the predicted probability value of highway slope instability.

[0036] In this embodiment, the initial feature sequence data extracted by the trained TCN model is analyzed through the trained LSTM model, and its gating mechanism is used to effectively capture the long-term dependency of the data, adapt to the dynamic changes of the slope state, and deeply explore the potential associations based on high-quality feature input. The quantitative instability prediction probability value is output, the accuracy of slope instability prediction is improved, and a scientific decision-making basis is provided for highway management departments, so that they can reasonably allocate resources and prevent risks in advance, thereby effectively ensuring highway traffic safety and reducing various losses caused by slope instability.

[0037] In this embodiment, the model training of the TCN model and the LSTM model includes: Obtain historical data on the inclination angle, soil moisture, displacement, geology and actual instability assessment value of the highway slope, and perform data preprocessing on the acquired data to obtain processed historical data on the highway slope; The processed highway slope historical data are divided into a number of highway slope time series historical data of equal time periods according to a preset time period, and then a highway slope time series historical data set is constructed; The highway slope time series historical data set is divided into a training set and a test set; The initial feature sequence data of the highway slope is obtained by extracting the initial features of the training set through the causal convolution layer, dilated convolution layer, residual connection layer, activation layer and attention mechanism layer of the TCN model to be trained; Use the LSTM model to be trained to predict the instability of highway slopes based on the initial feature sequence data of highway slopes, and obtain the instability prediction probability value of highway slopes; According to the instability prediction probability value and the actual evaluation value of highway slope instability, calculate the loss value, and optimize the parameters of the TCN model and LSTM model to be trained according to the loss value until the loss value converges, and obtain the preliminarily trained TCN model and LSTM model; Input the validation set into the preliminarily trained TCN model, extract the initial features according to the validation set through the preliminarily trained TCN model to obtain the initial feature sequence data of the highway slope, and use the preliminarily trained LSTM model to predict the instability of the highway slope based on the initial feature sequence data of the highway slope, and obtain the instability prediction probability value of the highway slope output during model validation. Calculate the evaluation index according to the instability prediction probability value of the highway slope output during model validation and the actual evaluation value of instability in the validation set, and optimize the parameters of the preliminarily trained TCN model and LSTM model according to the evaluation index until the evaluation index reaches the preset standard, and obtain the trained TCN model and LSTM model.

[0038] In this embodiment, the data preprocessing includes but is not limited to data cleaning and data normalization; among them, the data cleaning includes missing value cleaning and outlier cleaning; the missing value cleaning can be filled by interpolation methods (such as linear interpolation, spline interpolation); the outlier cleaning can be identified and corrected according to the statistical characteristics of the data (such as mean, standard deviation), or directly deleted; the data normalization methods include min-max normalization method and Z-score normalization method.

[0039] In this embodiment, the loss function includes but is not limited to mean square error loss function, mean absolute error loss function and cross entropy loss function.

[0040] In this embodiment, the ways of parameter optimization include but are not limited to gradient descent method and optimizer optimization.

[0041] In this embodiment, the evaluation indexes include but are not limited to precision, recall rate and average F1.

[0042] Step S5: Determine the hierarchical warning result according to the instability prediction probability value of the highway slope, and then process the highway slope according to the hierarchical warning result.

[0043] In this embodiment, the determining the hierarchical warning result according to the instability prediction probability value of the highway slope, and then processing the highway slope according to the hierarchical warning result includes: Compare the instability prediction probability value of the highway slope with the upper and lower index values of the preset classification warning intervals to determine the warning level. Among them, the preset classification warning intervals include: low-risk warning interval, medium-risk warning interval, relatively high-risk warning interval, and high-risk warning interval. The upper index value of the low-risk warning interval is less than the lower index value of the medium-risk warning interval. The upper index value of the medium-risk warning interval is less than the lower index value of the relatively high-risk warning interval. The upper index value of the relatively high-risk warning interval is less than the lower index value of the high-risk warning interval. When the warning level is low risk, conduct routine monitoring of the highway slope. When the warning level is medium risk, shorten the monitoring period of the highway slope, notify technical personnel to conduct on-site inspections, and formulate a preliminary emergency response plan. When the warning level is relatively high risk, conduct continuous monitoring of the highway slope, implement traffic control in the surrounding area of the slope, and notify the emergency rescue team. When the warning level is high risk, notify the personnel within the influence range of the slope to evacuate, and notify the emergency rescue team to conduct emergency rescue.

[0044] Embodiment 2: Refer to Figure 2 , which is a module diagram of a monitoring and warning system for highway slope stability provided by an embodiment of the present invention. In order to solve the problem in the prior art that the direct use of the LSTM model for prediction fails to fully consider the long-range dependence relationship and the extraction of local mutation characteristics in the time series data, resulting in low accuracy of highway slope instability prediction. The system at least includes: a data acquisition module, an image acquisition module, a feature extraction module, a slope instability prediction module, and a slope instability warning module. The data acquisition module is used to acquire the inclination angle data, soil humidity data, displacement data, and geological data of the highway slope at each moment within the currently preset time period, and construct an initial feature dataset. The image acquisition module is used to collect remote sensing images of the highway slope in real time, divide the remote sensing images into cells, and mark corresponding coordinate values for each cell. According to the initial feature dataset, weight calculation is performed on each coordinate value, so that the coordinate values corresponding to each cell are updated, and the highway slope time series data composed of the corresponding values of each cell is obtained. The feature extraction module is used to input the highway slope time series data into the trained TCN model, and perform initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer, and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope. The slope instability prediction module is used to input the initial feature sequence data into the trained LSTM model, and predict the highway slope instability through the trained LSTM model for the initial feature sequence data, so as to obtain the instability prediction probability value of the highway slope; The slope instability warning module is used to determine the hierarchical warning result according to the instability prediction probability value of the highway slope, and then process the highway slope according to the hierarchical warning result.

[0045] 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. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A monitoring and early warning method for highway slope stability, characterized in that: include: Obtain the inclination angle data, soil moisture data, displacement data and geological data of the highway slope at each moment in the current preset time period, and construct an initial feature number set; Collecting remote sensing images of the road slope in real time, dividing the remote sensing images into cells and marking corresponding coordinate values ​​for each cell; Performing weight calculation on each coordinate value according to the initial feature number set, so that the coordinate value corresponding to each cell mark is updated, and obtaining the highway slope time series data composed of the values ​​corresponding to each cell; The highway slope time series data is input into the trained TCN model, and the initial features of the highway slope time series data are extracted through the causal convolution layer, dilated convolution layer, residual connection layer and activation layer of the trained TCN model to obtain the initial feature sequence data of the highway slope; The initial feature sequence data is input into the trained LSTM model, and the highway slope instability prediction is performed on the initial feature sequence data through the trained LSTM model to obtain the highway slope instability prediction probability value; According to the predicted probability value of highway slope instability, the graded warning results are determined, and then the highway slope is processed according to the graded warning results.

2. A monitoring and early warning method for highway slope stability according to claim 1, characterized in that: The step of collecting the remote sensing image of the highway slope in real time, dividing the remote sensing image into cells and marking the corresponding coordinate value of each cell specifically includes: Normalizing the collected remote sensing images, identifying preset reference points on the remote sensing images, setting the preset reference points identified in each frame of remote sensing images as alignment points, and performing feature alignment on each frame of remote sensing images; By using a rectangular coordinate system, taking the alignment point of each frame of remote sensing image as the origin of the rectangular coordinate system, dividing the coordinate axis, performing numerical unit analysis on the initial feature number set, and determining the cell interval size of the coordinate axis division according to the change range of the initial feature number set within a unit time; The coordinate value corresponding to each cell mark is determined according to the interval size after each cell is divided.

3. A monitoring and early warning method for highway slope stability according to claim 2, characterized in that: The step of performing weight calculation on each coordinate value according to the initial feature number set so as to update the coordinate value corresponding to each cell mark and obtain the highway slope time series data composed of the values ​​corresponding to each cell specifically includes: According to the coordinate values ​​corresponding to each cell mark, respectively calculating the tilt angle data, soil moisture data, displacement data and geological data, and the spatial distance values ​​between the alignment points; Each coordinate value is sequentially superimposed and calculated according to the spatial distance value, so that the interval size of each cell changes, thereby updating the coordinate value corresponding to each cell mark; According to the updated data sorting, the highway slope time series data consisting of the corresponding values ​​of each cell are obtained.

4. A monitoring and early warning method for highway slope stability according to claim 1, characterized in that: The initial feature sequence data of the highway slope is obtained by performing initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer and activation layer of the trained TCN model, including: The time series features of the highway slope time series data are extracted through the causal convolution layer to obtain the time series features of each moment in the highway slope time series data; The time series features of each moment in the highway slope time series data are expanded through the dilated convolution layer to obtain the expanded features of each moment in the highway slope time series data; The time series features and expansion features of each moment in the highway slope time series data are added by the residual connection layer according to the time step, so as to obtain the connection features of each moment in the highway slope time series data; The activation layer is used to perform nonlinear activation on the connection features at each moment in the highway slope time series data to obtain the initial feature sequence data of the highway slope.

5. A monitoring and early warning method for highway slope stability according to claim 4, characterized in that: The expression of the causal convolutional layer is: ； in, is the time series feature of the highway slope time series data at the tth moment; t is the tth moment of the highway slope time series data; K is the convolution kernel size of the causal convolution layer; is the convolution kernel weight of the causal convolution layer; is the tkth highway slope data of the highway slope time series data; is the bias term of the causal convolutional layer; The expression of the dilated convolutional layer is: ； in, is the expansion feature at the tth moment in the highway slope time series data; t is the tth moment in the highway slope time series data; K is the convolution kernel size of the expansion convolution layer; is the time series characteristic of the td*kth moment in the highway slope time series data, and d is the expansion factor; is the convolution kernel weight of the dilated convolution layer, and ; is the bias term of the dilated convolutional layer, and ; The expression of the residual connection layer is: ； in, is the connection feature at time t in the highway slope time series data; is the time series feature of the highway slope time series data at the tth moment; is the expansion characteristic at the tth moment in the highway slope time series data; The expression of the activation layer is: ； in, is the initial feature of the initial feature sequence data of the highway slope at the tth moment; is the connection feature at time t in the highway slope time series data; is the ReLU activation function.

6. A monitoring and early warning method for highway slope stability according to claim 5, characterized in that: After the activation layer, it also includes: an attention mechanism layer; The attention mechanism layer is used to calculate the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculate the attention weight of the initial feature data at each moment based on the attention score of the initial feature data at each moment, and then multiply the attention weight of the initial feature data at each moment by the initial feature data at each moment to obtain the enhanced initial feature data at each moment.

7. A monitoring and early warning method for highway slope stability according to claim 6, characterized in that: The method of calculating the attention score of the initial feature data at each moment in the initial feature sequence data of the highway slope, and calculating the attention weight of the initial feature data at each moment according to the attention score of the initial feature data at each moment, comprises: For the initial feature data at each moment, the initial feature data is linearly transformed through the learnable weight matrix and bias vector of the attention mechanism to obtain the transformed initial feature data; The transformed initial feature data is nonlinearly activated by a hyperbolic tangent activation function to obtain activated initial feature data; The activated initial feature data is converted into a scalar score through a preset learnable weight vector to obtain an attention score of the initial feature data; The attention score of each initial feature data is indexed to obtain the attention score index value of each initial feature data, and the attention score index and value of all initial feature data are calculated; The attention score index value of each initial feature data is divided by the sum of the attention score indexes of all initial feature data to obtain the attention weight of each initial feature data.

8. A monitoring and early warning method for highway slope stability according to claim 7, characterized in that: The LSTM model includes: an input gate, a forget gate, a candidate memory unit, an update memory unit, an output gate, a hidden state update unit and a fully connected layer; The initial feature sequence data is input into the trained LSTM model, and the trained LSTM model is used to predict the instability of the highway slope, so as to obtain the instability prediction probability value of the highway slope, including: For each initial feature sequence of the initial feature sequence data, the input gate is linearly transformed and nonlinearly activated according to the initial feature data at the current moment and the hidden state at the previous moment through the input gate to obtain the input gate value at the current moment; The forget gate is used to perform linear transformation and nonlinear activation of the forget gate according to the initial feature data at the current moment and the hidden state at the previous moment, and the forget gate value at the current moment is obtained; The candidate memory unit is linearly transformed and hyperbolic tangent nonlinearly activated according to the initial feature data at the current moment and the hidden state at the previous moment to obtain the candidate memory data at the current moment; The input gate value at the current moment is multiplied by the candidate memory data at the current moment to obtain new memory increment data, and the forget gate value at the current moment is multiplied by the memory update data at the previous moment to obtain the retained memory data, and the memory update data of the updated memory unit at the current moment is obtained by adding the new memory increment data and the retained memory data; The output gate is used to perform linear transformation and nonlinear activation according to the initial feature data at the current moment and the hidden state at the previous moment, and the output gate value at the current moment is obtained; Performing hyperbolic tangent nonlinear activation on the memory update data at the current moment to obtain activated memory update data, and calculating the hidden state of the hidden state update unit at the current moment according to the activated memory update data and the output gate value at the current moment; The hidden state of the last moment of the initial feature time series data is input into the fully connected layer, and the hidden state at the current moment is linearly transformed and nonlinearly activated through the fully connected layer to generate the instability prediction probability value of the highway slope.

9. A monitoring and early warning method for highway slope stability according to claim 8, characterized in that: The step of determining the graded warning result according to the predicted probability value of the instability of the highway slope, and then processing the highway slope according to the graded warning result, includes: The predicted probability value of instability of the highway slope is compared with the upper limit index and the lower limit index of the preset graded warning interval to determine the warning level; wherein the preset graded warning interval includes: a low-risk warning interval, a medium-risk warning interval, a relatively high-risk warning interval and a high-risk warning interval; the upper limit index of the low-risk warning interval is less than the lower limit index of the medium-risk warning interval; the upper limit index of the medium-risk warning interval is less than the lower limit index of the relatively high-risk warning interval; the upper limit index of the relatively high-risk warning interval is less than the lower limit index of the high-risk warning interval; When the warning level is low risk, routine monitoring of the highway slopes is carried out; When the warning level is medium risk, the monitoring period of the highway slope is shortened, and technical personnel are notified to conduct on-site inspections and formulate preliminary emergency response plans; When the warning level is high risk, the highway slopes are continuously monitored, traffic control is implemented in the area around the slopes, and emergency teams are notified; When the warning level is high risk, people within the slope impact area will be notified to evacuate, and the rescue team will be notified to carry out emergency rescue.

10. A monitoring and early warning system for highway slope stability, characterized in that: include: Data acquisition module, image acquisition module, feature extraction module, slope instability prediction module and slope instability warning module; The data acquisition module is used to acquire the inclination angle data, soil moisture data, displacement data and geological data of the highway slope at each moment in the current preset time period, and to construct an initial feature number set; The image acquisition module is used to acquire remote sensing images of the road slope in real time, divide the remote sensing images into cells and mark the corresponding coordinate values ​​for each cell; Performing weight calculation on each coordinate value according to the initial feature number set, so that the coordinate value corresponding to each cell mark is updated, and obtaining the highway slope time series data composed of the values ​​corresponding to each cell; The feature extraction module is used to input the highway slope time series data into the trained TCN model, perform initial feature extraction on the highway slope time series data through the causal convolution layer, dilated convolution layer, residual connection layer and activation layer of the trained TCN model, and obtain the initial feature sequence data of the highway slope; The slope instability prediction module is used to input the initial feature sequence data into the trained LSTM model, and perform highway slope instability prediction on the initial feature sequence data through the trained LSTM model to obtain the highway slope instability prediction probability value; The slope instability warning module is used to determine the graded warning results according to the predicted probability value of the highway slope instability, and then process the highway slope according to the graded warning results.

Images