Slope safety and stability analysis method based on integrated sensor and intelligent algorithm

Abstract

A slope safety and stability analysis method based on an integrated sensor and an intelligent algorithm comprises the following steps: acquiring initial values of geometric parameters (including height, gradient and width) of a slope and basic physical parameters (weight, internal friction angle and cohesive force) of a soil body through a method of combining in-situ static sounding and an indoor test; an improved integrated sensor slope state monitoring device is adopted to continuously obtain state parameters such as water content, matrix suction, internal three-way displacement, an inclination angle and soil pressure of a slope soil body, and multi-dimensional dynamic monitoring is achieved; basic physical parameters of a soil body are updated in real time based on data collected by a sensor, and the slope stability safety coefficient is predicted through a semi-supervised learning model; safety early warning is triggered in a grading mode according to prediction results, model prediction values and measured values are recorded and analyzed, and the system evaluates the long-term stability and applicability of the model; through multi-source data fusion and intelligent analysis, theoretical support and technical guarantee are provided for safety monitoring and early warning of the loess slope.

Description

Technical Field

[0001] The present invention relates to the field of soil slope status monitoring and analysis along railways, and in particular to a slope safety and stability analysis method based on integrated sensors and intelligent algorithms. Background Art

[0002] At present, my country's railway transportation has transitioned from large-scale construction to healthy operation and maintenance. Loose shallow slopes are widely distributed along mountain railways. With the frequent occurrence of extreme rainfall in recent years, the frequency and harm caused by landslide disasters have become more and more significant. Therefore, it is of great practical significance to obtain the physical and mechanical state of the slope in a timely manner, conduct efficient data analysis and processing, and reasonably predict the safety and stability of the slope. Most of the existing slope disaster warning methods are based on single monitoring data. The slope monitoring device has low integration and can only reflect the displacement or moisture content change of the slope. It ignores the physical-mechanical mechanism of soil slope sliding, resulting in poor slope state monitoring accuracy, inaccurate landslide warning time, and especially difficult to accurately predict the sliding surface. Using integrated sensors and intelligent algorithm technology to analyze the safety and stability of soil slopes can improve the problem of single traditional sensor monitoring equipment and low data accuracy, improve the accuracy of landslide warning time, and reduce human and material investment, so as to realize the promotion and application of intelligent slope monitoring systems. Summary of the invention

[0003] In order to solve the defects in the prior art, the present invention discloses a slope safety stability analysis method based on integrated sensors and intelligent algorithms, and its technical solution is as follows: S1, obtain the height, slope and width geometric dimension information of the slope through machine vision and drone measurement technology, and obtain the unsaturated shear strength parameters of the slope soil, such as the initial values ​​of the internal friction angle and cohesion, through a portable unsaturated direct shear-pressure plate integrated machine; S2, using an improved sensor-integrated slope condition monitoring device to continuously obtain the soil moisture content, matrix suction, inclination angle, overburden soil pressure and other state parameters at each measuring point of the slope; S3, based on the data collected by the sensors, calculates and updates the basic physical parameters of the soil in real time, and uses a semi-supervised learning model to predict the stability safety factor of the loess slope; S4, trigger safety warnings in different levels according to the model prediction results. And continuously record the model prediction results and measured results to evaluate the long-term stability of the model.

[0004] The present invention also discloses a slope safety and stability analysis device based on integrated sensors and intelligent algorithms, which is suitable for monitoring soil slope state parameters. The device is based on the above-mentioned slope safety and stability analysis method based on integrated sensors and intelligent algorithms, and is characterized by: It includes multiple measurement modules, which are evenly arranged along the slope surface. According to the on-site survey results, the soil state parameters of the shallow landslide-prone area are monitored by the measurement modules within a certain depth range from the slope surface, and transmitted to the signal base station via wireless signals, and then uploaded to the cloud for real-time data processing and analysis; The measurement module includes an inclination measurement module, a moisture content measurement module, a matrix suction measurement module, and an earth pressure measurement module, wherein: Inclination measurement module: used to monitor the inclination changes of the soil at each measuring point on the slope; the main components of the measurement module include a fixed steel frame, a three-axis accelerometer, a signal transmission system and a power supply system. The working principle is: when the soil slides downward, the soil drives the integrated device to move together, and the device as a whole tilts at a certain angle. The spatial displacement and angle change of the soil relative to the initial position are measured by the inclinometer; Moisture content measurement module: used to monitor the change of soil moisture content within a certain range of each measuring point on the slope. The main components of this module include a signal transmitter, a signal receiver and an oscillator chip. The modules are evenly arranged along the vertical direction of the device according to user needs, and the position is flexible and adjustable. The working principle is: the signal transmitter first sends a sweep test signal of a specific frequency band to the surrounding soil; the upper and lower ends of the signal transmitter are composed of a pair of circular metal rings to form a ring capacitor structure, the surrounding soil acts as a dielectric, and the sensor capacitance is proportional to the dielectric constant of the surrounding soil medium; when the moisture in the soil increases, its dielectric constant increases accordingly, and the measured capacitance value also increases, resulting in a change in the measurement frequency, thereby measuring the moisture content of the soil; The matrix suction measurement module is used to monitor the changes in soil matrix suction within a certain range of each measuring point on the slope; the main components of this module include a sealed plastic tube, a porous ceramic head, a negative pressure gauge and a wireless signal transmitter. The working principle is: after the porous ceramic head comes into contact with the surrounding soil, the moisture in the soil and the plastic tube is exchanged according to the water potential difference; when the moisture content of the soil increases, the moisture enters the plastic tube through the ceramic head, and when the moisture content of the soil decreases, the moisture enters and exits the plastic tube through the ceramic head, and a vacuum negative pressure is generated in the tube; based on the monitoring data of the negative pressure gauge and the calibration table, the matrix suction of the soil can be inversely calculated.

[0005] The invention also discloses a slope safety stability analysis system based on integrated sensors and intelligent algorithms, which is characterized in that the system comprises the above method or device. Beneficial Effects

[0006] (1) Based on the multi-sensor integrated device, the parameters required for slope safety and stability analysis can be obtained efficiently and conveniently, meeting the needs of the back-end software to predict the slope safety factor; (2) Based on the semi-supervised learning model, a large amount of real-time sensor data is used to predict the slope stability safety factor and perform graded safety warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a schematic diagram of the field application of the slope data acquisition system of the present invention; Figure 2 It is a structural diagram of a multi-sensor integrated device; Figure 3 Flowchart of data transmission and processing for multi-sensor integrated device; Figure 4 A schematic diagram of the semi-supervised learning model framework used in the present invention; Figure 5 It is the flow chart of the intelligent algorithm of the present invention. DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. These embodiments are provided to make the present invention more thoroughly and completely conveyed to those skilled in the art.

[0009] In the first aspect, the embodiment of the present invention provides a multi-sensor integrated device suitable for monitoring soil slope state parameters, the field application of the device and the internal structure schematic diagram are respectively as follows: Figure 1 and Figure 2 As shown, specifically including: The multi-sensor integrated soil slope state parameter monitoring device can be arranged according to user needs, and can generally be evenly arranged along the slope surface. Based on the results of on-site surveys, the device can monitor the soil state parameters of shallow landslide-prone areas within a certain depth range from the slope surface. Various measuring elements such as inclination, moisture content, saturation, and soil pressure are transmitted to the signal base station through the wireless signal device on the top of the device, and then uploaded to the cloud for real-time data processing and analysis. The device is powered by solar panels.

[0010] The specific measurement modules are as follows: (1) Inclination measurement module, used to monitor the inclination changes of the soil at each measuring point on the slope. The main components of the measurement module include a fixed steel frame, a three-axis accelerometer, a signal transmission system, and a power supply system. The working principle of the module is as follows: when the soil slides downward, the soil drives the integrated device to move together, and the device as a whole tilts at a certain angle. The spatial displacement and angle change of the soil relative to the initial position are measured by the inclinometer.

[0011] The calculation process of the spatial displacement and angle change value of any point inside the slope is as follows: (1) (2) (3) (4) in, x 1, x 2 and x 3 are the displacements in three directions in the geodetic coordinate system; a 1, a 2 and a 3 are the acceleration measurements along the three directions in the geodetic coordinate system; t 1 and t 2 are the start and end times of the measurement respectively.

[0012] (5) in, α i (i=1, 2) are the rotation angles along the x and y directions of the geodetic coordinate system, respectively; ai(i=1, 2) are the accelerations along the x and y directions of the geodetic coordinate system, respectively; and L is the length of the measuring device.

[0013] (6) Among them, X i is the displacement of any point along the depth direction of the slope, L i is the vertical distance from the point to the anchor rod.

[0014] (2) Moisture content measurement module, which is used to monitor the changes in soil moisture content within a certain range of each measuring point on the slope. The main components of the measurement module include a signal transmitter, a signal receiver and an oscillator chip. The modules are evenly arranged along the vertical direction of the device according to user needs, and the position can be flexibly adjusted. The working principle of the module is as follows: the signal transmitter first sends a sweep test signal of a specific frequency band to the surrounding soil; the upper and lower ends of the signal transmitter are composed of a pair of circular metal rings to form a ring capacitor structure, the surrounding soil acts as a dielectric, and the sensor capacitance is proportional to the dielectric constant of the surrounding soil medium; when the moisture in the soil increases, its dielectric constant increases accordingly, and the measured capacitance value also increases, causing the measurement frequency to change, thereby measuring the moisture content of the soil.

[0015] The specific calculation process is as follows: In the module, the sine wave generator generates a sine signal as the signal source, and the phase conversion circuit and the impedance conversion circuit form a detection circuit. The signal source generates a current in the impedance input channel. i s , generates a current in the reference channel i r, the impedance is converted to Z x Connect the probe to the impedance input channel. Z x It can be equivalent to a capacitor and a conductor in parallel, and its phase will change with the reference current. i r changes with the changes of.

[0016] When the reference current passes through a phase converter resistor or capacitor, it generates an impedance converter output voltage u s There is a certain voltage u r , these two voltages can be multiplied to obtain a continuous sinusoidal signal containing the second harmonic.

[0017] (7) (8) (9) By filtering out the DC component through a low-pass filter, equation (9) becomes: (10) By setting the phase angle of the reference circuit β , so that it is equal to 0 and π / 2 respectively, we get: (11) (12) The equivalent capacitance and conductance of the probe impedance can be calculated by phase transformation of the reference channel. The size of the equivalent capacitance and conductance is ultimately reflected in the dielectric constant and conductivity. For the measurement of the internal moisture content of the slope, the dielectric constant is mainly determined by the soil moisture content. Therefore, the soil moisture content can be obtained by calculating the equivalent capacitance and conductance of the probe impedance.

[0018] (3) Matrix suction measurement module, used to monitor the changes in soil matrix suction within a certain range of each measuring point on the slope. The main components of the measurement module include a sealed plastic tube, a porous ceramic head, a negative pressure gauge, and a wireless signal transmitter. The working principle of the module is as follows: after the porous ceramic head contacts the surrounding soil, the water in the soil and the plastic tube is exchanged according to the water potential difference; when the soil moisture content increases, the water enters the plastic tube through the ceramic head; when the soil moisture content decreases, the water enters and exits the plastic tube through the ceramic head, and a vacuum negative pressure is generated in the tube; based on the monitoring data of the negative pressure gauge and the calibration table, the matrix suction of the soil can be inverted.

[0019] Furthermore, in order to avoid the measurement error caused by a single fixed measurement position, by setting up lateral and vertical traction systems and based on the soil pressure feedback information, such as the depth position of the soil pressure peak, the up and down movement of the suction meter is controlled by the built-in chip, so that the active measurement of the matrix suction at any point inside the slope can be realized, and the physical and mechanical information at the potential failure surface inside the slope can be effectively obtained.

[0020] (4) Soil pressure measurement module, used to monitor the size and distribution changes of the soil pressure of the overlying soil at each measuring point on the slope. The main components of the measurement module include a flexible pressure sensor and a signal transmitter. The working principle of the module is as follows: the flexible pressure sensor is attached to the outer surface of the integrated device and is in direct contact with the overlying soil. When the deadweight of the overlying soil changes, the overlying soil exerts soil pressure on the sensor element, and the piezoresistance and piezocaisture mixed response in the sensor element changes. The soil pressure value is obtained by measuring the change in the signal. Specifically, the flexible pressure sensor is a layered structure, which is made of a porous conductive nanocomposite material and an ultra-thin dielectric layer through a lamination process. It has a wide pressure range and high sensitivity. Furthermore, at present, flexible pressure sensors are widely used in soft robots, wearable devices and deformable touch panels, but have not been used in the monitoring of slope soil pressure.

[0021] In the second aspect, the embodiment of the present invention provides a slope safety stability analysis method based on multi-source monitoring data, the algorithm flow is as follows: Figure 5 As shown, including: S1, through machine vision and drone measurement technology, obtain the geometric dimension information of the slope, such as height, slope and width, and obtain the unsaturated shear strength parameters of the slope soil, such as soil weight, cohesion and initial value of internal friction angle, through a portable unsaturated direct shear instrument; at the same time, use sensors to continuously obtain the moisture content, matrix suction and effective normal stress state parameters of the soil at each measuring point of the slope to update the unsaturated shear strength parameters in real time, and create an algorithm data set; Specifically, a mapping relationship from state parameters to basic physical parameters is established, and the soil weight, internal friction angle and cohesion are updated through real-time calculation. The formulas for updating the soil weight, internal friction angle and cohesion are as follows: (13) in, For updated soil weight, For dry soil weight, is the moisture content, is the density of water (usually 9.8 ) (14) in, For the updated cohesion, is the initial cohesion, is the matrix suction, is the sensitivity coefficient of matrix suction to cohesion (calibrated through experiments).

[0022] (15) in, is the updated internal friction angle, is the initial internal friction angle, is the moisture content, is the effective normal stress, , It is an empirical parameter (calibrated through experiments).

[0023] Specifically, the real-time updated soil weight, cohesion and internal friction angle parameters can be divided into two categories to produce algorithm data sets: the first category is labeled data, that is, the theoretical value of slope stability coefficient verified by strict mathematical formulas and using existing historical data. The second category is unlabeled data, that is, a large amount of real-time sensor data, and the safety factor has not yet been calculated.

[0024] Furthermore, by establishing the mechanical equilibrium equation of sliding soil, the calculation formula of slope safety factor under the framework of unsaturated soil is derived. Specifically, on the determined sliding surface, the sliding soil can be divided into n Vertical bars of equal width, any i Mechanical analysis of the blocks: (1) Performing force balance analysis along the normal direction of the sliding surface, we have: (16) In the formula, is the normal force on the bottom of the sliding soil, is the deadweight of soil in different saturated states, and are the shear forces on the left and right sides of the strip, and are the normal forces on the left and right sides of the selected strip, is the angle between the bottom surface of the strip and the horizontal plane.

[0025] (2) Performing force balance analysis along the tangent direction of the sliding surface, we have: (17) In the formula, is the anti-sliding force of the bottom surface of the sliding soil, is the deadweight of soil in different saturated states, and are the shear forces on the left and right sides of the strip, and are the normal forces on the left and right sides of the selected strip, is the angle between the bottom surface of the strip and the horizontal plane, is the slope safety factor.

[0026] in, It can be obtained by multiplying the weight of soil under different water content conditions by the volume of the strips taken. (18) In the formula, is the height of the bar, is the width of the bar.

[0027] It can be obtained by multiplying the theoretical calculated value of unsaturated shear strength by the sliding surface area. Specifically, (19) In the formula, is the updated internal friction angle, For the updated cohesion, is the width of the bar taken.

[0028] Furthermore, the real-time update formulas for the weight, internal friction angle and cohesion of the soil under unsaturated conditions are shown in equations (13) to (15). Substituting equations (13) to (15) into The calculation formula of is, (20) Furthermore, when the anti-sliding force of the strip is less than the sum of its own weight and the mass of the overlying soil, the soil begins to become unstable, and the anti-sliding force continues to decrease until the slope failure surface is formed. The slope safety factor can be derived as: Substituting equation (16) and equation (20) into equation (17), we have: (twenty one) Furthermore, data processing can be performed through steps such as data cleaning, data synchronization, data normalization, and data sliding window. The outliers in the above data are processed, and the multi-sensor data is time-aligned and each variable is standardized to facilitate subsequent input into the model.

[0029] S2, using two types of data to train the semi-supervised learning model and establish a slope stability safety factor prediction model;

[0030] Specifically, a self-training model can be used to predict the slope safety factor. The model framework is as follows: Figure 5As shown in the figure. The model uses a small amount of labeled data to train the initial model, and uses the model to predict the pseudo labels of unlabeled data. Samples with high pseudo label credibility are added to the training set, and the model is continuously updated iteratively. The core idea is to mine information from unlabeled data through the model's own prediction results, thereby improving the model's generalization ability and prediction accuracy. It is particularly suitable for scenarios where labeled data is scarce.

[0031] Furthermore, the self-training semi-supervised model usually combines supervised loss and unsupervised loss to optimize the model, and the overall objective function is: (twenty two) in, is the loss function for supervised learning (such as cross entropy), is the loss function for unsupervised learning (such as consistency loss), is the weight of the unsupervised loss, controlling its contribution to the total loss.

[0032] (twenty three) in, is the number of samples of labeled data, is a sample of labeled data, is the label of the labeled data, is the prediction result of the model, are the parameters of the prediction function, is the loss function; Pseudo labels are generated using the model’s highly confident predictions: (twenty four) in, For unlabeled samples, is the probability distribution predicted by the model for unlabeled samples, are the parameters of the prediction function, is a pseudo label; Pseudo labels are used as supervisory signals for unlabeled data, and the optimization loss is: (25) in, is the number of samples of unlabeled data, 1 () is the indicator function, which is only used when the prediction confidence is higher than the threshold Participate in training, is the confidence threshold.

[0033] Specifically, use labeled data (such as 5%~20% of samples) to train the initial model, and then use the initial model to predict pseudo labels for unlabeled data. The pseudo-label data will be screened according to the confidence of the prediction results, and the high-confidence pseudo-label data will be added to the labeled data. The model will be retrained and the process will be repeated.

[0034] Specifically, consistency regularization can be used to add disturbances to unlabeled data, such as data augmentation and noise disturbance, to force the model to output the same value for the original data and the perturbed data, thereby improving the robustness of the model. The consistency regularization formula is: (26) in, For unlabeled samples random perturbations added (such as data augmentation or noise), is the model’s prediction result for random disturbances, is the parameter of the prediction function; Usually it is the mean square error (MSE), and other parameters are as described above.

[0035] Furthermore, cross-validation is used to evaluate the model performance. Conventional indicators of regression models, such as mean square error (MSE) and mean absolute error (MAE), can be used. At the same time, the accuracy of pseudo-label prediction is verified and low-confidence samples are eliminated.

[0036] S3, real-time evaluation of slope stability using safety factor prediction model;

[0037] Specifically, it is necessary to deploy the model to the slope monitoring system, receive real-time sensor data, output the safety factor prediction value based on the input data, and classify the safety factor. When the safety factor is too small, an early warning will be issued.

[0038] Furthermore, the unlabeled data are regularly used as input, i.e., the real-time inflow of updated parameters such as soil density, internal friction angle, and cohesion, and pseudo-labels are generated for the new data. High-confidence samples are screened and added to the training. The new data are used to perform small-scale training on the existing model to avoid underfitting or overfitting of the model and achieve incremental updating of the model.

[0039] Specifically, the self-training model continuously uses pseudo labels to iteratively optimize the model, where the update process can be expressed as: Use supervised data to optimize initial model parameters: (27) in, is the loss function for supervised learning, is the initial function of the model, To label the dataset; Generate pseudo labels for unlabeled data using the current model: (28) in, Iterate for the model tThe pseudo-label of For unlabeled samples, is the model parameter of the current iteration, representing the model state, is the predicted probability distribution of the model for unlabeled samples; Update model parameters using labeled data and pseudo-labeled data: (29) in, is the loss function for supervised learning, is the loss function for unsupervised learning, is the weight of the unsupervised loss, controlling its contribution to the total loss; is the model parameter of the current iteration, representing the model state, that is, For model parameters in iteration Update results when For unlabeled samples, is the probability distribution predicted by the model, Iterate for the model t The pseudo-label of To provide a reliable training basis for the labeled dataset for supervised learning. It is an unlabeled dataset.

[0040] Formulas (28) and (29) are repeated until the model converges or the predetermined number of iterations is reached.

[0041] Specifically, if the distribution of new data changes significantly from the distribution of old data, this problem can be solved by introducing adaptive technology to periodically retrain the model.

[0042] S4, trigger safety warnings in different levels according to the model prediction results. And continuously record the model prediction results and measured results to evaluate the long-term stability of the model.

[0043] Specifically, different safety threshold levels (for example, safety, warning, alert, and danger) are set according to the slope stability safety factor calculated by the semi-supervised learning model, and corresponding safety warnings are triggered in stages based on the comparison results between the predicted safety factor and the set threshold.

[0044] Furthermore, the slope stability safety factor predicted by the model is continuously recorded, and the triggering conditions, response conditions, and actual results of each safety warning are recorded. The deviation between the model prediction value and the measured value is statistically analyzed, and its error distribution and regularity are analyzed. As the monitoring data continues to accumulate, the prediction accuracy of the analysis model remains stable over time, and the model parameters are regularly updated and corrected to ensure that it can dynamically adapt to changes in actual working conditions.

[0045] The invention also discloses a slope safety stability analysis system based on integrated sensors and intelligent algorithms, which is characterized in that the system comprises the above method or device.

[0046] In summary, the present invention provides a slope safety stability analysis method based on multi-source monitoring data. The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may have various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements all fall within the scope of the present invention to be protected. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A slope safety and stability analysis method based on integrated sensors and intelligent algorithms, characterized by: S1, obtain the geometric dimension information of the height, slope and width of the slope, and obtain the initial values ​​of the unsaturated shear strength parameters of the slope soil weight, cohesion and internal friction angle through a portable unsaturated direct shear instrument; at the same time, use sensors to continuously obtain the moisture content, matrix suction and effective normal stress state parameters of the soil at each measuring point of the slope to update the unsaturated shear strength parameters in real time and create an algorithm data set; S2, using two types of data to train the semi-supervised learning model and establish a slope stability safety factor prediction model; S3, using the safety factor prediction model to perform real-time evaluation on slope stability; S4, based on the model prediction results, trigger safety warnings in a graded manner, and continuously record the model prediction results and measured results to evaluate the long-term stability of the model.

2. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 1 is characterized by: The step S1 includes the following contents: A mapping relationship from state parameters to basic physical parameters is established, and the soil weight, internal friction angle and cohesion are updated through real-time calculation. The formulas for updating the soil weight, internal friction angle and cohesion are as follows: （13） in, For updated soil weight, For dry soil weight, is the moisture content, is the weight of water; （14） in, For the updated cohesion, is the initial cohesion, is the matrix suction, is the sensitivity coefficient of matrix suction to cohesion; （15） in, is the updated internal friction angle, is the initial internal friction angle, is the moisture content, is the effective normal stress, , is an empirical parameter.

3. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 2 is characterized by: The step S1 further includes the following contents: The parameters of soil weight, cohesion and internal friction angle updated in real time are divided into two categories to produce algorithm data sets: the first category is labeled data, which uses the finite element simulation method or empirical formula to calculate the slope stability coefficient and actual results using existing historical data; The second category is unlabeled data, including a large amount of real-time sensor data and safety factors that have not yet been calculated; Data processing is performed through data cleaning, data synchronization, data normalization, and data sliding windows; outliers in the above data set are processed, and multi-sensor data are time-aligned and each variable is standardized to facilitate subsequent input into the model.

4. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 3 is characterized in that: step S2 comprises the following contents: The semi-supervised learning model combines supervised loss and unsupervised loss to optimize the model, and the overall objective function is: （16） in, is the loss function for supervised learning, is the loss function for unsupervised learning, is the weight of the unsupervised loss, controlling its contribution to the total loss; （17） in, is the number of samples of labeled data, is a sample of labeled data, is the label of the labeled data, is the prediction result of the model, are the parameters of the prediction function, is the loss function; Pseudo labels are generated using the model’s highly confident predictions: （18） in, For unlabeled samples, is the probability distribution predicted by the model for unlabeled samples, are the parameters of the prediction function, is a pseudo label; Pseudo labels are used as supervisory signals for unlabeled data, and the optimization loss is: （19） in, is the number of samples of unlabeled data, 1 () is the indicator function, which is only used when the prediction confidence is higher than the threshold Participate in training, is the confidence threshold.

5. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 4 is characterized by: The step S2 further includes the following contents: using the labeled data to train the initial model, then using the initial model to predict pseudo labels for the unlabeled data, filtering the pseudo label data according to the confidence of the prediction results, adding the high confidence pseudo label data to the labeled data, retraining the model and repeating the process; By adding disturbances to the unlabeled data through consistency regularization, the robustness of the model is improved. The consistency regularization formula is: （20） in, For unlabeled samples The random disturbance added, Usually it is the mean square error, is the model’s prediction result for random disturbances, is the parameter of the prediction function; Cross-validation was used to evaluate model performance, verify the accuracy of pseudo-label predictions, and remove low-confidence samples.

6. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 4 is characterized by: The step S3 includes the following contents: regularly taking unlabeled data as input, i.e., real-time inflow of updated soil weight, internal friction angle and cohesion parameters, generating pseudo labels for new data, screening high-confidence samples for training, and using the new data to perform small-scale training on the existing model to achieve incremental update of the model; By establishing the mechanical equilibrium equation of sliding soil, the calculation formula of slope safety factor under the framework of unsaturated soil is derived: on the determined sliding surface, the sliding soil is divided into n Vertical bars of equal width, any i Mechanical analysis of the strips: (1) Performing force balance analysis along the normal direction of the sliding surface, we have: （16） In the formula, is the normal force on the bottom of the sliding soil, is the deadweight of soil in different saturated states, and are the shear forces on the left and right sides of the strip, and are the normal forces on the left and right sides of the selected strip, is the angle between the bottom surface of the strip and the horizontal plane; (2) Performing force balance analysis along the tangent direction of the sliding surface, we have: （17） In the formula, is the anti-sliding force of the bottom surface of the sliding soil, is the deadweight of soil in different saturated states, and are the shear forces on the left and right sides of the strip, and are the normal forces on the left and right sides of the selected strip, is the angle between the bottom surface of the strip and the horizontal plane, is the slope safety factor; in, It can be obtained by multiplying the weight of soil under different water content conditions by the volume of the strips taken. （18） In the formula, is the height of the bar, is the width of the strip; The theoretical value of the unsaturated shear strength is multiplied by the sliding surface area to obtain: （19） In the formula, is the updated internal friction angle, For the updated cohesion, is the width of the strip; The real-time update formulas for the weight, internal friction angle and cohesion of soil under unsaturated state are shown in equations (13) to (15). Substituting equations (13) to (15) into The calculation formula of is, （20） When the anti-sliding force of the strip is less than the sum of its own weight and the mass of the overlying soil, the soil begins to become unstable, and the anti-sliding force continues to decrease until the slope failure surface is formed. The slope safety factor can be derived as: Substituting equation (16) and equation (20) into equation (17), we have: （21）。 7. The slope safety and stability analysis method based on integrated sensors and intelligent algorithms according to claim 6 is characterized in that: the step S3 further comprises the following content: the self-training model continuously uses pseudo-labels to iteratively optimize the model, wherein the update process is expressed as: Use supervised data to optimize initial model parameters: （27） in, is the loss function for supervised learning, is the initial function of the model, To label the dataset; Generate pseudo labels for unlabeled data using the current model: （28） in, Iterate for the model t The pseudo-label of For unlabeled samples, is the model parameter of the current iteration, representing the model state, is the predicted probability distribution of the model for unlabeled samples; Update model parameters using labeled data and pseudo-labeled data: （29） in, is the loss function for supervised learning, is the loss function for unsupervised learning, is the weight of the unsupervised loss, controlling its contribution to the total loss; is the model parameter of the current iteration, representing the model state, that is, For model parameters in iteration Update results when For unlabeled samples, is the probability distribution predicted by the model, Iterate for the model t The pseudo-label of To provide a reliable training basis for the labeled dataset for supervised learning. It is an unlabeled dataset.

8. A slope safety and stability analysis device based on integrated sensors and intelligent algorithms, which is suitable for monitoring soil slope state parameters. The device is based on the slope safety and stability analysis method based on integrated sensors and intelligent algorithms as described in claim 1, and is characterized by: It includes multiple measurement modules, which are evenly arranged along the slope surface. According to the on-site survey results, the soil state parameters of the shallow landslide-prone area are monitored by the measurement modules within a certain depth range from the slope surface, and transmitted to the signal base station via wireless signals, and then uploaded to the cloud for real-time data processing and analysis; The measurement module includes an inclination measurement module, a moisture content measurement module, a matrix suction measurement module, and an earth pressure measurement module, wherein: Inclination measurement module: used to monitor the inclination changes of the soil at each measuring point on the slope; the main components of the measurement module include a fixed steel frame, a three-axis accelerometer, a signal transmission system and a power supply system. The working principle is: when the soil slides downward, the soil drives the integrated device to move together, and the device as a whole tilts at a certain angle. The spatial displacement and angle change of the soil relative to the initial position are measured by the inclinometer; Moisture content measurement module: used to monitor the change of soil moisture content within a certain range of each measuring point on the slope. The main components of this module include a signal transmitter, a signal receiver and an oscillator chip. The modules are evenly arranged along the vertical direction of the device according to user needs, and the position is flexible and adjustable. The working principle is: the signal transmitter first sends a sweep test signal of a specific frequency band to the surrounding soil; the upper and lower ends of the signal transmitter are composed of a pair of circular metal rings to form a ring capacitor structure, the surrounding soil acts as a dielectric, and the sensor capacitance is proportional to the dielectric constant of the surrounding soil medium; when the moisture in the soil increases, its dielectric constant increases accordingly, and the measured capacitance value also increases, resulting in a change in the measurement frequency, thereby measuring the moisture content of the soil; The matrix suction measurement module is used to monitor the changes in soil matrix suction within a certain range of each measuring point on the slope; the main components of this module include a sealed plastic tube, a porous ceramic head, a negative pressure gauge and a wireless signal transmitter. The working process is as follows: after the porous ceramic head contacts the surrounding soil, the water in the soil and the plastic tube is exchanged according to the water potential difference; when the soil moisture content increases, the water enters the plastic tube through the ceramic head, and when the soil moisture content decreases, the water enters and exits the plastic tube through the ceramic head, and a vacuum negative pressure is generated in the tube; based on the monitoring data of the negative pressure gauge and the calibration table, the matrix suction of the soil can be inversely calculated; The soil pressure measurement module is used to monitor the size and distribution changes of the soil pressure of the overlying soil at each measuring point on the slope. The main components of the measurement module include a flexible pressure sensor and a signal transmitter. The working principle of the module is: the flexible pressure sensor is attached to the outer surface of the integrated device and is in direct contact with the overlying soil. When the deadweight of the overlying soil changes, the overlying soil exerts soil pressure on the sensor element, and the mixed response of piezoresistance and piezocaphicity in the sensor element changes. The soil pressure value is obtained by measuring the change of the signal.

9. A slope safety and stability analysis system based on integrated sensors and intelligent algorithms, characterized by: The system includes a slope safety and stability analysis method based on integrated sensors and intelligent algorithms as described in any one of claims 1-7.

10. A slope safety and stability analysis system based on integrated sensors and intelligent algorithms, characterized by: The system includes a slope safety and stability analysis device based on integrated sensors and intelligent algorithms as described in claim 8.

Images