A frozen soil block stone roadbed disease monitoring and early warning system, method and storage medium

Abstract

This invention relates to the field of permafrost roadbed monitoring technology, specifically to a monitoring and early warning system, method, and storage medium for permafrost block roadbed defects. The system includes a multi-source heterogeneous data acquisition module; a data transmission and preprocessing module, which normalizes the multi-source heterogeneous data; a data calculation and analysis module, which uses dynamic weight allocation to perform weighted summation on the normalized standard dataset to calculate the block roadbed health index (BSHI); and a system early warning and application module, which establishes BSHI health index threshold grading rules to achieve accurate risk classification and early warning output for defects. Compared to traditional single-data-source monitoring methods, the system provided by this invention can more comprehensively and accurately capture defect characteristics, reduce misjudgments and missed judgments, and provide a scientific basis for early warning and precise maintenance of roads in permafrost regions.

Description

Technical Field

[0001] This invention relates to the field of frozen soil subgrade monitoring technology, specifically to a frozen soil block roadbed disease monitoring and early warning system, method and storage medium. Background Technology

[0002] Permafrost is widely distributed along the Qinghai-Tibet Railway corridor. This high-temperature, high-ice-content permafrost is extremely sensitive to climate and environmental changes, with over 60% of the corridor area being high-risk zones for thaw settlement. To control thaw settlement of permafrost subgrades caused by climate warming and engineering disturbances, riprap subgrades are widely used in permafrost subgrade engineering. This special structure primarily utilizes the large-diameter pores of the riprap to generate a "thermal semiconductor" effect, reducing external heat intrusion during the warm season and continuously exchanging heat during the cold season, thus maintaining the stability of the permafrost beneath the subgrade and protecting the permafrost subgrade.

[0003] With the large-scale application of riprap subgrades, the riprap base layer is subjected to adverse factors such as vehicle vibration, severe freeze-thaw cycles, high-intensity radiation, and wind, sand, and snow accumulation. Problems such as riprap disintegration and cracking, weathering and pulverization of the stones, and blockage by foreign objects occur within the subgrade. This directly obstructs the effective ventilation and convection channels of the riprap layer, easily weakening its heat dissipation and regulation efficiency. Consequently, it leads to subgrade frost heave and thaw settlement deformation, riprap base layer displacement and collapse, and other defects, severely damaging the overall stability of the subgrade structure. Because riprap subgrade defects are characterized by their long-term, hidden, and abrupt nature, their development is often difficult to detect in a timely manner through conventional detection methods. Once they appear, they can rapidly deteriorate, posing a serious threat to road structural safety. Traditional monitoring methods mostly rely on manual inspections or single displacement and settlement monitoring, which have limited monitoring dimensions and cannot effectively monitor the dynamics of riprap layer pores, ventilation efficiency, and changes in the active layer of the underlying permafrost. Therefore, it is difficult to systematically, comprehensively, and in real-time monitor the development characteristics and trends of hidden subgrade defects. Therefore, the accurate identification, quantitative grading assessment, and risk warning of hidden defects in riprap roadbeds have become an urgent need. Summary of the Invention

[0004] To address the problem that traditional monitoring methods are insufficient for systematically and comprehensively monitoring the development characteristics and trends of hidden roadbed defects in real time, this invention proposes a method for monitoring and early warning of defects in frozen soil boulders roadbeds. By integrating multi-physical field coupled data such as roadbed temperature and humidity, vertical displacement deformation, stress and strain, pore water pressure, and porosity, dynamic monitoring and risk warning of road defects can be achieved.

[0005] In a first aspect, the present invention provides a monitoring and early warning system for permafrost block roadbed defects, characterized in that it comprises:

[0006] A multi-source heterogeneous data acquisition module, wherein the data acquisition module uses sensors to collect basic information about the interior of the rubble roadbed, the sensors including a rubble roadbed temperature sensor, a rubble roadbed settlement and deformation sensor, a rubble roadbed porosity sensor, a rubble roadbed inter-base pressure sensor, and a rubble roadbed pore water pressure sensor. The data transmission and preprocessing module performs normalization processing on multi-source heterogeneous data and uses the range normalization method to uniformly map each monitoring index to the [0,1] interval. The range normalization algorithm includes positive index normalization and negative index normalization. The positive index normalization is applied to the normalization process of the porosity of the boulders base course, and the negative index normalization is applied to the normalization process of roadbed temperature, deformation, interlayer pressure of boulders, and pore water pressure. The data calculation and analysis module uses dynamic weight allocation to perform weighted summation on the normalized standard dataset to calculate the slab roadbed health index (BSHI). The dynamic weight allocation is calculated using Principal Component Analysis (PCA) and a Long Short-Term Memory (LSTM) network, and its calculation formula is as follows: In the formula, Let be the dynamically adjusted weight of the i-th variable at time t. The baseline weights are calculated using the principal component analysis (PCA) method. The weights of the i-th variable output by the LSTM model are adjusted by normalizing the dynamically adjusted weights; that is, each weight is divided by the sum of all dynamically adjusted weights to obtain the final weights. The system early warning and application module establishes BSHI health index threshold grading rules to achieve accurate disease risk classification and early warning output.

[0007] Preferably, in the multi-source heterogeneous data acquisition module, the boulders subgrade temperature sensor is deployed along the depth direction inside the subgrade to monitor the longitudinal temperature gradient; the boulders subgrade settlement deformation sensor is deployed along the interface between the bottom of the boulders layer and the underlying layer to monitor the displacement deformation of the boulders layer; the boulders subgrade interlayer pressure sensor is deployed along the middle of the boulders layer to monitor the stress-strain state of the boulders layer; the boulders subgrade porosity sensor is deployed in the middle of the boulders layer to analyze and calculate the porosity of the boulders layer using the air velocity between the boulders pores; and the boulders subgrade pore water pressure sensor is deployed along the seasonally active frozen soil layer of the subgrade base to monitor the change in the upper limit of the frozen soil layer.

[0008] More preferably, the porosity sensor for the boulders roadbed includes an injection pipe and a monitoring device. One end of the injection pipe is connected to an air source generator, and the other end of the injection pipe is inserted into the boulders layer of the roadbed to be tested. The end of the injection pipe inserted into the roadbed to be tested is sealed. An air guide hole is formed on the side wall of the injection pipe, and the air guide hole is oriented towards the monitoring device. The monitoring device is set in the boulders layer of the roadbed to be tested. The monitoring device is used to monitor the wind speed and wind pressure injected into the roadbed to be tested, and to deduce the porosity of the boulders roadbed based on the wind speed and wind pressure.

[0009] Preferably, in the data transmission and preprocessing module, monitoring data is transmitted using one or more communication devices such as IoT, 5G / 6G, and BeiDou, and the monitoring data is cleaned in a process-oriented manner using Kalman filtering to remove noise and outlier interference.

[0010] Preferably, in the data transmission and preprocessing module, the positive index normalization uses the maximum and minimum value normalization method, that is, subtracting the minimum value from the variable value and dividing by the difference between the maximum and minimum values; and the negative index normalization uses the maximum and minimum value normalization method of the reverse index, that is, subtracting the minimum value from the variable value and dividing by the difference between the maximum and minimum values, and then subtracting the value by one.

[0011] Preferably, in the data calculation and analysis module, the principal component analysis (PCA) includes the following calculation steps: calculating the covariance matrix among each monitoring indicator and solving for its eigenvalues ​​and corresponding eigenvectors; selecting a predefined number of principal components with a cumulative contribution rate greater than a preset threshold as the core feature subset after PCA dimensionality reduction by calculating the variance contribution rate of each principal component, for subsequent calculation of the benchmark weights; calculating the comprehensive importance score of each indicator based on the variance contribution rate of each monitoring indicator in the effective principal components, normalizing the comprehensive importance score, and calculating the benchmark weights.

[0012] Preferably, in the data calculation and analysis module, before the Long Short-Term Memory Network (LSTM) is trained, a dataset construction process is also included. The dataset construction process includes the following steps: acquiring historical monitoring data covering at least two complete freeze-thaw cycles with multiple working conditions including the cold season, warm season, and freeze-thaw transition period; using the baseline weights calculated by the principal component analysis (PCA) as initial values, and combining them with the actual disease situation that occurred within 30 days after each time point, generating basic training labels; and dividing the labeled dataset into a training set, a validation set, and a test set in an 8:1:1 ratio.

[0013] Preferably, the BSHI index ranges from [0,1]. Furthermore, in the system's early warning and application module, a BSHI index in the range of [0,0.3) indicates a sudden drop in all monitored indicators, suggesting that the roadbed structure is about to undergo shear failure and requires immediate emergency reinforcement measures. A BSHI index in the range of [0.3,0.6) indicates a high risk of blockage failure in the riprap layer, suggesting the roadbed faces instability and requiring timely local structural maintenance to prevent the heat accumulation effect from exacerbating permafrost degradation. A BSHI index in the range of [0.6,0.8) indicates a continuously increasing risk of roadbed thaw settlement deformation during the warm season, which can easily lead to frost heave deformation and roadbed structural damage during the cold season, requiring strengthened periodic monitoring and preventative maintenance to inhibit the cumulative damage to the roadbed structure caused by freeze-thaw cycles. A BSHI index in the range of [0.8,1] indicates that the roadbed structure is in good working order.

[0014] In a second aspect, the present invention provides a method for monitoring and early warning of diseases in frozen soil boulders roadbed, characterized by comprising the following steps: S1: Multi-source heterogeneous data acquisition, using sensors to collect basic information of the rubble roadbed, including rubble roadbed temperature sensor, rubble roadbed settlement and deformation sensor, rubble roadbed porosity sensor, rubble roadbed inter-base pressure sensor, and rubble roadbed pore water pressure sensor. S2: Data transmission and preprocessing. Normalization processing is performed on multi-source heterogeneous data. The range normalization method is used to uniformly map each monitoring index to the [0,1] interval. The range normalization algorithm includes positive index normalization and negative index normalization. The positive index normalization is applied to the normalization process of the porosity of the boulders roadbed, and the negative index normalization is applied to the normalization process of roadbed temperature, deformation, interlayer pressure of boulders, and pore water pressure. S3: Data Calculation and Analysis. A weighted summation is performed on the normalized standard dataset using dynamic weight allocation to calculate the BSHI (Brick Roadbed Health Index). The dynamic weight allocation is calculated using Principal Component Analysis (PCA) and a Long Short-Term Memory (LSTM) network. The calculation formula is as follows: In the formula, Let be the final dynamic weight of the i-th variable at time t. The baseline weights are calculated using the principal component analysis (PCA) method. The weights of the i-th variable output by the LSTM model are adjusted by normalizing the dynamically adjusted weights; that is, each weight is divided by the sum of all dynamically adjusted weights to obtain the final weights. S4: System early warning and application, establish BSHI health index threshold classification rules, and realize accurate classification of disease risks and early warning output.

[0015] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described in the second aspect.

[0016] This invention proposes a method for monitoring and early warning of roadbed defects in frozen soil with boulders. Through the fusion analysis of multi-source heterogeneous data, it significantly improves the accuracy and efficiency of identifying hidden road defects in cold regions. Compared to traditional single-data source monitoring methods, it can more comprehensively and accurately capture defect characteristics, reducing misjudgments and omissions, and providing a scientific basis for early warning and precise maintenance of roads in frozen soil regions. Compared to existing technologies, this invention has the following innovative points: 1. A system architecture of "multi-source sensing - data processing - multi-modal fusion - health index calculation - disease classification" was constructed, realizing long-term monitoring and dynamic intelligent assessment of hidden diseases of boulder roadbed; 2. Establish a method for calculating the health index of riprap subgrade. The method adopts the normalization calculation method of positive and negative exponential range, PCA principal component analysis combined with LSTM deep learning time series fitting algorithm, adaptively optimizes the weight of each monitoring index, eliminates the influence of dimensional differences, and constructs a dimensionless riprap subgrade stability technical state index to realize the quantitative and continuous evaluation of the overall health status of the subgrade, replacing the traditional discrete threshold judgment mode. 3. Construct a multi-parameter feature coupling discrimination model, based on the technical status index, to achieve graded evaluation of defects and autonomously identify four types of defect risks: blockage of the boulders base course, decline in convective heat transfer efficiency, roadbed freeze-thaw deformation, and boulders layer slippage and collapse. Simultaneously output first-level (emergency response), second-level (alarm), and third-level (early warning) graded results, so that the early warning results can be interpreted and can be implemented to guide maintenance. Attached Figure Description

[0017] Figure 1 This is a block diagram of the frozen soil block roadbed disease monitoring and early warning system of the present invention.

[0018] Figure 2 This is a diagram showing the monitoring cross-sectional layout of the multi-source sensor described in this invention.

[0019] Figure 3 This is a flowchart of the principal component analysis (PCA) method described in this invention.

[0020] Figure 4 This is a flowchart of the method for monitoring and early warning of diseases in frozen soil boulders roadbed according to the present invention.

[0021] Figure 5 This is a schematic block diagram illustrating a computer system for executing a computer program contained in a computer-readable medium according to embodiments of the present invention.

[0022] Marked in the image: 100 is the monitoring and early warning system for frozen soil and rocky roadbed defects; 101 is the multi-source heterogeneous data acquisition module; 102 is the data transmission and preprocessing module; 103 is the data calculation and analysis module; and 104 is the system early warning and application module. 200 is the monitoring section of the multi-source sensor; 201 is the temperature sensor for the rubble roadbed; 202 is the pressure sensor between the rubble roadbed layers; 203 is the wind speed sensor serving as a porosity sensor for the rubble roadbed; 204 is the pore water pressure sensor for the rubble roadbed; 205 is the settlement and deformation sensor for the rubble roadbed; 206 is the data transmission line; 207 is the data acquisition box; 208 is the rubble layer of the roadbed; and 209 is the melting plate at the bottom of the roadbed. 300 is the principal component analysis (PCA) method; S301~S305 are the method steps. 400 describes the monitoring and early warning method for frozen soil and rocky roadbed defects, and S401~S404 describe the method steps. 500 is a computer system for executing a computer program contained in a computer-readable medium according to an embodiment of the present invention, 510 is a processor, and 520 is a memory. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0024] Unless otherwise specified, the terms "upper," "lower," "left," "right," "center," "inner," and "outer," etc., used in the description of specific embodiments of the present invention to indicate orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings, or the orientation or positional relationship in which the product / equipment / device is usually placed during use. These terms are merely for the purpose of facilitating the description of the present invention or simplifying the description in specific embodiments, and for enabling those skilled in the art to quickly understand the solution, and do not indicate or imply that a particular device / component / element must have a specific orientation, or be constructed and operated in a specific positional relationship. Therefore, they should not be construed as limitations on the present invention.

[0025] Furthermore, the use of terms such as "first," "second," and "third" in terminology is merely for distinguishing descriptions of identical or similar components and should not be interpreted as emphasizing or implying the relative importance of a particular component.

[0026] Furthermore, in the description of the embodiments of the present invention, "several", "more than", and "a number of" represent at least two. The number can be any number, such as two, three, four, five, six, seven, eight, or nine, and can even exceed nine.

[0027] Next, combine Figures 1 to 3 This invention provides a detailed description of the frozen soil boulders roadbed disease monitoring and early warning system proposed in the embodiments of the present invention.

[0028] Figure 1 The diagram shows the modules of the frozen soil block roadbed disease monitoring and early warning system 100. The disclosed frozen soil block roadbed disease monitoring and early warning system 100 includes a multi-source heterogeneous data acquisition module 101, a data transmission and preprocessing module 102, a data calculation and analysis module 103, and a system early warning and application module 104. Figure 2 for Figure 1 The layout diagram of the multi-source sensor monitoring section 200 used in the frozen soil and rock roadbed disease monitoring and early warning system 100. Figure 3 for Figure 1 The flowchart of the principal component analysis (PCA) method 300 used in the multi-source heterogeneous data frozen soil block roadbed disease monitoring and early warning system 100.

[0029] exist Figure 1 In the process, the multi-source heterogeneous data acquisition module 101 uses sensors to collect basic information of the rubble roadbed. The sensors include a rubble roadbed temperature sensor, a rubble roadbed settlement and deformation sensor, a rubble roadbed porosity sensor, a rubble roadbed inter-base pressure sensor, and a rubble roadbed pore water pressure sensor.

[0030] Among them, the temperature sensor of the rubble roadbed is deployed along the depth direction inside the roadbed to monitor the longitudinal temperature gradient; the settlement and deformation sensor of the rubble roadbed is deployed along the interface between the bottom of the rubble layer and the underlying layer to monitor the displacement and deformation of the rubble layer; the pressure sensor between the rubble roadbed layers is deployed along the middle of the rubble layer to monitor the stress and strain state of the rubble layer; the porosity sensor of the rubble roadbed is deployed in the middle of the rubble layer to analyze and calculate the porosity of the rubble layer using the air velocity between the pores of the rubble; and the pore water pressure sensor of the rubble roadbed is deployed along the seasonal active layer of frozen soil at the roadbed base to monitor the change of the upper limit of frozen soil.

[0031] More specifically, the porosity sensor for the rubble roadbed mentioned above can be, for example, a wind speed sensor, which includes an air injection pipe and a monitoring device. One end of the air injection pipe is connected to an air source generator, and the other end is inserted into the rubble layer of the roadbed to be tested. The end of the air injection pipe inserted into the roadbed is sealed. An air guide hole is formed on the side wall of the air injection pipe, with the guide hole facing the monitoring device. The monitoring device is installed within the rubble layer of the roadbed to be tested and is used to monitor the wind speed and wind pressure injected into the roadbed, and to deduce the porosity of the rubble roadbed based on the wind speed and wind pressure.

[0032] Figure 2The diagram shows the layout of an example monitoring section 200 of the multi-source sensors used in the multi-source heterogeneous data acquisition module 101. Specifically, 201 is a temperature sensor for the rubble roadbed, 202 is a pressure sensor between the rubble roadbed layers, 203 is a wind speed sensor serving as a porosity sensor for the rubble roadbed, 204 is a pore water pressure sensor for the rubble roadbed, 205 is a settlement and deformation sensor for the rubble roadbed, 206 is a data transmission line, 207 is a data acquisition box, 208 is the rubble layer of the roadbed, and 209 is the melting plate at the bottom of the roadbed.

[0033] for Figure 2 As shown in the diagram, during implementation, a complete monitoring section is set up every 50-100m along the transverse direction of the roadbed, and a temperature measurement point is set up at a depth of 0.5-1m at each monitoring section; settlement deformation sensors are set up synchronously at the temperature monitoring sections.

[0034] Furthermore, the temperature sensor has an accuracy of ≤0.1℃ and a sampling frequency of ≥1 time / h; the settlement deformation sensor has an accuracy of ≤0.1mm and a sampling frequency of ≥1 time / day; the interlayer pressure sensor and pore water pressure sensor have an accuracy of ≤1kPa and a sampling frequency of ≥1 time / h; and the porosity sensor has an accuracy of ≤0.1m / s and a sampling frequency of ≥1 time / h. Using these parameter settings ensures the accuracy and timeliness of the monitoring data.

[0035] exist Figure 1 In this process, the data transmission and preprocessing module 102 performs normalization processing on multi-source heterogeneous data. Optionally, the data transmission and preprocessing module 102 uses one or more communication devices, such as the Internet of Things, 5G / 6G, and BeiDou, to transmit monitoring data, and uses Kalman filtering to perform process-oriented cleaning of the monitoring data to remove noise and outlier interference.

[0036] Furthermore, the data transmission and preprocessing module 102 uses the range normalization method to uniformly map each monitoring indicator to the [0,1] interval. The range normalization algorithm includes positive indicator normalization and negative indicator normalization.

[0037] For positive index normalization, it is applied to the porosity of rubble roadbeds. Since higher porosity results in a more complete ventilation structure and better heat dissipation efficiency, positive index normalization is suitable. The positive index normalization process for porosity uses a maximum-minimum normalization method, which involves subtracting the minimum value from the variable value and dividing by the difference between the maximum and minimum values. Let the porosity of the rubble layer be... K， The normalization calculation process is shown in the following formula:

[0038] In the formula, Let be the normalized standard monitoring value at time t, with a value range of [0,1]. This corresponds to the original real-time monitoring data. and These are the maximum and minimum values ​​of each monitoring indicator within a 7-day sliding time window.

[0039] For negative index normalization, it is applied to the normalization process for subgrade temperature, settlement deformation, interlayer pressure of boulders, and pore water pressure. Since higher temperatures, settlement deformation, interlayer pressure of boulders, and pore water pressure all indicate higher risks of freeze-thaw damage and thaw settlement, and worse structural stability, negative index normalization is suitable. For example, negative index normalization can employ the maximum-minimum normalization method, i.e., subtracting the minimum value from the variable value, dividing by the difference between the maximum and minimum values, and then subtracting the remaining value from the result. The formula for calculating the negative index normalization, which represents the subgrade temperature, settlement deformation, interlayer pressure of boulders, and pore water pressure, is as follows:

[0040] In the formula, Let be the normalized standard monitoring value at time t, with a value range of [0,1]. This corresponds to the original real-time monitoring data; among which Any one of the following indicators: subgrade temperature T, deformation D, interlayer pressure of boulders P1, and pore water pressure P2. and These are the maximum and minimum values ​​of each monitoring indicator within a 7-day sliding time window.

[0041] exist Figure 1 In the middle, the data calculation and analysis module 103 uses dynamic weight allocation to perform weighted summation on the normalized standard dataset to calculate the Block-Stone Subgrade Health Index (BSHI).

[0042] BSHI can achieve accurate fusion and quantitative evaluation of multiphysics data. Its specific calculation formula is as follows:

[0043] In the formula, The seasonal adaptive dynamic weights are, in order, temperature, deformation, interlayer pressure of boulders, pore water pressure, and porosity. The value of BSHI(t) is strictly constrained to [0,1]. The closer the value is to 1, the better the overall stability of the subgrade and the lower the risk of hidden defects; the closer the value is to 0, the more severe the multi-field performance degradation of the subgrade is and the extremely high risk of instability.

[0044] In the calculation process of BSHI, the dynamic weights used are first calculated using principal component analysis (PCA) to calculate the baseline weights, and then calculated using long short-term memory (LSTM) network to obtain the time-series dynamic weight correction coefficients. The final dynamic weights are obtained by combining the two.

[0045] PCA is used to calculate the benchmark weights, and the process is as follows: Figure 3 As shown.

[0046] First, in step S301, the multi-source data that has been preprocessed and normalized by the data transmission and preprocessing module 102 is input.

[0047] Next, in step S302, a covariance matrix is ​​constructed, and eigenvalues ​​are generated. The data matrix A has a dimension of m×n, where m is the number of detection indicators (e.g., m=5, representing porosity K, temperature T, deformation D, interlayer pressure P1, and pore water pressure P2, respectively), and n is the number of data points within the window (e.g., for a seven-day window, hourly sampling n=168, and daily sampling data is completed to hourly levels through linear interpolation). The covariance matrix C between each detection indicator is calculated, and its eigenvalues ​​λ and corresponding eigenvectors are solved. The formula for calculating the covariance matrix C is as follows:

[0048] In the formula, The normalized data matrix, for The transpose of the matrix, elements within the covariance matrix This represents the covariance between the i-th and j-th indicators, reflecting the degree of linear correlation among the monitoring indicators.

[0049] The covariance matrix C is decomposed into eigenvalues, and the QR algorithm is used to solve for the eigenvalues ​​λ and their corresponding eigenvectors v.

[0050] In the formula, For the i-th eigenvalue, Let be the eigenvector corresponding to the i-th eigenvalue, with dimension m×1, satisfying || ||=1.

[0051] After calculating the covariance matrix and eigenvalues, in step S303, principal components are selected, and variance contribution rate analysis is performed. In this step, by calculating the variance contribution rate of each principal component, a predefined number of principal components with a cumulative contribution rate greater than a preset threshold (e.g., the top five principal components with a cumulative contribution rate greater than 85%) are selected as the core feature subset after PCA dimensionality reduction, used for subsequent calculation of baseline weights. Specifically, the calculation formula is as follows:

[0052] In the formula, The variance contribution rate of the i-th principal component represents the proportion of the original data variation that can be explained by this principal component to the total variation. Let i be the i-th eigenvalue; This represents the cumulative variance contribution rate of the first p principal components.

[0053] Further, in step S304, based on the variance contribution rate of each monitoring indicator in the effective principal components, the comprehensive importance score of each indicator is calculated, and the comprehensive importance score is normalized to calculate the benchmark weight. The comprehensive importance score of each indicator is calculated using the following formula:

[0054] In the formula: The overall importance score for the i-th monitoring indicator; Let | be the variance contribution rate of the j-th principal component. | represents the absolute value of the eigenvector corresponding to the i-th eigenvalue.

[0055] The normalization of the overall importance score is calculated using the following formula:

[0056] In the formula: is the benchmark weight for the i-th monitoring indicator.

[0057] In step S305, the benchmark weight results obtained after PCA calculation are output.

[0058] After obtaining the baseline weights through PCA, the time-series dynamic weight correction coefficients are further calculated using LSTM. The specific construction and training process of LSTM are described below.

[0059] The LSTM architecture employs a three-layer structure: an input gate, a forget gate, and an output gate. The input gate has 13 input dimensions, including the following indicators: normalized values ​​of five monitoring indicators (rock layer porosity K, subgrade temperature T, deformation D, rock interlayer pressure P1, and pore water pressure P2, 5-dimensional), seasonal condition labels (cold season / warm season, unique thermal encoding, 2-dimensional), the 7-day moving average change rate of each indicator (reflecting the temporal deterioration trend, 5-dimensional), and the statistical value of the number of freeze-thaw cycles in the past 30 days (1-dimensional). The forget gate is configured similarly to the input gate, controlling the degree of retention of historical information and selectively forgetting non-critical factors to ensure the model can capture key temporal features in long-term and short-term dependencies, avoiding errors. The output gate has 5 output dimensions, corresponding to the dynamic weight correction coefficients of the five indicators: temperature, deformation, rock interlayer pressure, pore water pressure, and porosity.

[0060] For the LSTM training process, the dataset is first constructed. The data source for the dataset is historical monitoring data from at least two complete freeze-thaw cycles (no less than two years), covering various operating conditions such as the cold season, warm season, and freeze-thaw transition period. For data labeling, PCA baseline weights are used as initial values, combined with actual disease occurrences within 30 days after each time point, as basic training labels to improve detection accuracy. After labeling the dataset, it is divided into training, validation, and test sets in an 8:1:1 ratio.

[0061] After obtaining the dataset, the model training step is performed. In this step, for example, the Adam optimizer is used, and the total number of iterations is set to 100 to prevent overfitting of the training data.

[0062] Using a trained LSTM model, the temporal dynamic weight correction coefficients are calculated and combined with the baseline weights calculated by PCA through a dynamic correction process to obtain the final dynamic weights. The combination method is shown in the following formula:

[0063] In the formula, Let be the final dynamic weight of the i-th indicator at time t; The baseline weights calculated for the PCA method; This represents the weight correction coefficient for the i-th index output by the LSTM model. Further, the dynamically corrected weights are normalized by dividing each weight by the sum of all dynamically corrected weights to obtain the final weights.

[0064] The dynamic weighting system is set to update once a day, automatically executing at midnight every day. The updated dynamic weights will be used for subsequent health index calculations.

[0065] exist Figure 1 In the system, the early warning and application module 104 establishes the BSHI health index threshold classification rules to achieve accurate disease risk classification and early warning output.

[0066] The BSHI health index threshold grading rules can be divided according to the following criteria: When the BSHI index falls within the range of [0, 0.3), it indicates a sudden decrease in all monitoring indicators, a significant acceleration in the deformation rate, and an abnormal increase in temperature within the boulders layer, suggesting that the roadbed structure is about to undergo shear failure; the integrity of the roadbed structure faces a severe challenge, and emergency reinforcement measures need to be initiated immediately.

[0067] When the BSHI index falls within the range of [0.3, 0.6), it indicates that as the porosity of the boulders layer continues to decrease and the temperature within the boulders layer gradually increases, the risk of blockage failure of the boulders layer is high, the roadbed faces the risk of instability, and local structural maintenance needs to be carried out in a timely manner to prevent the heat accumulation effect from exacerbating permafrost degradation.

[0068] The BSHI index falls within the range of [0.6, 0.8), indicating that during the warm season, the rubble layer absorbs heat and warms up, leading to an increase in the upper limit temperature of the frozen soil, increased pore water pressure, and a deeper active temperature layer. This results in a continuous increase in the risk of roadbed thaw settlement and deformation, and an expansion of the thaw plate. During the cold season, the frozen soil refreezes, and the pore water pressure changes periodically, which can easily cause frost heave and deformation of the roadbed and damage to the roadbed structure. It is necessary to strengthen periodic monitoring and preventive maintenance to inhibit the cumulative damage of freeze-thaw cycles to the roadbed structure.

[0069] A BSHI index within the range of [0.8, 1] indicates that the roadbed structure is in good working order and continuous monitoring is sufficient.

[0070] Based on the above classification criteria, corresponding warning levels can be output according to the BSHI index value: a BSHI index less than 0.3 is a Level 1 (emergency response) warning, between 0.3 and 0.6 is a Level 2 (alarm) warning, and between 0.6 and 0.8 is a Level 3 (warning) warning.

[0071] The permafrost roadbed disease monitoring and early warning system proposed in this invention significantly improves the accuracy and efficiency of identifying hidden road diseases in cold regions through the fusion analysis of multi-source heterogeneous data. Compared with traditional single-data source monitoring systems, it can capture disease characteristics more comprehensively and accurately, reduce misjudgments and omissions, and provide a scientific basis for early warning and precise maintenance of roads in permafrost regions.

[0072] The permafrost boulders subgrade disease monitoring and early warning system proposed in this invention constructs a system architecture of "multi-source sensing—data processing—multi-modal fusion—health status index calculation—disease classification and risk early warning," realizing long-term monitoring and dynamic intelligent assessment of hidden diseases in boulders subgrades. Simultaneously, based on the system proposed in this invention, a method for calculating the health index of boulders subgrades is established. This method employs a normalized calculation method for positive and negative exponential ranges, combined with PCA principal component analysis and LSTM deep learning time series fitting algorithm, adaptively optimizing the weights of each monitoring indicator to eliminate the influence of dimensional differences. A dimensionless boulders subgrade stability technical status index is constructed, achieving a quantitative and continuous evaluation of the overall health status of the subgrade, replacing the traditional discrete threshold judgment mode. Furthermore, the system proposed in this invention also constructs a multi-parameter feature coupling discrimination model. Based on the technical status index, it realizes graded evaluation of diseases and autonomously identifies four types of disease risks: blockage of the boulders base course, decline in convective heat transfer efficiency, roadbed freeze-thaw deformation, and boulders layer slippage and collapse. It simultaneously outputs first-level (emergency response), second-level (alarm), and third-level (early warning) graded results, achieving the effect of interpretable and practical guidance for maintenance based on early warning results.

[0073] like Figure 4 As shown, this embodiment also provides a method 400 for monitoring and early warning of diseases in frozen soil boulders roadbed, including steps S401 to S404. This method 400 can be... Figure 1 The system 100 shown is executed, but this is merely an example. Those skilled in the art will understand that the above method steps can also be executed by other devices or system architectures with corresponding processing capabilities.

[0074] exist Figure 4 In step S401, multi-source heterogeneous data acquisition is performed. In this step, sensors are used to collect information about the rubble roadbed foundation. These sensors include a rubble roadbed temperature sensor, a rubble roadbed settlement and deformation sensor, a rubble roadbed porosity sensor, a rubble roadbed inter-subgrade pressure sensor, and a rubble roadbed pore water pressure sensor. Specifically, in the process of... Figure 1 The description already includes various types of sensors, such as porosity sensors for rubble roadbeds. The above description of the sensors also applies to... Figure 4 The process of acquiring multi-source heterogeneous data in S401.

[0075] exist Figure 4 In step S402, data transmission and preprocessing are performed. In this step, multi-source heterogeneous data is normalized by using a range normalization method to map each monitoring index to the [0,1] interval. The range normalization algorithm includes positive index normalization and negative index normalization. Positive index normalization is applied to the normalization process of porosity in the boulders subgrade, while negative index normalization is applied to the normalization process of subgrade temperature, deformation, interlayer pressure of boulders, and pore water pressure. Figure 1 The data transmission and preprocessing processes have already been described in the description. The above description of data transmission and preprocessing processes also applies to... Figure 4 The steps in S402.

[0076] exist Figure 4 In step S403, the data calculation and analysis process is described. In this step, the data calculation and analysis module uses dynamic weight allocation to perform a weighted summation on the normalized standard dataset to calculate the Rockbed Health Index (BSHI). The dynamic weight allocation first uses Principal Component Analysis (PCA) to calculate the baseline weights, then uses a Long Short-Term Memory (LSTM) network to calculate the time-series dynamic weight correction coefficients. The final dynamic weight allocation is obtained by combining these two methods. [Further details regarding the process are needed for a complete translation.] Figure 1 The description already details the specific calculation process for dynamic weights. This same process also applies to... Figure 4 The process of data calculation and analysis in S403.

[0077] exist Figure 4 In this section, S404 outlines the steps for system early warning and application. This step establishes BSHI health index threshold grading rules to achieve accurate disease risk classification and early warning output. [Further details regarding the steps are needed for a complete translation.] Figure 1 The description already outlines the specific rules for categorizing BSHI health index thresholds and the process for issuing warnings based on these categorizations. These rules for categorizing BSHI health index thresholds and the corresponding warning levels also apply to… Figure 4 The S404 process in [the context of the original text].

[0078] The method for monitoring and early warning of roadbed defects in frozen soil with boulders proposed in this invention significantly improves the accuracy and efficiency of identifying hidden road defects in cold regions through the fusion analysis of multi-source heterogeneous data. Compared with traditional single-data source monitoring methods, it can capture defect characteristics more comprehensively and accurately, reduce misjudgments and omissions, and provide a scientific basis for early warning and precise maintenance of roads in frozen soil regions.

[0079] The following examples further illustrate the method for monitoring and early warning of diseases in frozen soil and rocky roadbeds included in this invention.

[0080] In this embodiment, the application of monitoring and early warning of roadbed defects on the Gongyu Expressway is taken as an example. The Gongyu Expressway is 635.6 kilometers long and traverses 227 kilometers of permafrost areas. High-temperature unstable permafrost is widely distributed along the route. Among them, the section from K429+800 to K429+950 is 150m long and adopts a roadbed structure with boulders (boulder layer thickness 1.2m, particle size 15~30cm).

[0081] In step one of the embodiments, multi-source heterogeneous data is collected. Complete monitoring sections are set up at three locations: K429+800, K429+875, and K429+950. Sensors are deployed at each section as follows: Figure 2 As shown, the system includes a riprap subgrade porosity sensor 203, a subgrade temperature sensor 201, a settlement deformation sensor 205, a riprap subgrade inter-base pressure sensor 202, and a pore water pressure sensor 204. The system uses a multi-channel data acquisition instrument to collect data from multiple sensors, achieving synchronous acquisition and local caching of multi-source data.

[0082] Specifically, for the riprap subgrade temperature sensor 201, one point is placed every 0.5m along the depth, for a total of 10 points, with an accuracy of ±0.05℃, and data is collected once per hour. For the riprap subgrade settlement deformation sensor 205, one sensor is placed at the bottom of the riprap layer, with an accuracy of ±0.05mm, and data is collected once per day. For the riprap subgrade interlayer pressure sensor 202, one sensor is placed in the middle of the riprap layer, with an accuracy of ±0.5kPa, and data is collected once per hour. For the micro-wind speed sensor 203, which serves as the riprap subgrade porosity sensor, one sensor is placed in the middle of the riprap layer to calculate porosity, with an accuracy of ±0.05m / s, and data is collected once per hour. For the riprap subgrade pore water pressure sensor 204, one sensor is placed within the active layer (0.5m depth), with an accuracy of ±0.5kPa, and data is collected once per hour.

[0083] In step two of the embodiment, data transmission and preprocessing are performed.

[0084] After preliminary verification, the collected data is uploaded to the cloud server in real time via a 4G / 5G wireless communication module, realizing remote transmission of wireless data.

[0085] Taking the monitoring data of section K429+875 (the location of maximum settlement) from July 15, 2025 to July 21, 2025 (7 consecutive days) as an example, a sliding time window was constructed, and the statistical characteristic values ​​of each monitoring indicator were extracted as shown in the table below: Table 1: Statistical Characteristic Values ​​of Each Monitoring Indicator

[0086] Among them, the settlement deformation is a cumulative value (based on the opening of the road in 2020), and the porosity is calculated from the actual flow velocity measured by the micro wind speed sensor.

[0087] Subsequently, range normalization data processing (taking the measured data on July 20 as an example) provides a standardized dataset for subsequent data fusion.

[0088] Positive normalized index (porosity K of the boulder layer):

[0089] The calculation method for negative normalized indicators is similar to that of positive indicators. It uses the maximum and minimum value normalization method for negative indicators, which involves subtracting the minimum value from the variable value, dividing by the difference between the maximum and minimum values, and then subtracting the value from one. The daily average values ​​of each indicator after normalization within a 7-day window are as follows: Table 2: Normalized daily average values ​​of each monitoring indicator within the 7-day window

[0090] It should be noted that the values ​​in the table are normalized daily averages, which differ from the values ​​used in the actual subsequent PCA calculations. These are hourly data (n=168). After normalization, all indicators take values ​​in the range [0,1]. The closer a value is to 1, the better the indicator's condition; the closer it is to 0, the worse the condition. In step three of the embodiment, data fusion calculation and analysis are performed. This step can be further divided into three steps: PCA baseline weight calculation, LSTM time-series dynamic analysis method for dynamic weight allocation, and basalt roadbed health index (BSHI) calculation.

[0091] Furthermore, the calculation of PCA benchmark weights can be divided into three steps: covariance matrix and eigenvalue decomposition, principal component selection and variance contribution rate calculation, and benchmark weight allocation calculation.

[0092] In the covariance matrix and eigenvalue decomposition steps, the covariance matrix C (5×5) among the 168-hour data points within the 7-day window is calculated to obtain the covariance matrix C among the 5 monitoring indicators: Table 3: Covariance Matrix among Monitoring Indicators

[0093] As can be seen from the covariance matrix, there is a strong correlation among the indicators: the correlation coefficient between deformation and pore water pressure is the highest (0.835), and porosity is negatively correlated with the other four indicators, which is consistent with the law of multi-physical field coupling of rubble roadbed.

[0094] Eigenvalue decomposition of the covariance matrix yields five eigenvalues ​​(sorted from largest to smallest): λ1=3.872, λ2=0.615, λ3=0.312, λ4=0.146, λ5=0.055 In the principal component selection and variance contribution rate calculation steps, the variance contribution rate α... i The proportion of the eigenvalue of the i-th principal component to the sum of all eigenvalues: by For example, calculation =77.44% Similarly, we can calculate α2 = 12.30%, α3 = 6.24%, α4 = 2.92%, and α5 = 1.10%. Cumulative variance contribution rate β p The sum of the variance contribution rates of the first p principal components: When p=1 = =77.44% Similarly, when p=2 or 3, β2=89.74% and β3=95.98%. In this embodiment, the top two principal components (β2=89.74%) with a cumulative variance contribution rate of over 85% are selected as effective principal components.

[0095] In the benchmark weight allocation calculation step, based on the variance contribution rate of each monitoring indicator in the effective principal components, the comprehensive importance score of each indicator is calculated, and the comprehensive importance score is normalized to calculate the benchmark weight.

[0096] Firstly, the eigenvectors corresponding to the first two principal components whose cumulative variance contribution rate reaches over 85% are: V1=[0.447,0.472,0.421,0.485,-0.403] T V2=[-0.273,-0.125,0.618,-0.086,-0.725] T Based on the variance contribution rates calculated in the first two steps, the comprehensive importance scores of each indicator (porosity K of the boulders layer, subgrade temperature T, deformation D, interlayer pressure P1 of the boulders layer, and pore water pressure P2) can be obtained as follows: S T =77.44%×0.447+12.30%×0.273=0.379 S D =77.44%×0.472+12.30%×0.125=0.381 S P1 =77.44%×0.421+12.30%×0.618=0.402 S P2 =77.44%×0.485+12.30%×0.086=0.386 S K =77.44%×0.403+12.30%×0.725=0.401 Normalizing the overall importance score yields the baseline weights. The normalization formula is as follows:

[0097] Where variable i corresponds to temperature, deformation, interlayer pressure of boulders, pore water pressure, and porosity in turn, and variable m corresponds to the total number of indicators (5 in this embodiment).

[0098] Therefore, the baseline weights obtained after normalization are:

[0099] After completing the PCA baseline weight calculation step, the LSTM time-series dynamic analysis method is used to dynamically allocate weights. Furthermore, the dynamic weight allocation using the LSTM time-series dynamic analysis method can be divided into three steps: input feature construction, model training output, and dynamic adjustment of baseline weights.

[0100] In the input feature construction step, the input data is integrated into a form that conforms to the input requirements of LSTM. Specifically, the input features include the following variables: Normalized monitoring values ​​include the porosity K of the boulders layer, the subgrade temperature T, the deformation D, the interlayer pressure P1 of the boulders, and the pore water pressure P2. Taking the measured data from July 22nd as an example, T... =0.381, D =0.426, P1 =0.452, P2 =0.394, K =0.412 Seasonal operating condition labels use two-dimensional one-hot encoding to distinguish between cold and warm seasons. In the measured data involved in this embodiment, it is set as warm season, and the encoding is [0, 1].

[0101] The input features also include the 7-day moving average rate of change of the measured data, as detailed below: , , , ,

[0102] In addition, the input features also include the number of freeze-thaw cycles in the past 30 days, counted as 2 times.

[0103] After constructing the input features, the model training and output step is performed. In this step, the feature vectors are input into the trained LSTM model to obtain the weight correction coefficients for each indicator: , , , ,

[0104] After obtaining the indicator weight adjustment coefficient, the benchmark weight is dynamically adjusted using the following formula:

[0105] The dynamically adjusted weights are then normalized by dividing each weight by the sum of all dynamically adjusted weights, resulting in the following adjusted weights:

[0106] It can be seen that, compared with the baseline weight, the porosity weight decreases due to the thermal expansion effect of the boulders layer caused by the rise in temperature during the warm season, while the weights of subgrade temperature, deformation, interlayer pressure of boulders, and pore water pressure increase. This reflects the physical law that the melting and deformation and pressure changes of the boulders structure layer become the dominant factors during the warm season.

[0107] After dynamically assigning weights and renormalizing using the LSTM time-series dynamic analysis method, the BSHI (Brick Subgrade Health Index) is calculated. Specifically, the formula used for BSHI calculation is:

[0108] Taking the measured data from July 22 as an example, substituting the above data...

[0109] In step four of the embodiment, risk classification warnings and outputs are performed based on the calculated BSHI.

[0110] According to the BSHI health index threshold classification rules, the calculated BSHI health index of the rubble roadbed is 0.413, which is between 0.3 and 0.6. The roadbed structure is at risk of instability and local structural maintenance needs to be carried out in a timely manner. A level-two alarm should be initiated.

[0111] Correspondingly, the road damage survey along the route revealed continuous wavy settlement in this section, with two consecutive settlements within 150m, and the maximum settlement at a single location was 6cm. On-site excavation revealed that part of the boulders roadbed material was disintegrating stone, which had disintegrated into needle-like and flaky pieces under high-frequency freeze-thaw cycles and high-frequency wet-dry cycles, resulting in a decrease in the porosity of the boulders layer and obstruction of ventilation convection channels.

[0112] Therefore, it can be seen that the calculated results of the on-site monitoring indicators are consistent with the results of the artificial disease investigation and excavation, further verifying the effectiveness of the method.

[0113] According to another aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, which, when executed by a processor, implements a method for monitoring and early warning of permafrost block roadbed defects according to an embodiment of the present invention.

[0114] Figure 5This is a schematic block diagram illustrating a computer system 500 for executing a computer program contained in a computer-readable medium according to an embodiment of the present invention.

[0115] like Figure 5 As shown, a computer system 500 for executing a computer program contained in a computer-readable medium according to embodiments of the present invention may include a processor 510 and a memory 520. The memory 520 stores the computer program.

[0116] Processor 510 is any processing device such as a microprocessor, which operates according to a computer program installed / stored in memory 520. Memory 520 is any volatile or non-volatile storage element, such as a hard disk, solid-state drive, ROM, or RAM. The program executed by processor 510, etc., is stored in memory 520. Figure 5 The computer device 500 shown can be used to implement a method for monitoring and early warning of frozen soil block roadbed defects according to an embodiment of the present invention.

[0117] The hardware computing device or its components described in this invention can be implemented by various suitable hardware means, including but not limited to FPGA, ASIC, SoC, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The devices, apparatuses, methods, and systems involved in this invention are not limited to any specific hardware architecture or configuration. Components in the disclosed devices, apparatuses, and systems may be discrete or integrated, and may be combined in different ways and / or replaced or supplemented by other components. It should be understood that the teachings of this invention can be implemented in various forms, including hardware, software, firmware, dedicated processors, or combinations thereof.

[0118] The block diagrams of the devices, equipment, methods, systems, etc., involved in this invention are merely exemplary and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these circuits, devices, equipment, and systems can be connected, arranged, and configured in any manner that achieves the desired purpose.

[0119] In the foregoing description, the present invention has been described based on embodiments. The various embodiments described above are merely illustrative, and those skilled in the art should understand that the combination of constituent elements and processes of these embodiments can be modified in various ways, and such modifications are also within the scope of the present invention.

Claims

1. A monitoring and early warning system for frozen soil boulders roadbed defects, characterized in that, include: A multi-source heterogeneous data acquisition module, wherein the data acquisition module uses sensors to collect basic information of the rubble roadbed, the sensors including rubble roadbed temperature sensor, rubble roadbed settlement and deformation sensor, rubble roadbed porosity sensor, rubble roadbed inter-base pressure sensor, and rubble roadbed pore water pressure sensor. The data transmission and preprocessing module performs normalization processing on multi-source heterogeneous data and uses the range normalization method to uniformly map each monitoring index to the [0,1] interval. The range normalization algorithm includes positive index normalization and negative index normalization. The positive index normalization is applied to the normalization process of the porosity of the boulders base course, and the negative index normalization is applied to the normalization process of roadbed temperature, deformation, interlayer pressure of boulders, and pore water pressure. The data calculation and analysis module uses dynamic weight allocation to perform weighted summation on the normalized standard dataset to calculate the slab roadbed health index (BSHI). The dynamic weight allocation is calculated using Principal Component Analysis (PCA) and a Long Short-Term Memory (LSTM) network, and its calculation formula is as follows: In the formula, Let be the dynamically adjusted weight of the i-th variable at time t. The baseline weights are calculated using the principal component analysis (PCA) method. The weights of the i-th variable output by the LSTM model are adjusted by normalizing the dynamically adjusted weights; that is, each weight is divided by the sum of all dynamically adjusted weights to obtain the final weights. The system early warning and application module establishes BSHI health index threshold grading rules to achieve accurate disease risk classification and early warning output.

2. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, In the multi-source heterogeneous data acquisition module, the boulders subgrade temperature sensor is deployed along the depth direction inside the subgrade to monitor the longitudinal temperature gradient; the boulders subgrade settlement deformation sensor is deployed along the interface between the bottom of the boulders layer and the underlying layer to monitor the displacement deformation of the boulders layer; the boulders subgrade interlayer pressure sensor is deployed along the middle of the boulders layer to monitor the stress-strain state of the boulders layer; the boulders subgrade porosity sensor is deployed in the middle of the boulders layer to analyze and calculate the porosity of the boulders layer using the air velocity between the boulders pores; and the boulders subgrade pore water pressure sensor is deployed along the seasonally active frozen soil layer of the subgrade base to monitor the change in the upper limit of the frozen soil layer.

3. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 2, characterized in that, The porosity sensor for the boulders roadbed includes an injection pipe and a monitoring device. One end of the injection pipe is connected to an air source generator, and the other end of the injection pipe is inserted into the boulders layer of the roadbed to be tested. The end of the injection pipe inserted into the roadbed to be tested is sealed. An air guide hole is opened on the side wall of the injection pipe, and the opening direction of the air guide hole is towards the monitoring device. The monitoring device is installed in the boulders layer of the roadbed to be tested. The monitoring device is used to monitor the wind speed and wind pressure injected into the roadbed to be tested, and to deduce the porosity of the boulders roadbed based on the wind speed and wind pressure.

4. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, In the data transmission and preprocessing module, monitoring data is transmitted using one or more communication devices, such as the Internet of Things, 5G / 6G, and BeiDou. The Kalman filter method is used to perform process-oriented cleaning of the monitoring data to remove noise and outlier interference.

5. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, In the data transmission and preprocessing module, the positive index normalization uses a maximum-minimum-value normalization method, that is, subtracting the minimum value from the variable value and dividing by the difference between the maximum and minimum values; and The negative index normalization uses the maximum and minimum value normalization method of the inverse index, that is, subtract the minimum value from the variable value, divide by the difference between the maximum and minimum values, and then subtract the value by one.

6. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, In the data calculation and analysis module, the principal component analysis (PCA) includes the following calculation steps: Calculate the covariance matrix among the monitoring indicators, and solve for its eigenvalues ​​and corresponding eigenvectors; By calculating the variance contribution rate of each principal component, a predefined number of principal components with a cumulative contribution rate greater than a preset threshold are selected as the core feature subset after dimensionality reduction by principal component analysis (PCA) and used for subsequent calculation of benchmark weights. Based on the variance contribution rate of each monitoring indicator in the effective principal components, the comprehensive importance score of each indicator is calculated, and the comprehensive importance score is normalized to calculate the benchmark weight.

7. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, In the data computation and analysis module, before the Long Short-Term Memory (LSTM) network is trained, a dataset construction process is also included. The dataset construction process includes the following steps: Acquire historical monitoring data for at least two complete freeze-thaw cycles covering various working conditions including cold season, warm season, and freeze-thaw transition period; use the baseline weights calculated by the principal component analysis (PCA) as initial values, and combine them with the actual damage situation that occurred within 30 days after each time point to generate basic training labels. The labeled dataset was divided into training, validation, and test sets in an 8:1:1 ratio.

8. The monitoring and early warning system for frozen soil boulders roadbed defects according to claim 1, characterized in that, The BSHI index ranges from [0,1], and in the system early warning and application module, A BSHI index within the range of [0, 0.3) indicates a sudden drop in all monitored indicators, suggesting that the roadbed structure is about to undergo shear failure and that emergency reinforcement measures must be initiated immediately. A BSHI index in the range of [0.3, 0.6) indicates a high risk of blockage failure in the riprap layer, a risk of roadbed instability, and the need for timely local structural maintenance to prevent the heat accumulation effect from exacerbating permafrost degradation. The BSHI index falls within the range of [0.6, 0.8), indicating that the risk of roadbed thaw settlement deformation continues to increase during the warm season and is prone to frost heave deformation and roadbed structural damage during the cold season. It is necessary to strengthen periodic monitoring and preventive maintenance to inhibit the cumulative damage of freeze-thaw cycles to the roadbed structure. A BSHI index within the range of [0.8, 1] indicates that the roadbed structure is in good working order.

9. A method for monitoring and early warning of diseases in frozen soil boulders roadbed, characterized in that, Includes the following steps: S1: Multi-source heterogeneous data acquisition, using sensors to collect basic information of the rubble roadbed, including rubble roadbed temperature sensor, rubble roadbed settlement and deformation sensor, rubble roadbed porosity sensor, rubble roadbed inter-base pressure sensor, and rubble roadbed pore water pressure sensor. S2: Data transmission and preprocessing. Normalization processing is performed on multi-source heterogeneous data. The range normalization method is used to uniformly map each monitoring index to the [0,1] interval. The range normalization algorithm includes positive index normalization and negative index normalization. The positive index normalization is applied to the normalization process of the porosity of the boulders roadbed. The negative index normalization is applied to the normalization process of roadbed temperature, deformation, interlayer pressure of boulders and pore water pressure. S3: Data Calculation and Analysis. A weighted summation is performed on the normalized standard dataset using dynamic weight allocation to calculate the BSHI (Brick Roadbed Health Index). The dynamic weight allocation is calculated using Principal Component Analysis (PCA) and a Long Short-Term Memory (LSTM) network, and its calculation formula is as follows: In the formula, Let be the final dynamic weight of the i-th variable at time t. The baseline weights are calculated using the principal component analysis (PCA) method. The weights of the i-th variable output by the LSTM model are adjusted by normalizing the dynamically adjusted weights; that is, each weight is divided by the sum of all dynamically adjusted weights to obtain the final weights. S4: System early warning and application, establish BSHI health index threshold classification rules, and realize accurate classification of disease risks and early warning output.

10. A computer-readable storage medium, characterized in that, It contains a computer program that, when executed by a processor, implements the method as described in claim 9.

Images