A method and system for predicting thermal thaw settlement of a saline permafrost railway
By constructing a multi-field coupling model and physical constraint machine learning, the problem of multi-factor coupling in traditional methods was solved, and accurate prediction and risk assessment of railway thermal thawing settlement in saline permafrost were achieved, improving the accuracy and stability of the prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2026-05-21
- Publication Date
- 2026-07-21
AI Technical Summary
Traditional methods struggle to simultaneously address the multi-factor coupling effects of changes in ground temperature, ice content, salinity, and climate boundaries, making it impossible to construct a spatialized prediction model covering the entire railway line. Furthermore, they lack a unified description reflecting the thermal thawing potential and mechanical response of permafrost, resulting in discrepancies between the predicted thermal thawing settlement results and the actual conditions of railways in saline permafrost.
A multi-field coupled model based on ground temperature, ice content, salinity, and InSAR deformation data is constructed. By combining the equivalent thermal fusion potential energy and thermal fusion settlement sensitivity coefficient with a physical constraint machine learning model, accurate prediction of thermal fusion settlement along railway lines is achieved.
It significantly improves the accuracy and stability of thermal settlement prediction, enabling the assessment of settlement risk across the entire region and multiple time scales, and supporting maintenance decisions and structural safety assurance during railway operation.
Smart Images
Figure CN122432708A_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of engineering construction, and more specifically, this application relates to a method and system for predicting thermal thawing settlement of railways in saline permafrost. Background Technology
[0002] Permafrost, a unique geological medium widely distributed in high-altitude and cold regions, is influenced by temperature, ice content, salinity, and groundwater conditions in its engineering properties. Under the backdrop of climate warming, permafrost generally exhibits a trend of rising upper limits, deepening of the active layer, and increasing temperature, posing varying degrees of risk of uneven settlement due to thaw during the operational cycle of railway subgrades. Furthermore, in the Qinghai-Tibet Plateau and the cold and arid northwest regions, some permafrost also exhibits significant salinization characteristics. The presence of salt not only lowers the freezing point and accelerates the ice-water phase transition but also weakens the permafrost's skeletal structure, making the mechanism of thaw settlement more complex and significantly increasing the difficulty of prediction.
[0003] Traditional methods for predicting thaw settlement of permafrost subgrades typically rely on empirical correction formulas, simplified one-dimensional or two-dimensional heat conduction models, or empirical statistical relationships based on a small number of monitoring points. However, these methods generally have the following limitations: First, they are difficult to simultaneously handle the multi-factor coupling effects of ground temperature, ice content, salinity, and climate boundary changes; second, they are difficult to construct a spatial prediction model covering the entire route; third, the models fail to effectively absorb high-precision deformation monitoring data such as InSAR (interferometric synthetic aperture radar), leading to deviations between prediction results and actual operating conditions; and fourth, they lack a unified description reflecting the relationship between permafrost thaw potential and mechanical degradation, making them difficult to adapt to the complex medium with a special structure, such as saline permafrost.
[0004] Therefore, how to construct a multi-field coupled model that can simultaneously reflect heat conduction, moisture migration, salt diffusion and foundation mechanical response, and integrate it with high-resolution monitoring data to achieve accurate prediction of thermal fusion settlement along railway lines is a technical problem that urgently needs to be solved in this field. Summary of the Invention
[0005] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.
[0006] Firstly, this application proposes a method for predicting railway thermal thaw settlement in saline-alkali permafrost, including: Based on ground temperature monitoring data, ice content data, salinity data, and InSAR deformation data, a basic database of mileage-depth-time along the railway line is constructed. Based on the above-mentioned mileage-depth-time basic database, the route is divided into saline permafrost thermal thawing control units. For each of the above-mentioned thermal fusion control units, the equivalent thermal fusion potential energy is calculated based on the positive surface accumulated temperature, ice content distribution, and salinity content. The corresponding thermal fusion settlement sensitivity coefficient was determined using indoor fusion settlement test data. Based on the above equivalent fusion potential energy and the above fusion settlement sensitivity coefficient, a thermal-water-salt-force coupling mechanism model for salinized permafrost is constructed. The above-mentioned thermal-water-salt-force coupling mechanism model of saline permafrost was driven by meteorological boundary conditions to obtain the basic thermal fusion settlement prediction results of the above-mentioned thermal fusion control unit. The above-mentioned basic thermal fusion settlement prediction results and the above-mentioned InSAR deformation data are input into the physical constraint machine learning model. The physical constraint machine learning model is dynamically calibrated, and the predicted amount and risk level of thermal fusion settlement at different time scales along the railway are output based on the calibrated model.
[0007] In one feasible implementation, the route is divided into saline permafrost thawing control units based on the aforementioned mileage-depth-time baseline database, including: Based on the information on ground temperature distribution, ice content distribution, salinity distribution and landform type in the above-mentioned mileage-depth-time basic database, a comprehensive analysis of the permafrost structure characteristics of each mileage section along the railway is conducted. Based on the similarity of the above-mentioned geothermal distribution, the continuity of the above-mentioned ice content distribution, and the similarity of the above-mentioned salinity distribution, a partitioned clustering process is performed to obtain the partitioned clustering results. Based on the above-mentioned zoning clustering results, the railway line is divided into multiple saline permafrost thermal thawing control units with similar heat-ice-salt properties.
[0008] In one feasible implementation, the calculation of the equivalent thermal fusion potential energy for each of the aforementioned thermal fusion control units, based on positive surface accumulated temperature, ice content distribution, and salinity, includes: The distribution of positive surface accumulated temperature, ice content as a function of depth, and salinity as a function of depth of the above-mentioned thermal melting control unit within the range from the upper limit of permafrost to the high ice layer were obtained. The above-mentioned positive surface accumulated temperature, ice content distribution, and salinity distribution are input into a preset heat-ice-salt coupling calculation model to output the equivalent thermomelting potential energy.
[0009] In one feasible implementation, the above-mentioned determination of the corresponding thermal fusion settlement sensitivity coefficient using indoor fusion settlement test data includes: The results of indoor thaw settlement tests on representative soil samples from the above-mentioned thermal thaw control unit were obtained, including the thaw settlement deformation under different temperature conditions. The results of the above indoor fusion deposition test and the above equivalent thermal fusion potential energy were fitted and analyzed to obtain the fitting relationship; Based on the above fitting relationship, the fusion settlement sensitivity coefficient, which characterizes the degree of fusion deformation of the fusion control unit, is determined.
[0010] In one feasible implementation, the above-mentioned model for the thermal-water-salt-mechanical coupling mechanism of salinized permafrost, based on the equivalent thermal fusion potential energy and the thermal fusion settlement sensitivity coefficient, includes: The above equivalent fusion potential energy is used as the fusion driving force, and the above fusion settlement sensitivity coefficient is used as the strength degradation control parameter. Coupled control equations for heat conduction, seepage, salt migration and foundation mechanical response are established on a typical profile of the above-mentioned thermal fusion control unit. Based on the above coupled control equations, a thermal-water-salt-force coupled mechanism model for saline permafrost was constructed to simulate the melting and settling process of the thermal melting control unit under the action of heat, ice, and salt.
[0011] In one feasible implementation, the above-mentioned thermo-water-salt-mechanical coupling mechanism model of saline permafrost driven by meteorological boundary conditions obtains the foundation thermo-melting settlement prediction results of the above-mentioned thermo-melting control unit, including: The temperature change, precipitation change and surface heat transfer conditions in the future meteorological elements are applied as boundary inputs to the above-mentioned coupled mechanism model; The above-mentioned thermal-water-salt-force coupling mechanism model of saline permafrost was used to simulate the temperature field changes and melting response of the thermal melting control unit at different time scales. Based on the above temperature field changes and melting response, the basic thermal melting settlement prediction results of the thermal melting control unit are output.
[0012] In one feasible implementation, the aforementioned basic thermal fusion settlement prediction results and the aforementioned InSAR deformation data are input into a physical constraint machine learning model, the physical constraint machine learning model is dynamically calibrated, and based on the calibrated model, the predicted thermal fusion settlement and risk level at different time scales along the railway line are output, including: The above-mentioned basic thermal fusion settlement prediction results are synchronized in time and spatially registered with the above-mentioned InSAR deformation data. The synchronized data is input into the physical constraint machine learning model mentioned above, and the parameters are updated based on the physical consistency constraints within the model. Based on the updated physical constraint machine learning model, the model outputs the predicted thermal fusion settlement along the railway line in the short, medium and long term, and calculates the corresponding thermal fusion settlement risk level based on the predicted amount.
[0013] In one feasible implementation, the above-mentioned physical constraint machine learning model includes: The graph construction unit is used to construct a graph structure based on the spatial adjacency relationship between the above-mentioned thermal fusion control units; The node feature encoding unit is used to encode the equivalent thermal fusion potential energy, the thermal fusion settlement sensitivity coefficient and the basic thermal fusion settlement prediction result corresponding to each of the above thermal fusion control units into a node embedding vector. The edge feature encoding unit is used to encode the edge features reflecting the spatial distance and geological differences between adjacent thermal fusion control units into edge embedding vectors. The graph convolutional inference unit is used to perform message passing and feature aggregation based on the node embedding vector and the edge embedding vector. The physical constraint embedding unit is used to apply physical consistency constraints, including monotonicity constraints and lower bound constraints, during model training to ensure that the model output satisfies the above-mentioned heat-water-salt-force coupling mechanism conditions. The output decoding unit is used to generate the predicted amount of thermal settlement of the thermal control unit based on the output of the graph convolution inference unit.
[0014] In one feasible implementation, parameter updates based on physical consistency constraints within the model include: A joint loss function is constructed, which includes a data fitting loss term and a physical consistency constraint term. The data fitting loss term is used to measure the difference between the thermal fusion settlement prediction output by the physical constraint machine learning model and the InSAR deformation data. The physical consistency constraint term includes a monotonicity constraint term to constrain the thermal fusion settlement prediction to not decrease as the equivalent thermal fusion potential energy increases, and a lower limit constraint term to constrain the thermal fusion settlement prediction to not be less than the basic thermal fusion settlement prediction result. The joint loss function is input into the gradient descent optimizer. Based on the gradient information calculated by the optimizer, the internal parameters of the physical constraint machine learning model are iteratively updated to obtain model parameters that satisfy the physical consistency constraints.
[0015] Secondly, the present invention also proposes a railway thermal thaw settlement prediction and control system for saline-frozen soil, used to execute the above-described method for predicting railway thermal thaw settlement in saline-frozen soil according to any one of the first aspects, including: The first building unit is used to construct a basic database of mileage-depth-time along the railway line based on ground temperature monitoring data, ice content data, salinity data, and InSAR deformation data. The division unit is used to divide the route into saline permafrost thermal thawing control units based on the above-mentioned mileage-depth-time basic database. The calculation unit is used to calculate the equivalent thermal fusion potential energy for each of the above-mentioned thermal fusion control units based on the positive surface accumulated temperature, ice content distribution, and salinity content. The unit is used to determine the corresponding thermo-melting settlement sensitivity coefficient using indoor melting settlement test data; The second building unit is used to construct a thermal-water-salt-force coupling mechanism model of salinized permafrost based on the above equivalent thermal fusion potential energy and the above thermal fusion settlement sensitivity coefficient. The acquisition unit is used to drive the above-mentioned thermal-water-salt-force coupling mechanism model of saline permafrost with meteorological boundary conditions to obtain the basic thermal fusion settlement prediction results of the above-mentioned thermal fusion control unit. The output unit is used to input the above-mentioned basic thermal fusion settlement prediction results and the above-mentioned InSAR deformation data into the physical constraint machine learning model, dynamically calibrate the above-mentioned physical constraint machine learning model, and output the thermal fusion settlement prediction amount and risk level at different time scales along the railway based on the calibrated model.
[0016] In summary, this invention constructs a multi-source data system integrating "heat-ice-salt" attributes and introduces equivalent thaw potential energy and thaw settlement sensitivity coefficient, achieving a quantitative characterization of the thaw potential of saline permafrost. This overcomes the limitations of traditional methods that rely solely on single temperatures or simple empirical relationships, which cannot reflect the internal coupling mechanism of permafrost. Based on this, the established "heat-water-salt-mechanism" coupling mechanism model can simultaneously simulate the coordinated evolution of temperature, moisture, salt, and mechanical fields, realistically reflecting the complex thaw settlement behavior of saline permafrost under thermal disturbance. This invention synchronizes the prediction results of the basic physical model with InSAR deformation data in time and space, and introduces a machine learning model with physical consistency constraints. Through monotonicity constraints, lower bound constraints, and spatial continuity constraints, the prediction model is dynamically calibrated, ensuring it conforms to the physical mechanism and maintains a consistent trend with real-time monitoring data. This method not only significantly improves the accuracy and stability of thaw settlement prediction results but also enables settlement prediction and risk level assessment across the entire railway line at different time scales. This invention has a more comprehensive ability to describe coupling mechanisms, higher prediction accuracy, and stronger engineering applicability, and can effectively support maintenance decisions and structural safety assurance during railway operation.
[0017] Other advantages, objectives and features of this application will be apparent in part from the description which follows, and in part from what those skilled in the art will understand through study and practice of this application. Attached Figure Description
[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting thermal thawing settlement of railways in saline permafrost, as provided by the present invention.
[0019] Figure 2 A flowchart illustrating a method for dividing permafrost into thermal thawing control units, provided by this invention. Figure 3 This is a flowchart illustrating a method for calculating equivalent fusion potential energy provided by the present invention. Figure 4 This is a flowchart illustrating a method for determining the corresponding thermal settling sensitivity coefficient provided by the present invention. Figure 5 This invention provides a flowchart illustrating a method for constructing a thermal-water-salt-mechanical coupling mechanism model of saline permafrost. Figure 6 This is a flowchart illustrating a method for obtaining the predicted thermal settlement of a foundation in a thermal fusion control unit, as provided by the present invention. Figure 7 A flowchart illustrating a method for outputting the predicted amount and risk level of thermal fusion settlement along a railway line at different time scales, provided by the present invention. Figure 8 This invention provides a structural schematic diagram of a railway thermal thawing settlement prediction and control system for saline permafrost. Detailed Implementation
[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.
[0021] Please see Figure 1 This is a flowchart illustrating a method for predicting railway thermal thaw settlement in saline permafrost, provided in an embodiment of this application. Specifically, it may include: S110. Based on ground temperature monitoring data, ice content data, salinity data, and InSAR deformation data, a basic database of mileage-depth-time along the railway line is constructed. S120. Based on the above-mentioned mileage-depth-time basic database, the route is divided into saline permafrost thermal thawing control units. S130. For each of the above-mentioned thermal fusion control units, calculate the equivalent thermal fusion potential energy based on the positive surface accumulated temperature, ice content distribution, and salinity content. S140. Determine the corresponding thermal fusion settlement sensitivity coefficient using indoor fusion settlement test data; S150. Based on the above equivalent fusion potential energy and the above fusion settlement sensitivity coefficient, a thermal-water-salt-force coupling mechanism model for salinized permafrost is constructed. S160. Drive the above-mentioned thermal-water-salt-force coupling mechanism model of saline permafrost with meteorological boundary conditions to obtain the basic thermal fusion settlement prediction results of the above-mentioned thermal fusion control unit. S170. Input the above-mentioned basic thermal fusion settlement prediction results and the above-mentioned InSAR deformation data into the physical constraint machine learning model, dynamically calibrate the above-mentioned physical constraint machine learning model, and output the thermal fusion settlement prediction amount and risk level at different time scales along the railway based on the calibrated model.
[0022] For example, multi-year geothermal sequences are obtained based on geothermal monitoring points deployed along the railway line, and ice content and salinity data are obtained by combining drilling and experimental data. Simultaneously, InSAR technology is used to invert surface deformation information along the railway line. By performing time registration, spatial interpolation, and depth unification processing on the above multi-source data, a mileage-depth-time basic database covering both the railway line mileage direction and the permafrost depth direction is constructed. Based on this database, the permafrost structure along the railway line is comprehensively analyzed according to the similarity of geothermal distribution, the continuity of ice content distribution, and the differences in salinity content. The railway line is then divided into multiple saline permafrost thermomelting control units with similar heat-ice-salt properties.
[0023] After obtaining the divided thermal thawing control units, this embodiment further calculates the surface positive accumulated temperature and determines the intensity of heat-ice-salt coupling at different depths based on the actual ground temperature conditions, ice content variations with depth, and salt content distribution of each control unit. Then, using a pre-set heat-ice-salt coupling calculation model, the equivalent thermal thawing potential energy is obtained to quantitatively describe the potential for thermal thawing and deformation of the control unit under external disturbances. Simultaneously, indoor thaw settlement tests are conducted using representative soil samples from the control units. By analyzing the relationship between thaw settlement deformation and equivalent thermal thawing potential energy under different temperature conditions during the tests, the thermal thawing settlement sensitivity coefficient of the control unit is determined, thereby characterizing the degree of strength degradation of the permafrost structure under thermal thawing.
[0024] Based on the aforementioned equivalent thaw potential energy and thaw settlement sensitivity coefficient, this embodiment establishes a coupled thermal-water-salt-mechanical mechanism model for saline permafrost. By constructing the coupling relationship between the heat conduction equation, seepage equation, salt migration equation, and foundation mechanical equilibrium equation, the nonlinear evolution process of frozen soil under the combined effects of heat, ice, water, and salt is simulated. Furthermore, future meteorological elements (including temperature changes, precipitation processes, and surface heat exchange conditions) are used as the upper boundary input to drive the coupled mechanism model, thereby obtaining the foundation thaw settlement prediction results for each thaw control unit during different prediction periods.
[0025] To improve the accuracy and physical plausibility of the predicted data, this embodiment incorporates the aforementioned basic thermo-melting settlement prediction results and InSAR deformation observations into a physically constrained machine learning model. Physical consistency conditions such as monotonicity constraints, lower limit constraints, and spatial continuity constraints are applied within the model. Dynamic calibration enables adaptive updates of the model parameters, ensuring that the settlement predictions output by the model conform to the statistical regularities of the monitoring data while adhering to the physical mechanisms of permafrost thermo-mechanical evolution. Ultimately, the calibrated physically constrained machine learning model can output predicted thermo-melting settlement along the railway line at different time scales in the short, medium, and long term. Based on the settlement amplitude and unevenness, the corresponding risk level is determined, providing a reliable decision-making basis for railway maintenance, reinforcement, and line adjustments during the operational phase.
[0026] In summary, this invention constructs a multi-source data system integrating "heat-ice-salt" attributes and introduces equivalent thaw potential energy and thaw settlement sensitivity coefficient, achieving a quantitative characterization of the thaw potential of saline permafrost. This overcomes the limitations of traditional methods that rely solely on single temperatures or simple empirical relationships, which cannot reflect the internal coupling mechanism of permafrost. Based on this, the established "heat-water-salt-mechanism" coupling mechanism model can simultaneously simulate the coordinated evolution of temperature, moisture, salt, and mechanical fields, realistically reflecting the complex thaw settlement behavior of saline permafrost under thermal disturbance. This invention synchronizes the prediction results of the basic physical model with InSAR deformation data in time and space, and introduces a machine learning model with physical consistency constraints. Through monotonicity constraints, lower bound constraints, and spatial continuity constraints, the prediction model is dynamically calibrated, ensuring it conforms to the physical mechanism and maintains a consistent trend with real-time monitoring data. This method not only significantly improves the accuracy and stability of thaw settlement prediction results but also enables settlement prediction and risk level assessment across the entire railway line at different time scales. This invention has a more comprehensive ability to describe coupling mechanisms, higher prediction accuracy, and stronger engineering applicability, and can effectively support maintenance decisions and structural safety assurance during railway operation.
[0027] In one feasible implementation, such as Figure 2 As shown, step S120 divides the area along the route into a saline permafrost thermal thawing control unit based on the aforementioned mileage-depth-time baseline database, including: S1201. Based on the information on ground temperature distribution, ice content distribution, salinity distribution and landform type in the above-mentioned mileage-depth-time basic database, a comprehensive analysis of the permafrost structure characteristics of each mileage section along the railway is conducted. S1202. Based on the similarity of the above-mentioned geothermal distribution, the continuity of the above-mentioned ice content distribution, and the similarity of the above-mentioned salinity distribution, perform partitioned clustering processing to obtain partitioned clustering results. S1203. Based on the above-mentioned partitioning and clustering results, the railway line is divided into multiple saline permafrost thermal thawing control units with similar heat-ice-salt properties.
[0028] For example, based on information such as ground temperature changes, ice content distribution characteristics with depth, salt content and geomorphic units recorded in the mileage-depth-time basic database, the permafrost structure state of different mileage sections along the railway is analyzed as a whole, and the differences in thermal field conditions, ice layer structure and salt enrichment degree in different regions are identified.
[0029] Based on the similarity of ground temperature distribution, the continuity of ice content distribution, and the similarity of salinity distribution, this embodiment uses a clustering algorithm to divide the permafrost along the route into zones, so that mileage sections with similar ground temperature, consistent ice content variation trends, and similar salinity levels are classified into the same zone.
[0030] Based on the clustering results, the railway line is divided into multiple saline permafrost thermal thaw control units with consistent characteristics in terms of thermal conditions, ice structure, and salinity composition. Each control unit can serve as a logical carrier for physical and statistical analysis in subsequent thermal thaw potential energy calculation, mechanism model construction, and settlement prediction, thereby improving the accuracy and regional adaptability of the overall prediction process.
[0031] In one feasible implementation, such as Figure 3 As shown, step S130 above calculates the equivalent thermal fusion potential energy for each of the aforementioned thermal fusion control units based on the positive surface accumulated temperature, ice content distribution, and salinity content, including: S1301. Obtain the distribution of surface positive accumulated temperature, ice content with depth, and salinity with depth of the above-mentioned thermal melting control unit within the range from the upper limit of permafrost to the high ice layer. S1302. Input the above-mentioned positive surface accumulated temperature, the above-mentioned ice content distribution and the above-mentioned salt content distribution into the preset heat-ice-salt coupling calculation model to output the equivalent thermal fusion potential energy.
[0032] For example, the surface positive accumulated temperature, ice content, and salinity distribution curves of the thermal thaw control unit within the range from the upper limit of permafrost to the highest ice-bearing layer are obtained as depth. These parameters varying with depth are then uniformly scaled and divided into depth intervals to ensure the continuity and physical consistency of subsequent calculations. Based on this, the surface positive accumulated temperature, ice content distribution, and salinity distribution are input into a preset heat-ice-salt coupled calculation model. By jointly quantifying the effects of heat input, ice layer phase change characteristics, and salinity on freezing point and strength characteristics, the equivalent thermal thaw potential energy, representing the comprehensive thaw potential energy of the control unit, is output.
[0033] The equivalent thermal potential energy can be calculated using an integral model of the following form: in, It represents the equivalent thaw potential energy, used to comprehensively evaluate the thaw potential of frozen soil; The positive accumulated surface temperature within the time window corresponding to this thermal melting control unit is determined by analyzing the surface temperature sequence that is higher than... The portion is obtained by time integration; For depth variables, This represents the upper limit of permafrost depth. This represents the lower boundary depth of the high ice-bearing layer. The ice content stratification weighting function is used to characterize the contribution of high ice content layers at different depths to thermomelting deposition. The higher the ice content, the stronger the amplification effect on thermomelting potential energy. The salt content correction factor reflects the impact of salt on the lowering of the freezing point and the exacerbation of soil structural damage. In a feasible expression, it can be represented as: in, For depth The salt content at that location, The salinity sensitivity coefficient can be calibrated through indoor experiments. As can be seen from the above formula, in the calculation model of this embodiment, the positive accumulated temperature of the surface provides the heat input, the ice content weighting function reflects the phase change potential of the ice layer, and the salinity correction coefficient further amplifies the influence of different salinity contents on the thermal behavior of permafrost, so that the final equivalent thermomelting potential energy can truly reflect the heat-ice-salt coupling characteristics within the thermomelting control unit. This calculation result can not only be used for the subsequent construction of a heat-water-salt-mechanical coupling mechanism model, but also provides a quantitative basis for the differences in thermomelting sensitivity among different control units.
[0034] In one feasible implementation, such as Figure 4 As shown, step S140 above uses indoor fusion settlement test data to determine the corresponding thermal fusion settlement sensitivity coefficient, including: S1401. Obtain the indoor fusion settlement test results of representative soil samples from the above-mentioned thermal fusion control unit, wherein the indoor fusion settlement test results include the fusion settlement deformation under different temperature conditions. S1402. The above indoor fusion test results and the above equivalent thermal fusion potential energy are fitted and analyzed to obtain the fitting relationship. S1403. Based on the above fitting relationship, determine the heat fusion settlement sensitivity coefficient that characterizes the degree of heat fusion deformation of the heat fusion control unit.
[0035] For example, representative undisturbed or remolded soil samples are selected for the thaw control unit, and indoor thaw settlement tests with multiple temperature gradients are conducted under controlled temperature conditions. During the test, the thaw settlement deformation of the soil sample at different temperatures is recorded, thus forming a temperature-thaw settlement test data sequence. Since the thaw settlement of frozen soil is closely related to its internal ice content, salt content, and heat input, this embodiment fits the above-mentioned indoor thaw settlement test results with the equivalent thaw potential energy obtained in step S130 to establish a mathematical relationship between the two.
[0036] In one feasible implementation, this embodiment can utilize a linear or nonlinear fitting model to describe the response relationship between the equivalent thermal fusion potential energy and the fusion deformation. One typical form can be expressed as follows: in, It represents the stable melt-sink strain obtained from indoor tests, used to characterize the final degree of melt-sinking of the sample under specific temperature conditions; This represents the equivalent thermal fusion potential energy calculated according to step S130, which is used to reflect the effective heat absorbed by the sample under the heat-ice-salt coupling effect. This represents the thermal fusion settlement sensitivity coefficient, which is the target parameter of this step and is used to characterize the deformation sensitivity of the frozen soil layer under the same thermal fusion potential energy input in the thermal fusion control unit. This is a fitting bias term used to absorb experimental errors or sample differences.
[0037] In another feasible approach, to more accurately reflect the nonlinear mechanical behavior of saline permafrost, this embodiment can also use a power function or exponential form for fitting, for example: or: Where the constant All data were obtained through indoor experimental data calibration and are used to describe the nonlinear melt-settling response under different ice and salinity conditions.
[0038] Through the above fitting process, the thaw settlement sensitivity coefficient obtained in this embodiment can truly reflect the sensitivity of the frozen soil structure of the thaw control unit to thaw action. When the equivalent thaw potential energy is the same but the test coefficients of different control units are significantly different, it can be determined that the control unit exhibits a stronger tendency for thaw settlement due to high ice content, strong salt sensitivity, or loose soil structure. Therefore, the thaw settlement sensitivity coefficient obtained in this step is not only an important parameter for subsequent coupled mechanism model calculations, but also provides a key constitutive basis for risk zoning along railway lines.
[0039] In one feasible implementation, such as Figure 5 As shown, step S150 above constructs a thermal-water-salt-mechanical coupling mechanism model for salinized permafrost based on the equivalent thermal fusion potential energy and the thermal fusion settlement sensitivity coefficient, including: S1501. The above equivalent fusion potential energy is used as the fusion driving force, and the above fusion settlement sensitivity coefficient is used as the strength degradation control parameter. S1502. Establish coupled control equations for heat conduction, seepage, salt migration and foundation mechanical response on the typical profile of the above-mentioned thermal fusion control unit. S1503. Based on the above coupled control equations, a thermal-water-salt-force coupled mechanism model for saline permafrost is constructed to simulate the melting and settling process of the thermal melting control unit under the action of heat-ice-salt.
[0040] For example, the equivalent fusion potential energy obtained in step S130 is used as the main fusion driving force of the control unit to characterize the intensity of external heat input, and the fusion settlement sensitivity coefficient obtained in step S140 is used as a control parameter to describe the soil strength degradation rate, thereby clarifying the quantitative relationship of thermo-mechanical coupling in the model. Taking the typical profile of the fusion control unit as the computational domain, heat conduction equations, seepage equations, salt migration equations, and foundation mechanical response equations are established, and a complete set of thermo-water-salt-mechanical multi-field coupled control equations is formed under the condition that the boundary conditions and initial conditions are unified. By numerically solving the above equations, the dynamic evolution of the temperature field, moisture field, salt field, and displacement field of the soil during the fusion process can be obtained, thereby simulating the real fusion settlement process of the control unit under the action of heat-ice-salt.
[0041] This embodiment uses the following governing equations to illustrate the coupling behavior of saline permafrost: 1. The governing equation for the temperature field is the heat conduction equation. The heat transfer in frozen soil can be expressed by the heat conduction equation based on energy conservation as follows: in, Density of frozen soil; This refers to the specific heat capacity of frozen soil. For temperature; Thermal conductivity; Latent heat of phase transition; The value represents the ice content; this equation describes the coupling relationship between temperature change and the endothermic phase transition of ice and water.
[0042] 2. The governing equation for the moisture field is the seepage equation, and the migration of moisture generated by melting can be described by Darcy's law and the water content equilibrium relationship: in, Moisture content; ρ is the permeability coefficient; h is the hydraulic potential; this equation reflects the influence of meltwater redistribution on the pore structure.
[0043] 3. The governing equation for the salt field is the salt migration equation. The salt variation under the combined effects of salt diffusion and water migration can be expressed as: in, Salt content; The salt diffusion coefficient; This refers to the water flow rate; salt migration can cause the freezing point to drop and affect the strength of the soil skeleton.
[0044] 4. The governing equations of the mechanical field are the foundation mechanical response equations. The mechanical response of frozen soil during the thawing and settlement process can be described by the equilibrium equations: in, For stress tensor; This is the acceleration due to gravity.
[0045] The constitutive relation of soil can be written as: Among them, the material stiffness tensor Sensitivity coefficient of heat melt sedimentation Salt content and ice content It degenerates due to changes.
[0046] The four types of governing equations mentioned above together constitute the "thermal-water-salt-mechanical" coupled system of permafrost. After solving them using numerical methods (such as the finite element method), this embodiment can obtain the complete evolution process of the temperature field, moisture field, salinity field, and displacement field at different time steps. Therefore, the constructed thermal-water-salt-mechanical coupled mechanism model of saline permafrost can accurately simulate the thaw settlement process in complex permafrost environments, providing a scientific physical basis for subsequent settlement prediction based on monitoring and calibration.
[0047] In one feasible implementation, such as Figure 6 As shown, step S160 above uses meteorological boundary conditions to drive the above-mentioned thermal-water-salt-mechanical coupling mechanism model of saline permafrost to obtain the basic thermal thaw settlement prediction results of the above-mentioned thermal thaw control unit, including: S1601. Apply the changes in temperature, precipitation and surface heat transfer conditions in future meteorological elements as boundary inputs to the above-mentioned coupled mechanism model. S1602. The above-mentioned thermal-water-salt-force coupling mechanism model of saline permafrost was used to simulate the temperature field change and melting response of the thermal melting control unit at different time scales. S1603. Based on the above temperature field changes and melting response, output the basic melting settlement prediction results of the thermal melting control unit.
[0048] For example, external driving forces such as temperature changes, precipitation changes, and heat transfer coefficients reflecting surface heat exchange relationships from climate forecasts or typical meteorological year data are applied as upper boundary conditions to the thermal-water-salt-mechanical coupling mechanism model of saline permafrost constructed in step S150, enabling the model to conduct numerical evolution simulations under real or predicted environmental changes. Under the above boundary conditions, the model continuously solves the coupled control equations of the temperature field, moisture field, salinity field, and mechanical field, thereby obtaining the changes in freezing state, phase change heat release, meltwater migration process, and the resulting foundation deformation response at each time step. In this embodiment, based on the correspondence between temperature field changes and melt settlement response, the cumulative melt settlement amount and settlement rate within each prediction period are extracted to form the foundation melt settlement prediction results of the thermal melt control unit.
[0049] In one feasible implementation, the driving force of future meteorological elements on the temperature field can be expressed in the following boundary form: in, The effective thermal conductivity of the frozen soil surface layer; It refers to the surface temperature; Temperature is the primary driver in meteorological data; The surface heat transfer coefficient is calculated by comprehensively considering radiation, convection, and surface properties.
[0050] Changes in precipitation can be influenced by adjusting surface infiltration flux, thereby affecting the moisture field. in, This refers to the infiltration flux; Permeability coefficient; External water head caused by precipitation infiltration; To freeze the internal hydraulic potential of the soil.
[0051] Under the aforementioned meteorological boundary inputs, the evolution of the temperature field, moisture field, and salinity field will further affect the soil's mechanical behavior. The settlement displacement can be obtained through constitutive relations and soil equilibrium equations, for example: in, This represents the cumulative thermal settling amount; The instantaneous melt-settling strain rate can be calculated from the temperature field change and the thermo-melt-settling sensitivity coefficient.
[0052] By running the above model under different future meteorological conditions, such as short-term (1-3 years), medium-term (10-20 years) and long-term (30-50 years), this embodiment can obtain the temperature field evolution, cumulative melting and settling values and settlement rates at each time scale, thereby outputting the basic thermal melting and settling prediction results of the thermal melting control unit, providing an initial physical scenario and trend judgment for subsequent dynamic calibration based on monitoring data.
[0053] In one feasible implementation, such as Figure 7 As shown, step S170 above inputs the above-mentioned basic thermal fusion settlement prediction results and the above-mentioned InSAR deformation data into the physical constraint machine learning model, dynamically calibrates the above-mentioned physical constraint machine learning model, and outputs the predicted amount and risk level of thermal fusion settlement at different time scales along the railway line based on the calibrated model, including: S1701. Synchronize and spatially register the above-mentioned basic thermal fusion settlement prediction results with the above-mentioned InSAR deformation data. S1702. Input the synchronized data into the above physical constraint machine learning model, and update the parameters based on the physical consistency constraints inside the model. S1703. Based on the updated physical constraint machine learning model, output the predicted thermal fusion settlement along the railway line in the short, medium and long term, and calculate the corresponding thermal fusion settlement risk level based on the predicted amount.
[0054] In one feasible implementation, the above-mentioned physical constraint machine learning model includes: The graph construction unit is used to construct a graph structure based on the spatial adjacency relationship between the above-mentioned thermal fusion control units; The node feature encoding unit is used to encode the equivalent thermal fusion potential energy, the thermal fusion settlement sensitivity coefficient and the basic thermal fusion settlement prediction result corresponding to each of the above thermal fusion control units into a node embedding vector. The edge feature encoding unit is used to encode the edge features reflecting the spatial distance and geological differences between adjacent thermal fusion control units into edge embedding vectors. The graph convolutional inference unit is used to perform message passing and feature aggregation based on the node embedding vector and the edge embedding vector. The physical constraint embedding unit is used to apply physical consistency constraints, including monotonicity constraints and lower bound constraints, during model training to ensure that the model output satisfies the above-mentioned heat-water-salt-force coupling mechanism conditions. The output decoding unit is used to generate the predicted amount of thermal settlement of the thermal control unit based on the output of the graph convolution inference unit.
[0055] In one feasible implementation, parameter updates based on physical consistency constraints within the model include: A joint loss function is constructed, which includes a data fitting loss term and a physical consistency constraint term. The data fitting loss term is used to measure the difference between the thermal fusion settlement prediction output by the physical constraint machine learning model and the InSAR deformation data. The physical consistency constraint term includes a monotonicity constraint term to constrain the thermal fusion settlement prediction to not decrease as the equivalent thermal fusion potential energy increases, and a lower limit constraint term to constrain the thermal fusion settlement prediction to not be less than the basic thermal fusion settlement prediction result. The joint loss function is input into the gradient descent optimizer. Based on the gradient information calculated by the optimizer, the internal parameters of the physical constraint machine learning model are iteratively updated to obtain model parameters that satisfy the physical consistency constraints.
[0056] For example, step S170 serves to fuse the basic thermomelting settlement prediction results obtained from the physical mechanism model with InSAR monitoring data, enabling the model to both follow the physical laws of permafrost and absorb observational information in real time, thus achieving continuous calibration and optimization of prediction capabilities. To this end, this embodiment first performs time synchronization and spatial registration between the basic prediction results and InSAR deformation data to ensure that the two types of data are within comparable scales and coordinate systems. Subsequently, the synchronized data is input into a physical constraint machine learning model, and the model parameters are dynamically updated by applying physical consistency constraints, maintaining coordination between the predicted values and actual monitoring trends. Finally, the calibrated model outputs settlement prediction values at different time scales, and risk levels are determined through threshold rules, thereby forming a complete spatialized, periodic thermomelting settlement risk assessment result.
[0057] In one feasible implementation, this step first involves the basic predicted settlement sequence. and Deformation sequence Perform uniform interpolation and registration to satisfy: in, These are discrete locations along the railway line. For monitoring time series.
[0058] Based on the synchronized data, it is input into a physical constraint machine learning model to construct a joint loss function: in, The fitting loss is used to measure the model's output sedimentation. and Differences between monitoring quantities: This is a monotonicity constraint term used to ensure that the predicted settlement does not decrease as the equivalent thermal fusion potential energy increases, such as: This is a lower bound constraint term, ensuring that the predicted quantity is not less than the basic physical prediction quantity, i.e.: This is a smoothing constraint term used to suppress unreasonable abrupt changes in the prediction results across the line space, such as: These are weighting coefficients used to control the importance of physical constraints.
[0059] Use gradient descent or its improved algorithms (such as Adam) to analyze the model parameters. Update: in, For learning rate, For the first Model parameters for the next iteration.
[0060] Through the aforementioned dynamic calibration process, the physical constraint machine learning model can simultaneously consider the reliability of monitoring data and the physical consistency of permafrost evolution, thereby improving the stability and reliability of predictions.
[0061] Finally, based on the calibrated model, this embodiment outputs the predicted thermal fusion settlement along the railway line in the short term (1-3 years), medium term (10-20 years), and long term (30-50 years). Based on the predicted settlement amplitude, settlement rate, and degree of uneven deformation in adjacent sections, the risk level is divided into levels such as "normal, controllable, early warning, and severe early warning", providing a scientific basis for railway line operation, maintenance, and engineering intervention.
[0062] The second aspect, such as Figure 8 As shown, the present invention also proposes a railway thermal thaw settlement prediction and control system for saline-frozen soil, used to execute the above-described method for predicting railway thermal thaw settlement in saline-frozen soil, comprising: The first building unit 21 is used to build a basic database of mileage-depth-time along the railway line based on ground temperature monitoring data, ice content data, salinity data and InSAR deformation data. Division unit 22 is used to divide the route into saline permafrost thermal thawing control units based on the above-mentioned mileage-depth-time basic database. Calculation unit 23 is used to calculate the equivalent thermal fusion potential energy for each of the above-mentioned thermal fusion control units based on the positive surface accumulated temperature, ice content distribution and salinity content; Unit 24 is used to determine the corresponding thermal fusion settlement sensitivity coefficient using indoor fusion settlement test data; The second building unit 25 is used to construct a thermal-water-salt-force coupling mechanism model of salinized permafrost based on the above equivalent thermal fusion potential energy and the above thermal fusion settlement sensitivity coefficient. Acquisition unit 26 is used to drive the above-mentioned heat-water-salt-force coupling mechanism model of saline permafrost with meteorological boundary conditions to obtain the basic thermal fusion settlement prediction results of the above-mentioned thermal fusion control unit. Output unit 27 is used to input the above-mentioned basic thermal fusion settlement prediction results and the above-mentioned InSAR deformation data into the physical constraint machine learning model, dynamically calibrate the above-mentioned physical constraint machine learning model, and output the thermal fusion settlement prediction amount and risk level at different time scales along the railway based on the calibrated model.
[0063] The railway thermal thaw settlement prediction and control system for saline permafrost proposed in this invention can also perform the method described in any of the first aspects.
[0064] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for predicting thermal thawing settlement of railways in saline-alkali permafrost, characterized in that, include: Based on ground temperature monitoring data, ice content data, salinity data, and InSAR deformation data, a basic database of mileage-depth-time along the railway line is constructed. Based on the aforementioned mileage-depth-time database, the route is divided into saline permafrost thermal thawing control units. For each of the aforementioned thermal fusion control units, the equivalent thermal fusion potential energy is calculated based on the positive accumulated temperature of the surface, the distribution of ice content, and the salinity content. The corresponding thermal fusion settlement sensitivity coefficient was determined using indoor fusion settlement test data. A thermal-water-salt-force coupling mechanism model for salinized permafrost is constructed based on the equivalent thermal fusion potential energy and the thermal fusion settlement sensitivity coefficient. The thermal-water-salt-force coupling mechanism model of the saline permafrost is driven by meteorological boundary conditions to obtain the basic thermal fusion settlement prediction results of the thermal fusion control unit. The basic thermo-melting settlement prediction results and the InSAR deformation data are input into the physical constraint machine learning model. The physical constraint machine learning model is dynamically calibrated, and the predicted amount and risk level of thermo-melting settlement at different time scales along the railway are output based on the calibrated model.
2. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, Based on the aforementioned mileage-depth-time baseline database, the area along the route is divided into saline permafrost thermal thawing control units, including: Based on the information on ground temperature distribution, ice content distribution, salinity distribution and landform type in the mileage-depth-time basic database, a comprehensive analysis of the permafrost structure characteristics of each mileage section along the railway is conducted. Based on the similarity of the geothermal distribution, the continuity of the ice content distribution, and the similarity of the salinity distribution, a partitioned clustering process is performed to obtain the partitioned clustering results. Based on the partitioning clustering results, the railway line is divided into multiple saline permafrost thermal thawing control units with similar heat-ice-salt properties.
3. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, The calculation of the equivalent thermal fusion potential energy for each of the aforementioned thermal fusion control units, based on positive surface accumulated temperature, ice content distribution, and salinity, includes: The distribution of positive surface accumulated temperature, ice content with depth, and salinity with depth of the thermal melting control unit within the range from the upper limit of permafrost to the high ice layer are obtained. The positive accumulated temperature of the land surface, the distribution of ice content, and the distribution of salinity are input into a preset heat-ice-salt coupling calculation model to output the equivalent thermomelting potential energy.
4. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, The determination of the corresponding thermal fusion settlement sensitivity coefficient using indoor fusion settlement test data includes: Obtain the indoor thaw settlement test results of representative soil samples from the thermal thaw control unit, wherein the indoor thaw settlement test results include the thaw settlement deformation under different temperature conditions; The indoor fusion test results and the equivalent thermal fusion potential energy are fitted and analyzed to obtain the fitting relationship; Based on the fitting relationship, a heat fusion settlement sensitivity coefficient, which characterizes the degree of heat fusion deformation of the heat fusion control unit, is determined.
5. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, The model for the thermal-water-salt-mechanical coupling mechanism of salinized permafrost, based on the equivalent fusion potential energy and the fusion settlement sensitivity coefficient, includes: The equivalent fusion potential energy is used as the fusion driving force, and the fusion settlement sensitivity coefficient is used as the strength degradation control parameter. Coupled control equations for heat conduction, seepage, salt migration, and foundation mechanical response are established on a typical profile of the thermal fusion control unit. Based on the aforementioned coupled control equations, a thermal-water-salt-force coupled mechanism model for saline permafrost was constructed to simulate the melting and settling process of the thermal melting control unit under the action of heat, ice, and salt.
6. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, The above-mentioned model of the thermo-water-salt-mechanical coupling mechanism of saline permafrost driven by meteorological boundary conditions obtains the basic thermo-melt settlement prediction results of the thermo-melt control unit, including: The future meteorological elements, such as temperature changes, precipitation changes, and surface heat transfer conditions, are applied as boundary inputs to the coupled mechanism model. The thermal-water-salt-force coupling mechanism model of the saline permafrost was used to simulate the temperature field changes and melting response of the thermal melting control unit at different time scales. Based on the temperature field change and melting response, the basic thermal melting settlement prediction result of the thermal melting control unit is output.
7. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 1, characterized in that, The process of inputting the basic thermal fusion settlement prediction results and the InSAR deformation data into a physical constraint machine learning model, dynamically calibrating the physical constraint machine learning model, and outputting the predicted thermal fusion settlement and risk level at different time scales along the railway line based on the calibrated model includes: The basic thermal fusion settlement prediction results are synchronized in time and spatially registered with the InSAR deformation data; The synchronized data is input into the physical constraint machine learning model, and the parameters are updated based on the physical consistency constraints within the model. Based on the updated physical constraint machine learning model, the predicted thermal fusion settlement along the railway line for the short, medium and long term is output, and the corresponding thermal fusion settlement risk level is calculated based on the predicted amount.
8. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 7, characterized in that, The physical constraint machine learning model includes: Graph construction unit, the graph construction unit is used to construct a graph structure based on the spatial adjacency relationship between the thermal fusion control units; A node feature encoding unit is used to encode the equivalent thermal fusion potential energy, the thermal fusion settlement sensitivity coefficient and the basic thermal fusion settlement prediction result corresponding to each thermal fusion control unit into a node embedding vector. An edge feature encoding unit is used to encode edge features reflecting the spatial distance and geological differences between adjacent thermal fusion control units into edge embedding vectors. The graph convolutional inference unit is used to perform message passing and feature aggregation based on the node embedding vector and the edge embedding vector; A physical constraint embedding unit is used to apply physical consistency constraints, including monotonicity constraints and lower bound constraints, during model training to ensure that the model output satisfies the heat-water-salt-force coupling mechanism conditions. An output decoding unit is used to generate the predicted amount of thermal settlement of the thermal fusion control unit based on the output of the graph convolution inference unit.
9. The method for predicting railway thermal thaw settlement in saline-alkali permafrost as described in claim 8, characterized in that, Parameter updates based on the physical consistency constraints within the model include: A joint loss function is constructed, which includes a data fitting loss term and a physical consistency constraint term. The data fitting loss term is used to measure the difference between the thermal fusion settlement prediction output by the physical constraint machine learning model and the InSAR deformation data. The physical consistency constraint term includes a monotonicity constraint term to constrain the thermal fusion settlement prediction to not decrease as the equivalent thermal fusion potential energy increases, and a lower limit constraint term to constrain the thermal fusion settlement prediction to not be less than the basic thermal fusion settlement prediction result. The joint loss function is input into the gradient descent optimizer, and the internal parameters of the physical constraint machine learning model are iteratively updated based on the gradient information calculated by the optimizer to obtain model parameters that satisfy the physical consistency constraint.
10. A system for predicting railway thermal thaw settlement in saline-fertile permafrost, used to execute the method for predicting railway thermal thaw settlement in saline-fertile permafrost as described in any one of claims 1 to 9, characterized in that, include: The first building unit is used to construct a basic database of mileage-depth-time along the railway line based on ground temperature monitoring data, ice content data, salinity data, and InSAR deformation data. A division unit is used to divide the route into a saline permafrost thermal thawing control unit based on the mileage-depth-time basic database. The calculation unit is used to calculate the equivalent thermal fusion potential energy for each of the thermal fusion control units based on the positive accumulated temperature of the surface, the distribution of ice content, and the salinity content. The unit is used to determine the corresponding thermo-melting settlement sensitivity coefficient using indoor melting settlement test data; The second building unit is used to construct a thermal-water-salt-force coupling mechanism model of salinized permafrost based on the equivalent thermal fusion potential energy and the thermal fusion settlement sensitivity coefficient. The acquisition unit is used to drive the thermal-water-salt-force coupling mechanism model of the saline permafrost with meteorological boundary conditions to acquire the basic thermal fusion settlement prediction results of the thermal fusion control unit. The output unit is used to input the basic thermal fusion settlement prediction results and the InSAR deformation data into the physical constraint machine learning model, dynamically calibrate the physical constraint machine learning model, and output the thermal fusion settlement prediction amount and risk level at different time scales along the railway based on the calibrated model.