A Roadbed Settlement Prediction Method and System Based on Multi-Source Data
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHENGZHOU MUNICIPAL ENG SURVEY DESIGN&RES INST
- Filing Date
- 2025-12-16
- Publication Date
- 2026-07-17
Smart Images
Figure CN121706475B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of civil engineering technology, and in particular to a method and system for predicting roadbed settlement based on multi-source data. Background Technology
[0002] With the continuous expansion of highway networks and intercity roads, roadbeds are prone to uneven settlement and accumulation of minor settlement during long-term service due to the combined effects of rainfall, groundwater level changes, and traffic loads, affecting driving safety and road lifespan. Traditional roadbed settlement monitoring and prediction technologies have mainly followed the following development path: In the early stages, manual inspections and experience-based judgments were the primary methods, making it difficult to detect early settlement hazards in a timely manner; subsequently, point monitoring methods such as leveling and settlement plates were introduced, which could obtain the settlement evolution of some sections, but the spatial coverage was limited, the monitoring cycle was long, and the response to sudden or cumulative settlement was delayed; in recent years, roadbed settlement calculation based on numerical analysis methods such as finite element method has been developed, but it mostly relies on design conditions and simplified boundary conditions, making it difficult to reflect the dynamic changes in actual rainfall processes and soil moisture content.
[0003] To address the impact of rainfall, existing technologies have begun to utilize rain gauge data or limited sensor data to conduct correlation analysis between rainfall intensity and roadbed deformation. However, these methods generally suffer from the following shortcomings: First, most methods rely on a single data source (such as ground rain gauges or a limited number of soil moisture monitoring points), lacking the comprehensive utilization of multi-source data such as satellite remote sensing precipitation, ground sensors, and historical settlement records. This makes it difficult to reflect the true hydrological and soil conditions of the roadbed area in a timely and comprehensive manner. Second, existing assessments often remain at the empirical correlation level of "rainfall intensity - settlement response," failing to systematically characterize the "precipitation process - spatial and temporal distribution of soil moisture - soil saturation" process. The dynamic mechanism chain of "state change - decreased support capacity - settlement" makes it difficult to quantitatively describe the differential impact between soil layers of different depths and different spatial locations; third, existing methods for rainfall-induced settlement risk mostly use static or single-period indicators, lacking modeling of the cumulative effect of rainfall and the probability evolution process, making it difficult to provide spatially resolved settlement prediction intervals and roadbed stability distributions at the regional scale; fourth, existing results rarely deeply integrate settlement prediction results with roadbed structural parameters, topography, and geographic information systems, making it difficult to provide a reliable basis for identifying minor settlement risks in specific road sections and making refined maintenance decisions.
[0004] Therefore, how to establish a roadbed stability assessment method that can dynamically reflect the precipitation infiltration process, accurately simulate the soil state evolution, quantitatively assess the attenuation of support capacity, and ultimately predict the settlement space interval, supported by multi-source data, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention
[0005] This application provides a method and system for predicting road subgrade settlement based on multi-source data. It aims to address the problems in existing technologies, such as insufficient utilization of multi-source data, difficulty in depicting the dynamic coupling mechanism of precipitation-soil-settlement, difficulty in considering time cumulative effects and spatial heterogeneity, and difficulty in integrating prediction results with road subgrade structural information to form usable stability assessment results. The application provides a technical solution that can achieve dynamic prediction and stability assessment of road subgrade settlement at a regional scale.
[0006] In a first aspect, this application provides a method for predicting roadbed settlement based on multi-source data, the method comprising:
[0007] S1. Acquire precipitation sequence and soil moisture distribution data of roadbed within a specified area through multi-source data acquisition, correct the acquired data, and obtain the corrected precipitation sequence and soil moisture distribution data.
[0008] S2. Analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data to obtain the classification results.
[0009] S3. Based on the classification results, extract the estimated seepage depth under the corresponding working conditions, and simulate the changing trend of soil saturation state based on the estimated seepage depth.
[0010] S4. Analyze the soil support capacity index based on the trend of soil saturation state changes, and determine the data on the decrease in soil support capacity.
[0011] S5. Construct a settlement initiation probability calculation model based on the descent amplitude data to obtain the settlement initiation probability distribution of the road subgrade;
[0012] S6. Based on the probability distribution of settlement initiation and combined with the cumulative impact factors of precipitation, predict the settlement range of the roadbed.
[0013] S7. Integrate the settlement prediction range with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability assessment results.
[0014] Secondly, this application provides a roadbed settlement prediction system based on multi-source data, the system comprising:
[0015] The data acquisition module is used to acquire precipitation sequences and soil moisture distribution data of the roadbed within a specified area through multi-source data acquisition, correct the acquired data, and obtain corrected precipitation sequences and soil moisture distribution data.
[0016] The risk classification module is used to analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and to classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data, and obtain the classification results.
[0017] The permeability simulation module is used to extract the estimated permeability depth under the corresponding working conditions based on the classification results, and to simulate the changing trend of soil saturation state based on the estimated permeability depth.
[0018] The capacity analysis module is used to analyze the soil support capacity index based on the changing trend of soil saturation state, and to determine the rate of decrease in soil support capacity.
[0019] The probability calculation module is used to construct a settlement-induced probability calculation model based on the descent data and obtain the settlement-induced probability distribution of the roadbed.
[0020] The settlement prediction module is used to predict the settlement range of roadbed based on the settlement initiation probability distribution and the cumulative impact of precipitation.
[0021] The fusion assessment module is used to merge the settlement prediction range with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability assessment results.
[0022] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows:
[0023] 1. By jointly analyzing satellite remote sensing precipitation data and soil moisture data collected by ground sensor networks, and constructing a correlation model for systematic bias correction, the shortcomings of insufficient spatial coverage and large measurement errors of a single data source are effectively overcome. This results in obtaining precipitation and soil moisture distribution data that are consistent in time and space and have higher accuracy, providing high-quality input for all subsequent analysis stages.
[0024] 2. By classifying the risk of precipitation infiltration in roadbed areas based on the cumulative distribution characteristics of precipitation and soil moisture distribution data, and further combining the estimation of infiltration depth, the trend of soil saturation state change, the rate of change of soil saturation and the increment of pore pressure, a dynamic mechanism chain model of "precipitation process - spatial and temporal distribution of soil moisture - change of soil saturation state - decrease in support capacity - probability of settlement" is established. This model can characterize the differences and evolution of settlement risk at different soil depths and different spatial locations, improving the accuracy of settlement prediction and the interpretability of the mechanism.
[0025] 3. By constructing a probabilistic model based on data on declining support capacity and calibrating it using historical settlement records, a settlement-induced probability distribution conforming to engineering statistical laws was generated. Furthermore, by combining the cumulative effect of precipitation with high-risk evolution zone identification technology for interval prediction, the prediction results not only reflect immediate risks but also capture long-term cumulative trends, significantly improving the timeliness of early warnings and adaptability to complex service conditions.
[0026] 4. By using a geographic information system (GIS) to perform weighted fusion and spatial analysis of the settlement prediction interval and the actual roadbed structure data, a visualized roadbed stability distribution map and risk characteristics of specific road sections are generated. The output assessment results integrate quantitative prediction, risk distribution, and stability indicators, providing intuitive, quantitative, and operable direct technical support for road maintenance, reinforcement design, and safe operation and maintenance. Attached Figure Description
[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0028] Figure 1 This is a flowchart of the roadbed settlement prediction method based on multi-source data according to this application;
[0029] Figure 2 This is a schematic diagram of the roadbed settlement prediction system based on multi-source data of this application. Detailed Implementation
[0030] 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 particular 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” or “having,” and any variations thereof, are intended to cover a 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.
[0031] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of the roadbed settlement prediction method based on multi-source data provided by the present invention. The flowchart specifically includes the following steps:
[0032] S1. Acquire precipitation sequence and soil moisture distribution data of the roadbed within the specified area through multi-source data acquisition, correct the acquired data, and obtain the corrected precipitation sequence and soil moisture distribution data.
[0033] In one specific embodiment, the process of performing step S1 may specifically include the following steps:
[0034] Continuous precipitation sequences within a designated area are obtained through satellite remote sensing data, and soil moisture distribution data at roadbed locations are collected through a ground sensor network deployed on the roadbed.
[0035] Based on soil moisture distribution data, sampling was conducted at different depths of the roadbed at multiple time periods to analyze the moisture retention characteristics of each depth layer, forming multi-time period variation data to characterize changes in moisture retention time and retention rate.
[0036] Time series analysis methods are used to process continuous precipitation sequences and multi-period variation data to generate basic characteristic data of cumulative changes in precipitation at different time scales;
[0037] Based on basic feature data, a correlation model is constructed between precipitation and the moisture distribution of soil at different depths of the roadbed.
[0038] The bias in the multi-source data acquisition is corrected based on the correlation model, and the corrected precipitation sequence and corrected soil moisture distribution data are output.
[0039] Specifically, during the data acquisition phase, continuous precipitation sequences within a specific time window are extracted from satellite precipitation inversion products within a designated road corridor. These sequences are discretized at fixed time steps and denoted as the regional average precipitation value, numbered by time step. Correspondingly, a sensor network deployed along the roadbed collects soil moisture information at multiple representative cross-sections and lateral locations. Each monitoring point is buried in layers, such as at depths of 0.5m, 1.5m, and 3m, forming a three-dimensional data grid of "time-depth-location". At the same time, moisture content readings at each depth layer can be obtained. After multiple monitoring cycles, a multi-time-period sample sequence reflecting the trajectory of moisture content changes at each depth layer is formed. By organizing these sequences, a moisture content curve ordered by time is constructed for the same depth at the same monitoring point, allowing observation of the retention and attenuation of moisture at different depths during the onset of rainfall, peak rainfall, and post-rain dissipation.
[0040] To extract quantities describing retention behavior from these raw monitoring curves, several features are calculated for the water content time series over an observation period for each depth layer. These features include the time interval during which water content rises from its initial level to its peak, the increase in peak water content relative to the initial water content, and the time required for water content to decay from its peak to a certain recovery threshold after rainfall. These features are used to construct quantitative indicators of "water retention time" and "water retention rate change," which are then archived by depth to form so-called multi-period variation data. This type of data has a one-to-one correspondence with precipitation time series on the time axis: for any rainfall event, there exists a corresponding set of depth-layer retention indicators, thus establishing a link between "a rainfall event – a water content response curve – a set of retention features."
[0041] After completing spatiotemporal registration, time-series analysis methods are used to jointly process precipitation sequences and multi-period variation data. For the precipitation sequences, cumulative calculations are performed at multiple typical time scales, such as calculating sliding cumulative rainfall within 1-hour, 6-hour, 24-hour, and 72-hour windows, obtaining a set of cumulative curves that vary over time. These curves constitute the basic characteristic data of "cumulative changes in precipitation at different time scales." The selection of different time scales can be determined by considering the typical duration of heavy rainfall and the soil permeability characteristics of the road area, making short-term windows sensitive to sudden heavy rainfall and long-term windows reflecting long-term humid backgrounds. Within the corresponding observation period, each retention characteristic sequence in the multi-period variation data corresponds to the cumulative precipitation curve on the same time axis. Therefore, correlation analysis, lag analysis, and other methods can be used to analyze the quantitative relationship between the water content response at a certain depth and cumulative rainfall at different time scales. For example, it can identify at what level of cumulative rainfall at a certain scale the water content at that depth layer significantly increases, or under what accumulation mode the retention time is significantly prolonged, thus mapping the "precipitation process" to "retention behavior."
[0042] After the basic feature data is constructed, the cumulative precipitation features of different time windows can be combined into an input vector R(t), and the water content or retention features of each depth layer of the road subgrade at corresponding times can be combined into an output vector W(t). Based on this, a correlation model between precipitation and soil moisture distribution at each depth layer can be constructed. The correlation model can adopt a multiple linear regression model. Preferably, a feedforward neural network model can be used. This model consists of an input layer, one or more hidden layers, and an output layer set according to the depth dimension. The input layer and hidden layers, and the hidden layers and output layers are fully connected. The model parameters include the connection weights and bias terms. During training, multi-time period sample pairs formed by previous rainfall events are used as the training set. R(t) and W(t) are normalized, mean square error is selected as the loss function, and hyperparameters such as learning rate are set. The weights and biases are iteratively updated through gradient descent or other numerical optimization algorithms so that the predicted soil moisture distribution output by the model approximates the actual distribution measured by the sensor. Thus, the mapping relationship between "cumulative rainfall pattern → depth water content level" is fitted on a large number of samples. This can infer the systematic bias of satellite data in terms of magnitude or time positioning, and at the same time identify the zero drift or range offset of some sensors during long-term use.
[0043] After the correlation model is built, when using this model to correct the deviations in multi-source data, on the one hand, the original satellite precipitation sequence is input into the model and compared with the "expected water content response" inferred from the sensor data. When a systematic difference is found between the satellite-inverted precipitation and the precipitation demand inferred from the soil response in long-term statistics, the satellite sequence is numerically adjusted by correction coefficients or offset terms to obtain a corrected precipitation sequence that is more consistent with the measured soil response at the road corridor scale. On the other hand, the residual distribution between the model prediction and the measured values is used to judge the water content time series of each sensor point. When a sensor continuously shows a deviation in the same direction from the model prediction in multiple rainfall events, while no similar phenomenon is found in the surrounding similar soil monitoring points, it can be determined that the sensor has measurement drift. The correction factor is inferred through residual statistics to adjust its water content data to a level consistent with the overall correlation, thereby generating corrected soil moisture distribution data.
[0044] At the regional scale, the corrected precipitation sequences for all time periods, along with the corrected soil moisture distribution data at the corresponding time and depth, are stored in a unified dataset. This allows for the direct retrieval of multidimensional information on the "precipitation-water content-retention characteristics" of any given time segment. At the temporal scale, these data form a continuous evolutionary trajectory, providing temporally continuous input for subsequent steps such as risk classification, infiltration depth estimation, and saturation trend simulation. Through this construction method, multi-source data with inconsistent spatiotemporal resolution and different physical meanings are transformed into a set of mutually registered and bias-corrected basic data. This solves problems such as the difficulty of using single rain gauge data to characterize the overall rainfall distribution of the road corridor, discrepancies between satellite precipitation inversion and actual ground conditions, and the inability to directly use a small amount of sensor point water content data for regional analysis.
[0045] S2. Analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data to obtain the classification results.
[0046] In one specific embodiment, the process of performing step S2 may specifically include the following steps:
[0047] Based on the corrected precipitation sequence, the cumulative distribution characteristics of precipitation at different time scales are extracted;
[0048] Based on the cumulative distribution characteristics of precipitation, we analyze short-duration heavy precipitation patterns and long-duration continuous precipitation patterns;
[0049] If the cumulative precipitation distribution characteristics show a short-term heavy precipitation pattern, the corresponding precipitation sequence will be marked as a high infiltration risk category; if the cumulative precipitation distribution characteristics show a long-term continuous precipitation pattern, the corresponding precipitation sequence will be marked as a low infiltration risk category.
[0050] For high permeability risk categories, water infiltration path characteristics are extracted by combining the corrected soil moisture distribution data.
[0051] Based on the characteristics of water infiltration pathways, the preferential infiltration areas for water distribution in deep layers are determined;
[0052] Using data classification technology, the cumulative distribution characteristics of precipitation, infiltration risk categories and information on priority infiltration areas are integrated to classify the corrected precipitation sequence into multiple precipitation infiltration risk levels, generating classification results of precipitation infiltration risk after classification.
[0053] Specifically, precipitation sequences and soil moisture distribution data registered and corrected for bias under a unified time coordinate have been obtained. For a specific road section, within an observation period ranging from one day to several weeks, at any time t, the precipitation per unit time, cumulative precipitation under different time windows, and moisture content distribution at different depths and planar locations of the roadbed can be read simultaneously. To identify the precipitation conditions and infiltration paths most sensitive to settlement from these data, during S2 execution, multi-scale cumulative calculations are first performed on the corrected precipitation sequence in the time dimension. Based on the rainfall intensity sequence with a fixed time step, short-term windows such as 1 hour, 3 hours, 6 hours, and 24 hours, and long-term windows such as 72 hours and one week are constructed. For each time t, the sliding cumulative precipitation within these windows is calculated, forming a multi-dimensional time series vector R1(t), where different components correspond to cumulative rainfall at different time scales. This vector serves as a quantitative expression of the "cumulative precipitation distribution characteristics".
[0054] Based on this, frequency domain and time series analyses are performed on R1(t). The multi-scale cumulative curve on the time axis is decomposed into different frequency band components through Fourier transform. By observing the energy ratio of high-frequency and low-frequency components in the spectrum and the persistence index obtained from the autoregressive moving average model, it can be determined whether the rainfall process in a certain period is dominated by sudden high-intensity pulses or slow, continuous replenishment. The autoregressive moving average model is a process of predicting future trends by fitting historical data. The persistence index of the model is calculated by combining p-order autoregressive and q-order moving average parameters. When a sharp increase in cumulative amount is found in a short time window and the proportion of high-frequency energy is higher than a set threshold (e.g., 0.6), the corresponding precipitation process is classified as a short-term heavy precipitation pattern. When the cumulative amount rises slowly in a long time window, low-frequency components dominate, and the persistence parameter of the autoregressive moving average model is high, the corresponding precipitation process is classified as a long-term continuous precipitation pattern. This model analysis adds a "permeability risk category" label to each precipitation sequence, marking short-duration heavy precipitation as high permeability risk and long-duration continuous precipitation as low permeability risk. In terms of data structure, this is represented by adding a discrete categorical variable C to the original R1(t) to distinguish the differences in subsequent permeability behavior under different rainfall conditions.
[0055] Because subgrade settlement is more likely to occur during rapid infiltration caused by short-duration heavy rainfall, spatial analysis is further conducted using corrected soil moisture distribution data within the high-permeability risk category. Under the same rainfall event, sensor points deployed along the route are divided according to mileage coordinates and lateral positions. The moisture content time series of each monitoring point at each depth layer is aligned with the corresponding high-risk rainfall label. By calculating the lag time from the start of a rainfall event to a significant increase in moisture content at different depths, the time difference of the peak moisture content at different mileage positions, and the propagation path of the peak moisture content along the longitudinal and lateral directions, the infiltration path characteristics of moisture in the subgrade can be derived. These characteristics can be represented as a set of "depth-time-location" trajectories. For example, the arrival time T(z, x) of the moisture front at different depths, the duration L(z, x) of the local saturation zone, and the connectivity index E(x) of the high moisture content zone along the road direction are recorded, where z and x are two coordinate quantities based on the spatial location of the subgrade, x is the longitudinal position, and z is the depth at the soil unit. This allows us to extract dynamic features describing the direction and speed of water migration from the original static water content distribution data. These features are then summarized according to the roadbed block units to form a permeation path feature vector P1 for each sub-region.
[0056] By statistical analysis based on these vectors, certain depth ranges and planar locations can be identified as having more frequent and rapidly rising water-bearing trajectories under high-permeability-risk rainfall, thus defining them as "preferred permeability areas." In terms of data, this is manifested by assigning higher weight coefficients or risk markers to the corresponding depth layers and spatial grid units, so that each subgrade sub-region has multi-dimensional characteristics such as "cumulative precipitation distribution characteristics R1", "permeability risk category C", "permeability path characteristics P1", and "preferred permeability weight W".
[0057] Subsequently, when using data classification techniques to grade the roadbed area, these multidimensional features are used as input to divide the entire road corridor into multiple spatial units according to mileage, with each unit corresponding to a set of [R1, C, P1, W]. When selecting supervised classification algorithms such as k-means clustering or tree-based models as classifiers, risk level labels can be set as outputs. For example, units that have historically experienced settlement or have been found to have significant deformation are labeled as high-level, while units that have not experienced settlement are labeled as low-level. By training on samples containing features of cumulative precipitation patterns, risk categories, and infiltration paths, the classifier learns under which combinations of precipitation accumulation patterns and infiltration paths are more likely to result in settlement. In the prediction phase, the [R1, C, P1, W] features of each unit within the current observation period are used as input, and the classification model outputs the corresponding precipitation infiltration risk level, realizing the division of the roadbed area into multiple levels from low to high. The output is a dataset of "classification results of precipitation infiltration risk after grading".
[0058] Through the above processing, precipitation sequences are no longer used as a single intensity indicator, but are transformed into cumulative distribution characteristics across multiple time scales, capable of distinguishing between short-term impacts and long-term replenishment. The corrected soil moisture distribution data is also transformed into directional and connected features through infiltration path analysis. A one-to-one correspondence is established between the two types of data in the classification model, ensuring that each risk level corresponds to a clear rainfall pattern, moisture migration path, and combination of preferred infiltration areas. This structure helps address the problem of making rough judgments based solely on simple rainfall or single-point water content data. It visualizes the relationship between precipitation, infiltration, and potential settlement spatially to specific road sections and depth ranges through a data-driven approach, thus providing risk inputs with temporal and spatial resolution for subsequent infiltration depth estimation and saturation state simulation.
[0059] S3. Based on the classification results, extract the estimated seepage depth under the corresponding working conditions, and simulate the changing trend of soil saturation state based on the estimated seepage depth.
[0060] In one specific embodiment, the process of performing step S3 may specifically include the following steps:
[0061] Based on the classification results, the corrected precipitation sequence data corresponding to the high permeability risk category are extracted;
[0062] Based on the corrected precipitation sequence data, the estimated infiltration depth under the corresponding working conditions is calculated using time series analysis methods;
[0063] Based on the estimated seepage depth and combined with the road subgrade soil parameters, the simulation parameters for the soil saturation state are determined.
[0064] Based on the soil saturation state simulation parameters, the rate of change of soil saturation over time is analyzed at different time scales to obtain the saturation change characteristics used to characterize the change of soil saturation state.
[0065] The soil saturation state simulation parameters and saturation change characteristics are input into the finite element analysis model, and the distribution dynamics of water in the road subgrade soil are simulated through finite element analysis technology.
[0066] Based on the distribution dynamics, a pore pressure distribution map of the road subgrade soil is generated;
[0067] Based on the pore pressure distribution map, the changes in saturation state at different depths of the road subgrade are analyzed to determine the distribution characteristics of the soil saturation state change trend at each depth, and soil saturation state change trend data are generated.
[0068] Specifically, the classification results are precipitation infiltration risk classifications divided by time and space. For each roadbed segment and each rainfall event, precipitation sequences belonging to the high infiltration risk category can be identified based on the classification results. These sequences correspond to short-duration heavy rainfall or conditions with significant cumulative effects, which are more likely to trigger rapid infiltration and abrupt changes in deep water content. Data processing uses these high-infiltration-risk precipitation sequences as the entry point, combining the corrected precipitation time series with previously established multi-source correlations to estimate the infiltration depth range of different road segments under this condition. This infiltration depth is then used as a control condition to drive the numerical simulation of unsteady seepage and saturation states of the roadbed soil.
[0069] For a specific high-permeability risk event, the corrected precipitation sequence for a designated road segment provides rainfall intensity and multi-scale cumulative amounts at equal time steps. Combined with soil physical and mechanical parameters such as permeability coefficient, initial moisture content, and void ratio obtained from field or laboratory tests, and using empirical formulas or calibrated infiltration models, the rainfall process can be transformed into time-varying infiltration flux and changes in the infiltration front position. Based on this flux-front position change relationship, an estimated infiltration depth is obtained for each subgrade unit under a given working condition. This depth corresponds to the typical lower boundary of the moisture influence zone under that working condition. For example, short-duration heavy rainfall may cause the infiltration front to advance to the middle of the fill layer or near the top surface of the subgrade within a few hours, while continuous light rain only forms a shallow wetting zone on the surface. The estimated infiltration depth corresponds to the subgrade structural layers, soil layer boundaries, and sensor deployment depth. In the data structure, this estimated value is paired one-to-one with "road segment number, working condition number, and high permeability category label".
[0070] For example, the infiltration model essentially maps the "surface rainfall history q(r,t)" into a relationship between "time-varying infiltration flux q(f,t) and infiltration front depth h(f,t)". This relationship can use commonly used empirical or semi-empirical infiltration formulas in engineering, and the parameters are calibrated through indoor infiltration tests or field double-loop infiltration tests to match the actual soil properties of the road subgrade. In this scheme, the infiltration model can be implemented by combining a calibrated Green-Ampt type infiltration model with experimental corrections. Specifically, representative soil samples from the road subgrade are subjected to infiltration tests and soil-water characteristic curve tests in the laboratory to obtain the saturated permeability coefficient K. s The study included fundamental parameters such as the wetting front suction parameter ψ and the initial-to-saturated volumetric water content difference Δθ. In the field, water content was monitored through artificial rainfall experiments in small plots or under natural rainfall conditions, recording surface infiltration and water content changes at different depths at different times. The observed cumulative infiltration and wetting front advance depth curves were compared with the Green-Ampt theoretical solution, and K was analyzed using least squares methods. sThe parameters ψ and Δθ are iteratively adjusted to ensure that the infiltration flux curve calculated by the model matches the measured curve, and that the evolution trajectory of the infiltration front depth over time corresponds to the moments when the water content at each depth increases significantly. For roadbeds with obvious stratification, parameters can be calibrated separately for each soil layer. During numerical calculations, the parameter set is switched according to the moment when the wetting front crosses the layer interface, thus forming an infiltration model suitable for the roadbed structure. This model is used to convert any high-permeability-risk rainfall process into a time-varying infiltration flux boundary and infiltration front function.
[0071] After obtaining the estimated seepage depth, it is combined with the road subgrade soil parameters to form a set of simulation parameters for the soil saturation state. This parameter set typically includes the functional relationship between the permeability coefficient of each soil layer and the water content or saturation, soil-water characteristic curves, the conversion relationship between volumetric water content and saturation, the relationship between compression modulus and effective stress, etc., as well as the geometric boundary of the embankment, surface drainage conditions, and drainage or impermeable boundary conditions of the underlying layer. Based on the estimated seepage depth, the depth range that needs to be simulated in the vertical direction is determined, and this range is discretized into several computational units or grid nodes. Corresponding soil parameters are assigned to each unit to ensure that the simulation parameters are spatially consistent with the road structure layering and soil layering.
[0072] In the simulation process control, time is used as the independent variable. Following the discrete time steps of the rainfall process, the infiltration flux at the surface boundary is input into the seepage equation over time. Simulation parameters of the soil saturation state are substituted into the unsaturated seepage control equation. Finite element analysis is used to solve for the water content or saturation increment of each unit at a given time step, thus obtaining a family of curves representing "saturation change over time" at multiple time scales. Analysis of these curves allows for the extraction of characteristic quantities such as the rate of saturation change, peak saturation, time to reach the peak, and duration of saturation exceeding a certain threshold for each spatial unit. These characteristic quantities constitute the "saturation change features" used to characterize changes in the soil saturation state. In terms of data organization, each road segment, each depth layer, and each time period has a corresponding set of saturation change features, which correspond to the high permeability risk category and the estimated permeability depth, describing the temporal pattern of "how water accumulates and dissipates at that depth under this condition."
[0073] For example, the above seepage equation is for the mathematical description of saturated seepage based on Darcy's law and the conservation of mass. When the roadbed soil is in a near-saturated or fully saturated state and the pore water content does not change significantly with time, it can be expressed by a controlling equation with hydraulic potential as the variable: Where Q is the seepage velocity vector, K is the saturated permeability coefficient, H = h + z1 is the hydraulic potential, h is the pressure head, and z1 is the elevation, combined with the mass conservation equation. This yields the elliptic governing equations satisfying the spatial distribution of hydraulic potential in the saturated seepage field. In the finite element model, these equations are discretized and solved within the roadbed computational domain to describe the spatial distribution of pore water within the seepage depth range. The aforementioned unsaturated seepage governing equations address the condition where the soil is partially saturated and its water content changes significantly over time. They are described using Richards-type equations with matrix suction or pressure head as the main variables. In a one-dimensional vertical infiltration scenario, this can be written as: Where h is the matrix suction or pressure head, t is time, z2 is the vertical coordinate, C(h) is the water content function, reflecting the sensitivity of water content to pressure head changes, and K(h) is the unsaturated permeability coefficient function that varies with pressure head. By substituting the soil saturation simulation parameters into C(h) and K(h), and using the surface flux obtained from the infiltration model as the boundary condition, the above unsaturated seepage control equations are discretized and numerically solved using finite element methods at a given time step. This yields the water content or saturation increment of each grid cell at different times, thus forming a family of curves representing "saturation changing with time".
[0074] To better reflect the impact of saturation state changes on the roadbed's support capacity, effective stress and pore water pressure calculations are introduced simultaneously with the seepage solution. The pore water pressure of each element within each time step is calculated. During the numerical solution process, the unsaturated seepage control equations are discretized and solved using the finite element method in time-step order, yielding the water content or saturation field and seepage velocity field of each grid element at different times. The changes in water content and seepage velocity over time and space represent the dynamic distribution of water in the roadbed soil. Based on this dynamic distribution, the pore water pressure of each element is calculated according to the effective stress principle, combined with the soil saturation state simulation parameters. The pore water pressure results at different locations and depths at each moment are spatially reconstructed and visualized, generating a dataset of pore pressure distribution maps of the roadbed soil. Each grid point corresponds to a pore water pressure value representing a specific location, depth, and time. Analyzing the pore pressure variation sequence of the same depth layer throughout the entire rainfall and receding period can yield indicators such as the pore pressure rise rate, peak value of excess hydrostatic pressure, and duration. These indicators have a functional or statistical correlation with the aforementioned saturation variation characteristics, reflecting the effective stress weakening process caused by water infiltration.
[0075] Based on the above seepage and pore pressure analysis output, the saturation state change behavior of each depth layer is summarized from two dimensions: time and depth. The saturation and pore pressure sequences of adjacent time steps are regarded as a time curve, and trend features are extracted from the curve. For example, whether there are multiple peaks in saturation at a certain depth layer, the interval between peaks, and the proportion of the saturation and pore pressure rise segment in the entire observation period are all considered. These trend features are stored as "soil saturation state change trend data" according to depth layer and road segment location. In terms of data structure, this trend data is indexed by road segment number and depth layer number. Each record contains saturation change rate characteristics, pore pressure evolution characteristics, and high permeability risk condition indicators, so that subsequent soil support capacity analysis can select the corresponding trend data for calculation at different depth layers.
[0076] Through this series of processes, the corrected rainfall sequences corresponding to the high permeability risk category are transformed into physically meaningful estimates of permeability depth. Soil parameters, in conjunction with the permeability depth, define the simulation domain and boundary conditions. The results of water content and pore pressure variations over time and depth generated by the finite element method are then compressed into trend data that can be used for subsequent probability and bearing capacity analysis. This data chain, from "risk classification – permeability depth estimation – saturation state trend," allows the relationship between rainfall, infiltration, and pore pressure evolution to be reflected in the subgrade structure in a time-series and spatially layered manner. This solves the problem that static safety factors or single monitoring sections cannot reflect the dynamic process of gradual saturation and weakening of bearing capacity within the subgrade over time.
[0077] S4. Analyze the soil support capacity index based on the trend of soil saturation state changes, and determine the data on the decrease in soil support capacity.
[0078] In one specific embodiment, the process of performing step S4 may specifically include the following steps:
[0079] Based on the trend of soil saturation state change, the rate of saturation change of different depth layers of road subgrade is extracted, and the characteristics of soil particle spacing change are analyzed accordingly.
[0080] If the rate of change of saturation exceeds the preset rate threshold, the corresponding depth layer will be marked as a high-risk depth layer.
[0081] For high-risk depth layers, the increment of soil pore pressure is calculated based on the variation characteristics of particle spacing;
[0082] Numerical regression techniques were used to establish the correlation between the increase in soil pore pressure and soil support capacity indicators;
[0083] Based on the correlation, the decrease in soil support capacity of the roadbed at different depths is calculated, and preliminary decrease data is generated.
[0084] The initial decline data was verified over multiple time periods, and error correction was performed on the initial decline data based on the verification results to obtain the decline data.
[0085] Specifically, soil saturation state change trend data organized by road segment, depth, and time are obtained. For each subgrade unit and each depth layer, a saturation or moisture content curve changing over time and the corresponding pore pressure change sequence can be obtained. Subsequently, analysis is carried out around this type of trend data. By transforming "moisture state change" into "soil mechanical support capacity change", a data path from saturation evolution to the magnitude of support capacity decline is constructed.
[0086] For a specific road segment, it is divided into several calculation units along the longitudinal and transverse directions. Within each unit, it is further divided into multiple layers based on depth. The soil saturation state change trend data provides a time-discrete saturation sequence for each layer. By differencing or fitting these sequences, the rate of change of saturation with respect to time is calculated within a pre-defined time step. Combined with rainfall events, these rates are labeled under specific rainfall events or specific high-permeability risk conditions. In the data structure, a three-dimensional correspondence of "saturation change rate – time – condition" is established for each depth layer. Then, a rate threshold is determined based on experience or numerical analysis. When the saturation change rate of a certain depth layer exceeds the threshold within a certain time period, that depth layer is marked as a high-risk depth layer, indicating that the redistribution of moisture among soil particles is relatively drastic, and the pore structure and stress state have undergone significant adjustments within this range.
[0087] In high-risk depth layers, the volumetric variation characteristics of interparticle gaps can be deduced from the saturation variation trend. For a given soil layer, given its void ratio, consolidation state, and soil-water characteristic curves, by mapping the saturation increment to the pore water volume increment, and combining this with solid-phase skeleton constraints, the increase or decrease in the amount of water filling the effective pore space per unit volume of soil can be estimated, thus obtaining quantitative indicators of interparticle gap compression or expansion over a certain time period. These indicators are expressed in the data as gap changes under the "depth-time-condition" index, which can be used to reflect the degree of change in the relative positions of water and soil within the internal structure of a certain depth layer under the influence of high-permeability-risk rainfall.
[0088] Based on these interstitial variation characteristics, the effective stress principle is introduced, using the change in pore water pressure as an intermediate quantity for the change in soil supporting capacity. For each high-risk depth layer, using the saturation increment and interstitial variation as inputs, the changes are converted into pore pressure increments using existing empirical relationships in seepage-consolidation analysis or pre-established numerical calibration models. This conversion result is recorded as a data table of "pore pressure increment – depth – time – working condition," forming a pore pressure increment sequence corresponding one-to-one with the high-risk depth layer.
[0089] For example, the numerical calibration model can be established according to the idea of "using experimental / high-precision numerical results as the true value and simplified relationships as the model to be calibrated". For representative subgrade sections, different amplitudes of water content or saturation increments and void ratio changes are applied in indoor triaxial consolidated drained tests or consolidated undrained tests. A pore pressure measurement device is used to obtain the pore pressure increment sequence of the same depth unit under a given confining pressure and stress path. This sequence is then cross-checked with the high-precision seepage-consolidation finite element calculation results under the corresponding working conditions. Using "saturation increment, void ratio change, and initial stress state" as inputs and the measured or high-precision calculated pore pressure increment as output, a mapping model to be calibrated is constructed using linear or nonlinear functions. Least squares or regularized regression methods are used to fit the model parameters on multiple sets of experimental / calculation samples, making the pore pressure increment output by the model statistically close to the measured value. In subsequent engineering applications, the saturation increment and void ratio change at the corresponding time step are input for high-risk depth layers, and the pore pressure increment can be calculated using this calibration model.
[0090] After obtaining the pore pressure increment data, numerical regression techniques are needed to construct a correlation model to establish a link between this intermediate variable and the soil support capacity index. The soil support capacity index can be represented by the equivalent deformation modulus, the reduction factor of the foundation bearing capacity characteristic value, the reduction factor of shear strength, or a safety factor based on effective stress. For each depth layer and working condition, a set of "pore pressure increment – change in support capacity index" sample pairs are obtained through field static load tests, dynamic testing equipment, or inversion using existing numerical models. The independent variable in the sample is the pore pressure increment, and the dependent variable is the change in the support capacity index relative to the initial state. Based on these samples, multiple regression or nonlinear regression methods are used to fit the functional relationship, simplifying the complex, multi-factor soil response under this high-risk working condition into a response function with the pore pressure increment as the core driving variable. This is represented by a set of regression parameters for each soil layer type or each type of subgrade structure. Through this process, a correlation model of "pore pressure increment - change in support capacity index" is formed and bound to the depth layer identifier and soil type identifier, so that for any high-risk depth layer, the corresponding change in support capacity can be calculated as long as the pore pressure increment is given.
[0091] After calibrating the correlation, the pore pressure increment sequence obtained in the previous step is input into the correlation model hourly. The changes in soil support capacity indices are calculated for each depth layer and each time step, with particular attention paid to the relative decrease in bearing capacity or deformation modulus. By taking the envelope or maximum value in the time dimension, the peak value of support capacity decrease under specific rainfall conditions can be extracted for each depth layer. This peak value represents the degree of influence of the condition on the soil bearing capacity at that depth layer and is recorded in the data structure as a three-dimensional dataset of "depth – condition – support capacity decrease," which is the preliminary decrease data. At this stage, each depth segment of each subgrade unit corresponds to a decrease value, reflecting the degree of weakening of mechanical support capacity caused by saturation changes and pore pressure redistribution under high permeability risk conditions.
[0092] To ensure that this data on the rate of decrease aligns with actual engineering performance, multi-period validation is necessary. This multi-period validation phase selects historical periods encompassing multiple rainfall-recovery cycles. Within these periods, in addition to simulation results of saturation and pore pressure, data from field settlement monitoring, road surface smoothness changes, or structural overall stiffness changes measured by a falling weight deflectometer can be obtained. These observations are then converted into a macroscopic representation of the overall foundation support capacity change, and compared with the decrease in support capacity at each depth layer calculated using a correlation model. Inversion techniques or optimization methods are used to assess the explanatory power of the current decrease data for the macroscopic response. If, under multiple rainfall events and different seasonal backgrounds, the decrease in a certain soil layer or depth range is found to be systematically larger or smaller, it can be assumed that there is a deviation in the correlation model parameters or pore pressure estimation corresponding to that depth layer. Therefore, error correction can be performed on the initial decrease data by adjusting regression coefficients, introducing correction factors, or reclassifying high-risk thresholds.
[0093] Through the above verification and correction, a more stable correspondence is established between the drop amplitude data of each depth layer and the long-term monitoring of settlement and deformation at the same location, and the temporal information in this relationship is preserved, so that the drop amplitude can reflect both the instantaneous weakening of the support capacity of a single rainfall event and the cumulative effect of multiple events superimposed.
[0094] S5. Construct a settlement initiation probability calculation model based on the descent amplitude data to obtain the settlement initiation probability distribution of the road subgrade.
[0095] In one specific embodiment, the process of performing step S5 may specifically include the following steps:
[0096] Key parameters characterizing the decline in soil support capacity are extracted from the data on the magnitude of the decline, and a calculation model for the probability of settlement initiation of road subgrade is constructed based on the key parameters.
[0097] Historical settlement data is obtained and input into a settlement-induced probability calculation model. The correlation between the settlement amplitude data and the historical settlement data is analyzed using probabilistic simulation technology to obtain simulation results.
[0098] Based on the simulation results, a cumulative settlement impact factor is generated to characterize the long-term settlement risk of road subgrade.
[0099] The model parameters of the settlement initiation probability calculation model are corrected based on the settlement accumulation impact factor;
[0100] Based on the corrected settlement initiation probability calculation model, the settlement initiation probability distribution of road subgrade is determined, and the probability distribution characteristics of different regions are analyzed.
[0101] Specifically, the data on the rate of decrease is organized according to "road segment-depth-time". For each calculation unit and each depth layer, the decrease in soil support capacity relative to the initial state under a certain rainfall condition can be obtained. Using this data as input, the mechanical concept of "weakening of support capacity" is transformed into the statistical concept of "probability of settlement event occurrence", constructing a settlement initiation probability calculation model suitable for the road subgrade. In the data processing stage, the three-dimensional decrease data in a single rainfall event is reduced. A certain time window is used as the analysis unit, such as a typical rainfall process or a flood season. The decrease in support capacity of each depth layer and each spatial grid within the same window is combined according to preset weights to extract key parameter sets characterizing the overall weakening degree of the road segment. These parameters may include the maximum decrease, depth-weighted average decrease, decrease duration, number of high-risk depth layers, etc. Through this compression method, the complex "depth-time-space" data is transformed into a feature vector of finite length, so that each road segment corresponds to a "support capacity decrease feature" vector within each analysis window, and an index is established with the road segment number and time period number.
[0102] When constructing a settlement-induced probability calculation model, these characteristics of declining support capacity are treated as independent or covariates of the probability model, and historical settlement records are introduced as response variables. Historical settlement records can originate from long-term settlement observation points, structural health monitoring systems, or periodic road condition inspections. For each road segment and time window, the system records whether a settlement exceeding the limit event occurred within that window and the settlement level, converting this data into classification labels or interval variables. For example, a settlement exceeding a certain service limit can be considered a "settlement event," denoted as 1, otherwise 0. Alternatively, the settlement can be divided into several levels, forming multi-category labels. Thus, for each road segment-time window combination, both a "support capacity decline feature vector" and a "settlement response label" are simultaneously generated, forming the sample set required to train the settlement-induced probability model. Based on this sample set, a settlement-induced probability calculation model can be constructed using logistic regression, generalized linear models, proportional hazards models, or probabilistic neural networks, so that the model output is the conditional probability of a settlement event occurring under given support capacity decline characteristics.
[0103] During model training, historical settlement records are input into the model to constrain parameters, ensuring that the "settlement magnitude – settlement probability" relationship depicted by the model aligns with real-world engineering performance. Specifically, all road segments with historical records and time windows are divided into training and validation sets. The settlement magnitude feature vector is used as input, and settlement event labels are used as output. Model parameters are solved using methods such as maximum likelihood estimation, cross-entropy loss minimization, or Bayesian inference, aiming to make the model's predicted settlement probability as statistically consistent as possible with historical observations. This process can incorporate probabilistic simulation techniques to sample uncertainties in the settlement magnitude data and historical settlement records. For example, Monte Carlo methods can be used to introduce perturbations into the settlement magnitude features, and historical settlement measurements can be randomly sampled according to measurement error ranges. This yields a set of probabilistic model outputs from multiple simulations, allowing analysis of the settlement probability distribution corresponding to a certain state of reduced support capacity under different perturbation conditions. These simulation outputs constitute the "simulation results" dataset.
[0104] Based on the simulation results, a cumulative settlement impact factor can be constructed to characterize long-term settlement risk. This cumulative settlement impact factor does not solely rely on the instantaneous settlement probability within a single time window. Instead, it integrates the settlement probability, bearing capacity reduction characteristics, and existing settlement events for the same road segment across multiple time windows. For example, it integrates or discounts the settlement probability at the same location under multiple heavy rainfall events, forming a scalar or vector indicator reflecting long-term cumulative risk. This indicator reflects the repeated effects of rainfall-bearing capacity weakening-settlement, and also incorporates the differences between "multiple instances of no settlement" and "a few settlement events" in historical observations. For each road segment, the cumulative settlement impact factor is compared with the long-term observed settlement frequency or settlement statistics for that segment. When a systematic deviation occurs, it indicates a bias in the current model's characterization of the long-term risk for that segment. In this case, parameter correction is performed by adjusting model parameters, such as regression coefficients, threshold parameters, or baseline risk functions, to make the model's prediction of settlement frequency more consistent with historical statistics over a longer time scale.
[0105] After the model parameters are corrected using the cumulative settlement impact factor, the settlement initiation probability calculation model can provide the settlement initiation probability under the current working conditions in new time windows where there are no complete historical settlement records, using the settlement amplitude data. For real-time prediction tasks, after each new rainfall process, the settlement amplitude data corresponding to that process is processed to extract a new support capacity decline feature vector, which is input into the corrected probability model. The output is the probability value of settlement exceeding the limit event for each road segment within that time window, which can be refined according to different depth layers or different risk levels to form the "settlement initiation probability distribution of road subgrade". Mapping these probability distributions spatially according to mileage coordinates or road network topology reveals the probability distribution characteristics of different areas. For example, a section of embankment subgrade may have a significantly higher probability value than an adjacent excavated section after multiple rainstorms, or a section with insufficient soft soil foundation improvement may still show a high settlement initiation probability under light to moderate rainfall conditions.
[0106] In this way, the "decrease in support capacity" derived from mechanical analysis is transformed into a "settlement initiation probability distribution" that can be compared on a regional scale. This solves the problem that relying on a single safety factor or empirical coefficient is difficult to express the cumulative effect over time and spatial heterogeneity. At the same time, it makes historical settlement records no longer just passive ex-post statistics, but embeds them into the prediction model through probability simulation and parameter correction, providing statistically significant input for subsequent settlement interval prediction and stability assessment.
[0107] S6. Based on the probability distribution of settlement initiation and combined with the cumulative impact of precipitation, predict the settlement range of the roadbed.
[0108] In one specific embodiment, the process of performing step S6 may specifically include the following steps:
[0109] Based on the probability distribution of settlement, roadbed areas with a continuously increasing probability value over a continuous time period are identified and defined as high-risk evolution zones.
[0110] For high-risk evolution zones, the dynamic weights of different water infiltration pathways are analyzed based on the corresponding estimated infiltration depth and soil moisture distribution characteristics.
[0111] By integrating the dynamic weights of differential paths with the cumulative influencing factors of precipitation through time-series prediction techniques, a sequence of cumulative influencing factors characterizing the cumulative effect of subsidence is obtained.
[0112] Adjust the time scale of the roadbed settlement prediction interval based on the cumulative influencing factor sequence;
[0113] If the probability distribution of subsidence in high-risk evolution zones shows a continuous upward trend, the parameter adjustment feedback mechanism will be activated to update the dynamic weights of differential paths.
[0114] Based on the updated dynamic weights of the differential paths, and combined with geospatial information data, the spatial distribution of the roadbed's action mechanism is determined.
[0115] Based on the spatial distribution of the action mechanism and combined with the cumulative influencing factor sequence, the initial settlement prediction range of the road subgrade is generated;
[0116] The initial settlement prediction range is validated in multiple dimensions, and the validated settlement prediction range is output.
[0117] Specifically, the subsidence initiation probability distribution data is organized by time and space, with a corresponding subsidence initiation probability value for each road segment unit and each time step. Based on this, instead of focusing solely on the probability magnitude at a single moment, the focus shifts to the evolutionary trend of the probability over time. This probability evolution is combined with factors influencing cumulative precipitation and previously obtained infiltration and saturation information to generate a subsidence prediction interval with both temporal and spatial attributes. For a given road corridor, the subsidence initiation probability distribution can be viewed as a two-dimensional sequence with time and mileage as independent variables. For each spatial unit, a "probability-time" curve is formed within the observation period. A sliding window is used to calculate the slope of the probability within the window, the number of consecutive upward steps within the window, and local variance, among other indicators. These are used to identify areas where the probability value shows a continuous upward trend over a continuous time period. Spatially, these areas are marked as high-risk evolution zones, indicating that the subsidence risk at these locations has accumulated and increased over multiple adjacent time windows rather than experiencing sporadic fluctuations.
[0118] For high-risk evolution zones, it is necessary to incorporate the estimated infiltration depth and soil moisture distribution characteristics obtained in previous steps. For a specific high-risk evolution zone, the estimated infiltration depth under corresponding rainfall conditions, along with the moisture distribution and infiltration path characteristics at each depth layer, can be traced. Combined with the roadbed geometry and structural stratification, seepage paths are categorized according to different types, such as infiltration paths along slope surfaces, infiltration paths along pavement cracks, and seepage paths through drainage facilities. The frequency of occurrence, corresponding infiltration depth, and number of times high-saturation zones are triggered for each path within the area are statistically analyzed, thereby defining the dynamic weights of differential paths. The dynamic weights of differential paths essentially reflect the relative contributions of different moisture infiltration paths to the evolution of saturation and weakening of support capacity within the high-risk evolution zone. Data-wise, this is represented by a set of path type weights for each high-risk evolution zone. These weights have a computational relationship with the estimated infiltration depth and soil moisture distribution characteristics, and can be updated over time, forming a dynamic sequence of "path type – weight – time".
[0119] After obtaining the dynamic weights of differential paths, they are combined with precipitation accumulation influencing factors. These factors originate from the multi-scale precipitation accumulation characteristics and settlement accumulation influencing factors obtained in the previous multi-source data analysis, providing a comprehensive intensity index of "rainfall-infiltration-historical settlement performance" for each time window. Through time-series forecasting techniques, the dynamic weight sequence of differential paths within the high-risk evolution zone is integrated with the corresponding precipitation accumulation influencing factor sequence to construct a time series characterizing the settlement accumulation effect. This series provides a numerical value for "the degree of settlement accumulation within the high-risk evolution zone in the current and future time steps" at each time window. Time-series forecasting techniques can employ autoregressive moving average models, state-space models, or recurrent neural networks. Using precipitation accumulation influencing factors and path weights from the past several time steps as input, the cumulative effect index for the future several time steps is predicted, resulting in a cumulative influencing factor sequence extending into the future. The shape of this sequence on the time axis determines the length and location of the subsequent settlement prediction interval.
[0120] After obtaining the cumulative influencing factor sequence, the time scale of the settlement prediction interval is adjusted according to the evolution of the cumulative effect in different regions. For a high-risk evolution zone, if the cumulative influencing factor sequence rises sharply in a short period of time and reaches a high value range within a short time window, the settlement prediction interval can be set to a relatively concentrated short-term segment to reflect the high risk of settlement triggering in the short term. If the cumulative influencing factors accumulate slowly over a longer period of time and form a high plateau, the prediction interval can be extended to a longer time scale to reflect the settlement risk under long-term cumulative effects. In this way, the start and end times of the settlement prediction interval are no longer preset fixed lengths, but are controlled by the temporal shape of the cumulative influencing factors. Through a mapping relationship, the "curve of cumulative effect changing with time" is transformed into "one or more intervals on the time axis".
[0121] After identifying high-risk evolution zones and establishing preliminary prediction intervals, a parameter adjustment feedback mechanism is needed to continuously correct the dynamic weights of differential paths. When it is observed that the probability distribution of settlement initiation in a certain high-risk evolution zone continuously increases over multiple update cycles and deviates from the risk level predicted by cumulative influencing factors, it can be considered that the current path weights are insufficiently or excessively reflect certain actual infiltration processes. By introducing a feedback mechanism, prediction errors and observation errors are used as adjustment signals to correct the weights of each path type. For example, when monitoring data shows that the contribution of slope infiltration paths to settlement development is greater than the model prediction within a certain time period, the dynamic weight of that path can be appropriately increased and the weights of other paths reduced in subsequent calculations, making the updated weight system closer to the actual mechanism of action. At the data level, the feedback mechanism continuously updates the "path type-weight-time" sequence with the difference between the new round of settlement initiation probability changes and the actual monitoring results, enabling the path weights of high-risk evolution zones to have adaptive adjustment capabilities.
[0122] After completing the temporal prediction and weight adjustment, geospatial information data is introduced to determine the spatial distribution of the action mechanisms of the road subgrade. Geospatial information includes road horizontal and vertical profiles, topography, groundwater distribution, geological stratification, and the distribution of adjacent structures. For each spatial unit, a "mechanism intensity" index can be constructed by comprehensively considering the proportion of time it is in a high-risk evolution zone, the corresponding dynamic weights of differential paths, the range of infiltration depth, and geological conditions. This index reflects the combined effect of rainfall-infiltration-support capacity weakening at that location. By rasterizing or vectorizing this index within the road corridor, the spatial distribution of action mechanisms can be obtained, representing a spatial map describing the degree of influence of multiple action mechanisms on different subgrade units, providing a basis for the spatial projection of settlement prediction intervals.
[0123] Given the spatial distribution of the action mechanism and the sequence of cumulative influencing factors, an initial settlement prediction interval for road subgrade can be generated. For any spatial unit, the spatial distribution of the action mechanism provides the strength index of the unit under the combined effects of seepage path, soil conditions, and structural form over a long period. The sequence of cumulative influencing factors provides the strength of the settlement driving force of the unit under different rainfall histories at each time window. Based on historical data, a sample set is established for multiple spatial units within multiple time windows, including the "intensity of the action mechanism," the "statistics of the cumulative influencing factors corresponding to the time window (such as the peak value, average value, or integral within the window)," and the measured or inverted settlement within the time window. The intensity of the action mechanism and the statistics of the cumulative influencing factors are used as independent variables, and the corresponding settlement or settlement level is used as the dependent variable. A regression model is used to fit the functional relationship between the two, and the expected value and fluctuation range of the settlement are simultaneously output in the model, thus obtaining a prediction model that can output the "settlement interval" and the "time window corresponding to the interval." During prediction, for each spatial unit, the cumulative influencing factor statistics for each window are calculated according to a sliding time window over the future prediction period. These statistics, along with the fixed intensity of the influencing mechanism for that unit, are input into the regression model to obtain the settlement range for each time window. When the upper limit of the settlement range for a certain time window exceeds the set serviceability threshold or structural safety threshold, that time window is marked as a high-risk settlement range for that unit. By unifying and rearranging all marked time windows along the time axis according to the spatial unit dimension, a set of strip-shaped or block-shaped high-risk ranges can be formed on the time-space plane. This set constitutes the initial settlement prediction range dataset for the road subgrade.
[0124] To avoid deviations between predicted results and engineering realities, the initial settlement prediction range requires multi-dimensional verification. Multi-dimensional verification involves cross-referencing information from multiple sources, including comparing the predicted range with historical settlement events not used in model training to check if it covers known settlement times and locations; combining recent monitoring of minor settlement amounts, road surface unevenness, or structural deflection changes to check if actual deformation within the predicted range exhibits a cumulative trend; and overlaying road maintenance records and geological disaster records to statistically analyze the match between the predicted range and actual treatment locations. Through these verification steps, if the predicted range is found to be systematically too wide or too narrow in certain areas, it can be further corrected by adjusting the threshold of cumulative influencing factors or modifying the weights of the spatial distribution of action mechanisms, outputting a settlement prediction range that remains consistent with monitoring data and historical records after multiple rounds of verification.
[0125] S7. Integrate the settlement prediction range with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability assessment results.
[0126] In one specific embodiment, the process of performing step S7 may specifically include the following steps:
[0127] Extract subgrade stability assessment indicators for road subgrade stability analysis from the settlement prediction range;
[0128] Based on the roadbed stability assessment index, the heterogeneity characteristics of the settlement prediction interval are analyzed;
[0129] Heterogeneity correction technique is used to correct the prediction deviation of the settlement prediction range, and the corrected settlement prediction range is obtained.
[0130] Construct a spatial distribution weight model for roadbed based on the spatial distribution of action mechanisms;
[0131] By using geographic information system overlay analysis technology, the corrected settlement prediction interval is fused with the actual road subgrade structure data according to the weights determined by the spatial distribution weight model to generate a stability distribution map of the road subgrade.
[0132] Based on the stability distribution map, the micro-settlement risk area of a specific road section is analyzed, and the spatial distribution characteristics of the micro-settlement risk area are determined.
[0133] Generate subgrade stability assessment results that include settlement prediction results for specific road sections, distribution characteristics of minor settlement risks, and stability assessment indicators.
[0134] Specifically, for a given road, the roadbed is divided into several spatial units. Each unit corresponds to one or more settlement prediction time intervals within the prediction period. These intervals may also include information such as the settlement initiation probability, predicted settlement range, and cumulative influencing factor levels for that unit. Through statistical analysis of multiple prediction intervals within the same unit, quantitative indicators such as interval length, start and end times, average probability level within the interval, peak probability, upper and lower limits of predicted settlement, and the proportion of high-probability periods within the interval can be extracted. These indicators are summarized into roadbed stability assessment indicators to characterize the intensity and persistence of settlement risk borne by the unit throughout the prediction period. Significant differences exist between different units in terms of geological conditions, structural form, and load levels. Within the same road, some sections are high-fill roadbeds, some are shallow-cut roadbeds, and some are bridge abutment sections. These differences lead to different engineering significance for the same settlement prediction interval in different units; therefore, it is necessary to analyze the "heterogeneity characteristics" of the settlement prediction intervals.
[0135] In the heterogeneity analysis process, using roadbed stability assessment indicators as input, roads are divided into several categories based on soil type, structural form, and topographic conditions. The distribution of indicators within and between each category is compared to identify which road sections exhibit overly wide, narrow, or overly high / low probability levels in their prediction intervals. Simultaneously, combining existing measured settlement data, pavement smoothness test results, and maintenance records, the frequency and magnitude of actual settlement in a particular road section category are statistically analyzed historically. These engineering performances are compared with the prediction interval indicators of the corresponding categories to identify the patterns of prediction deviations varying with region, structural type, and geological conditions. Based on this difference identification, heterogeneity correction techniques are introduced to correct deviations in settlement prediction intervals by region and structural type. For example, different correction coefficients are introduced for sections with different fill heights, and the interval length or probability level is adjusted for soft soil foundation sections and rigid base sections, respectively. This results in corrected settlement prediction intervals, making the same probability level have a more consistent engineering meaning in different regions.
[0136] After correcting for heterogeneity in the temporal dimension, the prediction interval and road subgrade structure data need to be spatially weighted and fused. To reflect the importance of each spatial unit in the overall action mechanism, a spatial distribution weight model of the road subgrade is constructed based on the previously obtained spatial distribution of action mechanisms. This weight model links the action mechanism intensity of each spatial unit with its structural characteristics. For example, units with large infiltration depth, drastic saturation changes, large decrease in support capacity, and located in high embankments or soft soil foundations are given higher weights, while units with large structural stiffness, good drainage conditions, and few historical settlement records are given lower weights. This forms a weight field that reflects "the degree of contribution of the unit to the overall settlement risk". Preferably, the spatial distribution weight model of the roadbed can be understood as a weight matrix arranged according to spatial units. The route is discretized into several spatial units along the mileage direction and (optionally) the lateral and depth directions, such as by "station number × lateral zone × depth zone". For each spatial unit, a weight value is calculated based on factors such as "permeability depth, saturation change, decrease in support capacity, foundation type, and cut / fill height", which represents the degree of contribution of this unit to the overall settlement risk or stability assessment. These weights are arranged according to spatial index to obtain a weight matrix or weight tensor, i.e., the spatial distribution weight model.
[0137] In a Geographic Information System (GIS) environment, the corrected settlement prediction interval data is mapped onto the road's horizontal location and mileage coordinates. Simultaneously, road subgrade structural data, including subgrade type, cut and fill height, foundation treatment method, structural layer thickness, and material parameters, are imported, along with a spatial distribution weighting model. For each spatial unit, its settlement prediction interval is overlaid with structural parameters on the prediction time axis. The weighting model determines the unit's weight coefficient in the fusion calculation. Settlement prediction information from different time periods and different units is weighted and summarized according to these weights to calculate an index value reflecting the unit's overall stability, such as stability level, remaining safety reserve coefficient, or comprehensive risk score. These indicators are then rendered spatially in a raster or linear format to generate a road subgrade stability distribution map. The map visually displays stability differences at different mileage locations along the route and spatially concentrated areas of potential weak points.
[0138] After obtaining the stability distribution map, minor settlement risk analysis can be conducted for specific road sections. For example, for road sections with high traffic volume, remote locations, high structural importance, or those that have historically experienced minor uneven settlement, spatial units corresponding to these sections can be extracted from the stability distribution map. The relative levels of their stability indices within adjacent units can be analyzed. Combined with the upper limit and probability level of predicted settlement within the corrected settlement prediction interval, areas that, although not reaching the large settlement risk threshold, have been in a state of low-intensity settlement accumulation for a long time can be identified. These areas are defined as minor settlement risk areas. By analyzing the spatial continuity, length, and correspondence with structural details (such as pavement joints, bridge abutment overlaps, and culvert sides) of these areas, the spatial distribution characteristics of minor settlement risk areas can be determined, providing a basis for refining monitoring point layout and preventive maintenance.
[0139] Based on the above, a comprehensive subgrade stability assessment result is output for a specific road segment. This assessment result includes settlement prediction interval information in the time dimension, providing the time window and corresponding settlement range for the possible settlement response of the road segment within the prediction period; it includes the distribution characteristics of minor settlement risks in the spatial dimension, indicating which locations in the lateral and longitudinal directions of the road segment are more prone to the accumulation of minor deformations; and it also includes stability assessment indicators from the stability distribution map, such as the overall stability level, redundancy of support capacity reduction, and risk level for the road segment. Through this data processing logic of "settlement prediction interval – structural parameters – spatial weights – stability distribution map – assessment result for a specific road segment," a clear correspondence is established between the prediction interval information in the time domain and the structural heterogeneity and intensity of the action mechanism in the spatial domain. This enables the road subgrade settlement prediction results to serve subgrade stability assessment and engineering management in a quantitative, segmented, and visualized manner.
[0140] The road subgrade settlement prediction method based on multi-source data in the embodiments of this application has been described above. The road subgrade settlement prediction system based on multi-source data in the embodiments of this application is described below. Please refer to [link / reference]. Figure 2 The present application provides a schematic diagram of the structure of a roadbed settlement prediction system based on multi-source data. The system includes:
[0141] The data acquisition module 10 is used to acquire precipitation sequence and soil moisture distribution data of roadbed within a specified area through multi-source data acquisition, correct the acquired data, and obtain the corrected precipitation sequence and soil moisture distribution data.
[0142] The risk classification module 20 is used to analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and to classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data, and obtain the classification results.
[0143] The permeability simulation module 30 is used to extract the estimated permeability depth under the corresponding working conditions based on the classification results, and to simulate the changing trend of soil saturation state based on the estimated permeability depth.
[0144] The capacity analysis module 40 is used to analyze the soil support capacity index based on the changing trend of soil saturation state, and to determine the decrease in soil support capacity.
[0145] The probability calculation module 50 is used to construct a settlement-induced probability calculation model based on the descent amplitude data and obtain the settlement-induced probability distribution of the road subgrade.
[0146] The settlement prediction module 60 is used to predict the settlement range of the roadbed based on the settlement initiation probability distribution and the cumulative impact of precipitation.
[0147] The fusion assessment module 70 is used to fuse the settlement prediction range with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability assessment results.
[0148] The above-described 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 roadbed settlement based on multi-source data, characterized in that, The method includes: S1. Acquire precipitation sequence and soil moisture distribution data of roadbed within a specified area through multi-source data acquisition, correct the acquired data, and obtain the corrected precipitation sequence and soil moisture distribution data. S2. Analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data to obtain the classification results. S3. Based on the classification results, extract the estimated seepage depth under the corresponding working conditions, and simulate the change trend of soil saturation state based on the estimated seepage depth. S4. Analyze the soil support capacity index based on the trend of soil saturation state change, and determine the decrease in soil support capacity. S5. Based on the aforementioned descent data, construct a settlement initiation probability calculation model to obtain the settlement initiation probability distribution of the roadbed; S6. Based on the settlement initiation probability distribution, combined with the cumulative precipitation influencing factors, predict the settlement prediction interval of the roadbed. The cumulative precipitation influencing factors are comprehensive intensity indicators used to characterize each time window, which are determined by the cumulative precipitation distribution characteristics and the cumulative settlement influencing factor. S7. The settlement prediction interval is fused with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability assessment results. S5 includes: Key parameters characterizing the decline in soil support capacity are extracted from the decline data, and a settlement initiation probability calculation model for road subgrade is constructed based on the key parameters. Historical settlement record data is acquired, and the uncertainty in the settlement magnitude data and the historical settlement record data is sampled using probabilistic simulation technology to obtain simulation results; Based on the simulation results, a cumulative settlement impact factor is generated to characterize the long-term settlement risk of road subgrade. The model parameters of the settlement initiation probability calculation model are corrected based on the settlement accumulation impact factor. Based on the corrected settlement initiation probability calculation model, the settlement initiation probability distribution of road subgrade is determined, and the probability distribution characteristics of different regions are analyzed. S6 includes: Based on the settlement-induced probability distribution, roadbed areas with a continuously increasing probability value over a continuous time period are identified and defined as high-risk evolution zones. For the high-risk evolution zone, based on the corresponding estimated infiltration depth and soil moisture distribution characteristics, the dynamic weights of different water infiltration paths are analyzed. By integrating the dynamic weights of the differential paths with the cumulative influencing factors of precipitation using time-series prediction technology, a sequence of cumulative influencing factors characterizing the cumulative effect of subsidence is obtained. Adjust the time scale of the roadbed settlement prediction interval according to the cumulative influencing factor sequence; If the subsidence-induced probability distribution of the high-risk evolution zone shows a continuous upward trend, then the parameter adjustment feedback mechanism is activated to update the dynamic weights of the differential paths; Based on the updated dynamic weights of the differential paths, the spatial distribution of the action mechanism of the roadbed is determined in conjunction with geospatial information data. Based on the spatial distribution of the aforementioned mechanism of action and in conjunction with the cumulative influencing factor sequence, the initial settlement prediction range for the road subgrade is generated. The initial settlement prediction interval is validated in multiple dimensions, and the validated settlement prediction interval is output.
2. The method according to claim 1, characterized in that, S1 includes: Continuous precipitation sequences within a designated area are obtained through satellite remote sensing data, and soil moisture distribution data at roadbed locations are collected through a ground sensor network deployed on the roadbed. Based on the soil moisture distribution data, sampling was conducted at different depths of the roadbed at multiple time periods to analyze the moisture retention characteristics at each depth, forming multi-time period variation data to characterize the changes in moisture retention time and retention rate. The continuous precipitation sequence and the multi-period variation data are processed using time series analysis methods to generate basic characteristic data of the cumulative variation of precipitation at different time scales; Based on the aforementioned basic feature data, a correlation model is constructed between precipitation and the moisture distribution of soil at various depths of the roadbed. Based on the correlation model, the deviations in the multi-source data acquisition are corrected, and the corrected precipitation sequence and corrected soil moisture distribution data are output.
3. The method according to claim 1, characterized in that, S2 include: Based on the corrected precipitation sequence, the cumulative distribution characteristics of precipitation at different time scales are extracted; Based on the aforementioned cumulative precipitation distribution characteristics, short-duration heavy precipitation patterns and long-duration continuous precipitation patterns are analyzed. If the cumulative precipitation distribution characteristics show a short-term heavy precipitation pattern, the corresponding precipitation sequence is marked as a high infiltration risk category; if the cumulative precipitation distribution characteristics show a long-term continuous precipitation pattern, the corresponding precipitation sequence is marked as a low infiltration risk category. For the high permeability risk category, water infiltration path features are extracted by combining the corrected soil moisture distribution data; Based on the characteristics of the water infiltration path, the preferential infiltration areas of water distribution in the depth layer are determined; Using data classification technology, the cumulative distribution characteristics of precipitation, infiltration risk categories, and information on priority infiltration areas are integrated to classify the corrected precipitation sequence into multiple precipitation infiltration risk levels, generating classification results of precipitation infiltration risk after classification.
4. The method according to claim 3, characterized in that, S3 include: Based on the classification results, the corrected precipitation sequence data corresponding to the high permeability risk category are extracted; Based on the corrected precipitation sequence data, the estimated infiltration depth under the corresponding working conditions is calculated using time series analysis methods. Based on the estimated seepage depth and combined with the road subgrade soil parameters, the simulation parameters for the soil saturation state are determined. Based on the soil saturation state simulation parameters, the rate of change of soil saturation over time is analyzed at different time scales to obtain saturation change characteristics that characterize the change of soil saturation state. The soil saturation state simulation parameters and the saturation change characteristics are input into the finite element analysis model, and the distribution dynamics of water in the road subgrade soil are simulated by finite element analysis technology. Based on the aforementioned distribution dynamics, a pore pressure distribution map of the road subgrade soil is generated; Based on the pore pressure distribution map, the saturation state changes of each depth layer of the road subgrade are analyzed, the distribution characteristics of the soil saturation state change trend in each depth layer are determined, and soil saturation state change trend data are generated.
5. The method according to claim 1, characterized in that, S4 include: Based on the trend of soil saturation state change, the saturation change rate of different depth layers of road subgrade is extracted, and the characteristics of soil particle spacing change are analyzed accordingly. If the rate of change of saturation exceeds a preset rate threshold, the corresponding depth layer will be marked as a high-risk depth layer. For the high-risk depth layer, the soil pore pressure increment is calculated based on the particle spacing variation characteristics; Numerical regression techniques were used to establish the correlation between the soil pore pressure increment and the soil support capacity index. Based on the aforementioned correlation, the decrease in soil support capacity of the roadbed at different depths is calculated, and preliminary decrease data is generated. The initial decline data is verified over multiple time periods, and error correction is performed on the initial decline data based on the verification results to obtain the decline data.
6. The method according to claim 1, characterized in that, S7 includes: Extract subgrade stability assessment indicators for road subgrade stability analysis from the settlement prediction interval; Based on the aforementioned roadbed stability assessment indices, the heterogeneity characteristics of the settlement prediction interval are analyzed. The prediction deviation of the settlement prediction range is corrected by using heterogeneity correction technology to obtain the corrected settlement prediction range. A spatial distribution weight model of the roadbed is constructed based on the spatial distribution of the described mechanism of action. By using geographic information system overlay analysis technology, the corrected settlement prediction interval is fused with the actual road subgrade structure data according to the weights determined by the spatial distribution weight model to generate a stability distribution map of the road subgrade. Based on the stability distribution map, analyze the micro-settlement risk areas of specific road sections and determine the spatial distribution characteristics of the micro-settlement risk areas; Generate subgrade stability assessment results that include settlement prediction results for specific road sections, distribution characteristics of minor settlement risks, and stability assessment indicators.
7. A roadbed settlement prediction system based on multi-source data, used to implement the method as described in any one of claims 1 to 6, characterized in that, The system includes: The data acquisition module is used to acquire precipitation sequences and soil moisture distribution data of the roadbed within a specified area through multi-source data acquisition, correct the acquired data, and obtain corrected precipitation sequences and soil moisture distribution data. The risk classification module is used to analyze the cumulative distribution characteristics of precipitation based on the corrected precipitation sequence, and to classify the precipitation infiltration risk of the roadbed area in combination with the corrected soil moisture distribution data, and obtain the classification results. The permeability simulation module is used to extract the estimated permeability depth under the corresponding working conditions based on the classification results, and to simulate the changing trend of soil saturation state based on the estimated permeability depth. The capacity analysis module is used to analyze the soil support capacity index based on the changing trend of the soil saturation state, and to determine the decrease in soil support capacity. The probability calculation module is used to construct a settlement-induced probability calculation model based on the decrease amplitude data, and obtain the settlement-induced probability distribution of the roadbed. The settlement prediction module is used to predict the settlement prediction range of the roadbed based on the settlement initiation probability distribution and the cumulative impact factors of precipitation. The fusion evaluation module is used to fuse the settlement prediction range with the actual road subgrade structure data to obtain the settlement prediction results of the road subgrade and the corresponding subgrade stability evaluation results.