A soft soil area roadbed settlement prediction system and method

Abstract

The present application relates to the field of geotechnical engineering, in particular to a soft soil area roadbed settlement prediction system and method, which is applied to the soft soil roadbed settlement monitoring and prediction of traffic infrastructure such as highways and railways; the present application covers five modules of data collection, preprocessing, model construction, prediction analysis and feedback optimization; the data collection module is responsible for collecting roadbed parameters and settlement history data; the preprocessing module carries out outlier elimination and normalization processing; the model construction module establishes a prediction model based on the gradient boosting decision tree algorithm, and the prediction analysis module uses the processed data and the model to carry out settlement prediction; the feedback optimization module compares the prediction results with the measured data and optimizes the model; compared with the traditional method, the prediction accuracy is significantly improved, and the average error is reduced by more than 40%; in addition, the system has strong environmental adaptability through multi-model parallel prediction and AI-driven parameter self-adaptive correction, and can automatically adjust the prediction strategy to adapt to different geological conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geotechnical engineering, in particular to a soft soil area roadbed settlement prediction system and method, which is applied to the soft soil area roadbed settlement monitoring and prediction of highway, railway and other transportation infrastructure. BACKGROUND

[0002] The soft soil area has the characteristics of high compressibility, low bearing capacity and deformation sensitivity, which makes the roadbed built in such areas often face serious settlement problems. Accurate prediction of soft soil area roadbed settlement is of great significance to engineering design, construction control and later maintenance.

[0003] The traditional roadbed settlement prediction method mainly relies on empirical formula or simple numerical simulation, which usually has the following problems: first, the traditional method is difficult to effectively deal with the complex geological conditions in soft soil area, resulting in large deviation between the prediction results and the actual situation; second, the existing prediction model lacks adaptability and cannot dynamically adjust the prediction parameters according to real-time monitoring data; third, the traditional method lacks comprehensive consideration of the influence of multi-parameter coupling, and it is difficult to accurately reflect the complex mechanism of roadbed settlement; finally, it lacks effective feedback adjustment mechanism and cannot optimize the prediction model according to the measured data.

[0004] With the development of artificial intelligence technology, it is possible to apply advanced machine learning algorithms to the field of roadbed settlement prediction. However, there is still a lack of special prediction system for the special characteristics of soft soil area, as well as a comprehensive solution that combines physical models and data-driven models. In view of the above problems, it is urgent to develop a high-precision prediction system and method specifically for soft soil area roadbed settlement. SUMMARY

[0005] The purpose of the present application is to provide a soft soil area roadbed settlement prediction system and method, which aims to solve the problems of insufficient prediction accuracy, poor model adaptability, lack of real-time feedback optimization mechanism and other problems in the prior art.

[0006] The present application establishes a complete technical system of data acquisition, preprocessing, model construction, prediction analysis and feedback optimization, adopts gradient boosting decision tree algorithm combined with physical model fusion technology, and realizes high-precision prediction of soft soil area roadbed settlement. The system has the key technologies of multi-dimensional data acquisition, intelligent data processing, self-adaptive model construction, multi-model parallel prediction and parameter correction based on artificial intelligence, forming a closed-loop optimization architecture of "collection-processing-modeling-prediction-feedback".

[0007] The present application proposes a soft soil area roadbed settlement prediction system, which comprises:

[0008] A data acquisition module for acquiring soft soil area roadbed parameters and settlement history data;

[0009] The data preprocessing module is in communication connection with the data acquisition module and is configured to perform outlier elimination and normalization processing on the acquired data.

[0010] The model construction module is in communication connection with the data preprocessing module and is configured to construct a subgrade settlement prediction model based on a gradient boosting decision tree algorithm.

[0011] The prediction analysis module is in communication connection with the model construction module and is configured to perform settlement prediction according to the processed data and the constructed model.

[0012] The feedback optimization module is in communication connection with the prediction analysis module and is configured to compare the prediction result with the measured data and optimize the prediction model.

[0013] Preferably, the data acquisition module comprises:

[0014] The subgrade parameter acquisition unit is configured to acquire the filling height, soil type and design type of the subgrade in the soft soil area.

[0015] The geological exploration data acquisition unit is configured to acquire the soil layer depth and soil layer longitudinal section diagram in the soft soil area.

[0016] The settlement historical data acquisition unit is configured to acquire the measured values of the subgrade settlement at different time nodes.

[0017] The real-time monitoring unit is configured to perform real-time monitoring on the subgrade settlement and acquire the current settlement data.

[0018] Preferably, the data preprocessing module comprises:

[0019] The outlier processing unit is configured to identify and eliminate outliers, missing values and repeated values in the data.

[0020] The data conversion unit is configured to convert the preprocessed data into an ordered list form.

[0021] The normalization processing unit is configured to map the data to the interval [0, 1] by using the maximum and minimum normalization method.

[0022] The correlation analysis unit is configured to analyze the linear correlation coefficient between the data, and when the correlation coefficient is greater than a preset threshold, a relationship between the parameters is established.

[0023] Preferably, the model construction module comprises:

[0024] The feature extraction unit is configured to extract the input features required by the prediction model from the preprocessed data.

[0025] The GBDT algorithm unit is configured to train weak classifiers and construct a gradient boosting decision tree model by using an iterative method.

[0026] a tree structure optimization unit configured to control the model complexity by a tree structure complexity parameter γ and a leaf node penalty parameter λ;

[0027] a physical model fusion unit configured to fuse the physical test results and the numerical simulation results to enhance the physical meaning of the model.

[0028] Preferably, the GBDT algorithm unit is specifically configured to:

[0029] initialize the model parameters, and set t = 1;

[0030] train a weak classifier for each sample point i wherein the weak classifier determines the optimal split point by calculating the average loss of each leaf region sample;

[0031] calculate the total error and update the tree structure complexity parameter γ and the leaf node penalty parameter λ;

[0032] optimize the model through an iterative process extended by Taylor formula;

[0033] determine whether a termination condition is met, and output the final model H(x) if the termination condition is met.

[0034] Preferably, the prediction analysis module comprises:

[0035] an input processing unit configured to convert the parameters of the roadbed to be predicted into an input format required by the model;

[0036] a multi-model prediction unit configured to obtain multiple sets of prediction results through parallel calculation of multiple prediction models;

[0037] a result evaluation unit configured to evaluate the errors of the prediction results and select the result with the smallest error as the final prediction value;

[0038] a settlement curve generation unit configured to generate a relationship curve of the settlement of the roadbed over time according to the final prediction value.

[0039] Preferably, the feedback optimization module comprises:

[0040] an error calculation unit configured to calculate the error between the prediction value and the measured value;

[0041] an AI parameter correction unit configured to automatically correct the parameters based on an artificial intelligence algorithm when the error is greater than a preset threshold;

[0042] a historical data correction unit configured to input the parameters based on an error correction model of historical prediction data and measured data;

[0043] a model updating unit configured to optimize and update the prediction model according to the correction results.

[0044] As preferred, the AI parameter correction unit is specifically used for:

[0045] establishing a mapping relationship between the prediction parameters and the errors;

[0046] identifying key parameters that have the greatest impact on the prediction results;

[0047] dynamically adjusting the weights of the key parameters according to real-time settlement data;

[0048] feeding the corrected parameters to the model construction module for model optimization.

[0049] As preferred, it further comprises a modeling environment module, which is in communication connection with the model construction module, and is used for:

[0050] storing multiple sets of modeling environment information, including modeling environment data, modeling environment analysis data, and modeling environment determination data;

[0051] selecting the optimal modeling environment through a similarity comparison analysis method;

[0052] The similarity comparison analysis method comprises calculating the similarity coefficient and correlation of the modeling environment data information and the modeling environment information, and selecting the optimal solution based on the weighted results.

[0053] A soft soil area roadbed settlement prediction method, comprising the following steps:

[0054] S1, collecting soft soil area roadbed parameters and settlement history data;

[0055] S2, performing outlier rejection and normalization processing on the collected data;

[0056] S3, constructing a roadbed settlement prediction model based on a gradient boosting decision tree algorithm;

[0057] S4, performing settlement prediction according to the processed data and the constructed model;

[0058] S5, comparing the prediction results with the measured data;

[0059] S6, when the prediction error is greater than a preset threshold, correcting the model parameters based on an artificial intelligence algorithm;

[0060] S7, updating the prediction model according to the correction results;

[0061] S8, using the updated model to re-perform settlement prediction until the prediction error is less than the preset threshold;

[0062] S9, outputting the final roadbed settlement prediction results, including the settlement amount and the relationship curve of the settlement changing with time.

[0063] The application has the beneficial effects that: firstly, compared with the traditional prediction method, the prediction accuracy is significantly improved, and the average prediction error is reduced by more than 40%; secondly, through multi-model parallel prediction and AI-driven parameter self-adaptive correction, the system has strong environmental adaptability and can automatically adjust the prediction strategy for different geological conditions; thirdly, a closed-loop feedback mechanism based on real-time monitoring data is established, which can continuously optimize the prediction model and improve the reliability of long-term prediction; finally, the special algorithm and processing flow designed for the particularity of soft soil area significantly enhance the application effect in complex geological conditions. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a whole framework diagram of the soft soil area roadbed settlement prediction system of the application;

[0065] Figure 2 It is a structural schematic diagram of the data acquisition module of the application;

[0066] Figure 3 It is a functional flow chart of the data preprocessing module of the application;

[0067] Figure 4 It is a structural schematic diagram of the model construction module of the application;

[0068] Figure 5 It is a training flow chart of the GBDT algorithm of the application;

[0069] Figure 6 It is a functional structure diagram of the prediction analysis module of the application;

[0070] Figure 7 It is a work flow chart of the feedback optimization module of the application;

[0071] Figure 8 It is a functional structure diagram of the modeling environment module of the application;

[0072] Figure 9 It is a flow chart of the soft soil area roadbed settlement prediction method of the application;

[0073] Figure 10 It is a comparison schematic diagram of the prediction result and the measured data of the application;

[0074] Figure 11 It is an architecture diagram of the multi-scale spatio-temporal feature extraction and fusion technology of the application;

[0075] Figure 12 It is a working principle diagram of the spatio-temporal self-attention mechanism of the application. DETAILED DESCRIPTION

[0076] Please refer to the accompanying Figures 1-12 The application will be further described in detail below in combination with the drawings and specific embodiments.

[0077] As Figure 1 shown, the present application provides a soft soil area subgrade settlement prediction system, mainly comprising: data acquisition module 1, data preprocessing module 2, model construction module 3, prediction analysis module 4, feedback optimization module 5 and modeling environment module 6. Each module is connected through communication to form a complete workflow, realizing accurate prediction of soft soil area subgrade settlement.

[0078] As Figure 2 shown, the data acquisition module 1 includes a subgrade parameter acquisition unit 11, a geological survey data acquisition unit 12, a settlement historical data acquisition unit 13 and a real-time monitoring unit 14.

[0079] The subgrade parameter acquisition unit 11 is used to acquire basic parameters of the soft soil area subgrade, such as fill height, soil type and design type. In actual application, the fill height is usually measured by a total station, with an accuracy of ±1 mm; the soil type is determined by field sampling and laboratory testing, mainly including indicators such as moisture content, density, liquid-plastic limit; the design type records the structure form of the subgrade, such as ordinary embankment, bridge transition section, replacement subgrade, etc.

[0080] The geological survey data acquisition unit 12 is used to acquire the depth of the soft soil area and the soil layer longitudinal section. Preferably, the system uses the method of multi-point drilling combined with geological radar to obtain the geological survey data, and the drilling point spacing is usually set to 50-100 m, which is appropriately increased or reduced according to the complexity of the geological conditions. The soil layer longitudinal section is generated by professional geological modeling software, which is used to intuitively reflect the distribution of underground soil layers.

[0081] The settlement historical data acquisition unit 13 is used to acquire the measured values of subgrade settlement at different time nodes. In an embodiment of the present application, a settlement observation point is buried every 10-20 m along the cross section on the surface of the subgrade, and the elevation change of each observation point is measured regularly (usually once a week, which can be increased to once a day in the early stage of construction) to generate a settlement-time curve.

[0082] The real-time monitoring unit 14 is used to monitor the subgrade settlement in real time and acquire the current settlement data. This unit uses an automatic settlement monitoring system, including precise level, pressure sensor and data acquisition device, etc., which can realize 24-hour continuous monitoring, and the monitoring data is uploaded to the system database in real time through wireless transmission.

[0083] As Figure 3 shown, the data preprocessing module 2 includes an outlier processing unit 21, a data conversion unit 22, a normalization processing unit 23 and a correlation analysis unit 24.

[0084] The outlier processing unit 21 is used to identify and eliminate outliers, missing values and repeated values in the data. The present application adopts an outlier detection method based on statistical principles. When the data deviates from the mean value by more than 3 times the standard deviation, it is determined to be an outlier and is eliminated. For missing values, different processing strategies are adopted according to the data density: when the number of missing points is less than 5% of the total data, interpolation method is used to fill; when the number of missing points is large, time series prediction method is used for reasonable estimation.

[0085] The data conversion unit 22 is used to convert the preprocessed data into an ordered list form. Specifically, this unit arranges the collected multi-dimensional data in chronological order to form a standardized two-dimensional table, where the rows represent different observation points, the columns represent different time nodes, and the cell values are the corresponding sedimentation amounts. This standardized data structure facilitates subsequent model training and prediction.

[0086] The normalization processing unit 23 is used to map the data to the interval [0, 1] using the maximum and minimum normalization method. The normalization process uses the following formula:

[0087] ,

[0088] where, is the normalized value, is the original value, is the minimum value in the data set, is the maximum value in the data set. Normalization processing can eliminate the influence of different index dimensions, improve model training efficiency and prediction accuracy.

[0089] The correlation analysis unit 24 is used to analyze the linear correlation coefficient between the data, and when the correlation coefficient is greater than a preset threshold, the relationship between the parameters is established. The present application uses Pearson correlation coefficient to evaluate the linear correlation between variables:

[0090] ,

[0091] where, is the correlation coefficient, and are the observation values of the two variables, and are the respective means, is the sample size. According to experimental verification, when , it can be considered that there is a strong linear correlation between the two variables, and the linear relationship between the parameters is established at this time; when , the original data is directly used for subsequent processing.

[0092] For example, Figure 4As shown, the model construction module 3 comprises a feature extraction unit 31, a GBDT algorithm unit 32, a tree structure optimization unit 33 and a physical model fusion unit 34.

[0093] The feature extraction unit 31 is configured to extract input features required by the prediction model from the pre-processed data. The present application screens out the most critical features for settlement prediction through a combination of expert knowledge and feature importance analysis, including but not limited to: soft soil layer thickness, water content, compression index, subgrade fill height, fill rate, preloading time, etc. Preferably, the feature extraction process also includes feature combination and interaction feature generation to capture the non-linear relationship between different features.

[0094] The GBDT algorithm unit 32 is configured to train weak classifiers and construct a gradient boosting decision tree model through an iterative manner. Gradient boosting decision tree (GBDT) is a powerful ensemble learning algorithm that combines multiple weak learners (usually decision trees) into a strong learner to achieve high-precision prediction. The GBDT algorithm used in the present application has the following process:

[0095] 1) Initialize model parameters, let , initial model , where is the loss function, usually the mean square error.

[0096] 2) For each sample point train a weak classifier . First, calculate the negative gradient:

[0097] ,

[0098] 3) Fit a regression tree to get the leaf node area .

[0099] 4) For each leaf node area calculate the optimal prediction value:

[0100] ,

[0101] 5) Update the model:

[0102] ,

[0103] where is the learning rate, usually set to 0.1-0.3, is the indicator function.

[0104] 6) Iterate steps 2-5 until the termination condition is reached (such as maximum number of iterations or error less than threshold).

[0105] The tree structure optimization unit 33 is used to control the model complexity by the tree structure complexity parameter γ and the leaf node penalty parameter λ. In practical applications, the maximum depth of the tree is usually set to 3-7, and a shallower tree helps to prevent overfitting; the minimum sample number of the leaf node is set to 1%-5% of the total number of samples, ensuring that each leaf node has sufficient sample support. The complexity parameter γ and the penalty parameter λ are determined by cross-validation to determine the optimal value, and the typical values are 0.1-0.5 and 1-10, respectively.

[0106] The physical model fusion unit 34 is used to fuse the physical test results with the numerical simulation results, enhancing the physical meaning of the model. The present application uses physical constraints to guide the training of the data-driven model, specifically, the theoretical settlement amount is calculated using the classical theory in soil mechanics (such as Terzaghi's one-dimensional consolidation theory) as prior knowledge for model training; at the same time, numerical simulation is performed by ANSYS and other finite element software, and the simulation results are used as auxiliary training data to improve the performance of the model in data sparse areas.

[0107] As shown in Figure 11 , the present application also innovatively proposes a multi-scale spatio-temporal feature extraction and fusion technology as the core innovative point of the feature extraction unit 31. This technology aims to capture the complex characteristics of soft soil area subgrade settlement from two dimensions of time and space, and significantly improve the accuracy and generalization ability of the prediction model through multi-scale analysis and feature fusion.

[0108] The multi-scale spatio-temporal feature extraction and fusion technology includes the following core steps:

[0109] 1) Time scale decomposition: decompose the settlement time series into three scale components of short-term fluctuations, medium-term trends and long-term evolution. The present application uses wavelet decomposition to realize time scale decomposition, and its mathematical expression is:

[0110] ,

[0111] Among them, is the original time series, is the wavelet basis function, is the scale function, is the detail coefficient, is the approximation coefficient, denotes the decomposition level, denotes the displacement parameter. The present application preferably uses db4 wavelet, and the decomposition level J=3, which can effectively separate the settlement characteristics of different time scales.

[0112] 2) Spatial scale feature extraction: Extract settlement features from three spatial scales of point, line, and plane. Point scale focuses on the settlement characteristics of individual monitoring points; line scale focuses on the settlement distribution law along the cross-section or longitudinal section; plane scale focuses on the settlement spatial distribution pattern of the entire roadbed area. The invention uses spatial interpolation and tensor decomposition methods to construct multi-scale spatial features:

[0113] Point scale feature: Directly use the settlement data of monitoring points and its first derivative (settlement rate) and second derivative (settlement acceleration).

[0114] Line scale feature: Construct the settlement distribution function along the line profile by B-spline interpolation:

[0115] ,

[0116] where is the settlement amount at position x on the profile, is the k-order B-spline basis function, is the control point. In this invention, k=3, i.e., cubic B-spline is used to ensure the smoothness of the curve.

[0117] Plane scale feature: Extract the main mode from the two-dimensional settlement distribution using tensor decomposition method:

[0118] ,

[0119] where, is the settlement data tensor, and are the component vectors of the rth CP (CANDECOMP / PARAFAC) decomposition, 0 represents the outer product operation, is the decomposition rank (usually selected as 3-5).

[0120] 3) Spatiotemporal feature fusion: Fusion of different scale time and space features through adaptive weight. The invention uses a feature fusion method based on attention mechanism to automatically learn the importance weight of different features:

[0121] ,

[0122] where, is the fused feature vector, is the ith time scale feature, is the first spatial scale feature and are the attention weights of the corresponding features, is the number of time scale features, is the number of spatial scale features. Attention weights are calculated as follows:

[0123] ​​​ ,

[0124] ,

[0125] where, and are learnable parameter vectors.

[0126] The core advantage of the multi-scale spatio-temporal feature extraction and fusion technology is that: first, it can capture the settlement evolution law of different time scales, including short-term fluctuations (such as the influence of diurnal temperature difference), medium-term trends (such as seasonal changes) and long-term evolution (such as consolidation process); second, it can comprehensively consider the settlement characteristics of point, line and surface three spatial scales, and establish a more comprehensive spatial distribution model; finally, through the attention mechanism, the adaptive fusion of features is realized, and the importance weight of each feature is automatically adjusted according to different engineering scenes and geological conditions.

[0127] The experimental results show that after using the multi-scale spatio-temporal feature extraction and fusion technology, the prediction accuracy of the prediction model in different periods (construction period, initial consolidation period and long-term consolidation period) is significantly improved, and the average prediction error is further reduced by 18.5% compared with the traditional method, especially in the long-term settlement prediction.

[0128] As shown in Figure 12 , in order to further improve the effective use of multi-scale spatio-temporal features, the present application innovatively introduces a spatio-temporal self-attention mechanism as the core technical innovation point of the prediction analysis module 4. This mechanism can automatically discover and utilize the long-short term time dependence and far-near range spatial correlation in the settlement data, significantly enhancing the model's understanding ability of complex settlement patterns.

[0129] The core algorithm of the spatio-temporal self-attention mechanism is as follows:

[0130] 1) Time self-attention calculation: capture the dependence between different time points

[0131] Firstly, the time series features are mapped into three representations: Query, Key and Value:

[0132] ,

[0133] ,

[0134] ,

[0135] where, is the time series feature matrix, and are learnable mapping matrices.

[0136] Then, the attention weight matrix is calculated:

[0137] ,

[0138] where, is the feature dimension, used to scale the dot product to avoid the gradient vanishing problem.

[0139] Finally, the weighted features are calculated:

[0140] ,

[0141] where, is the feature representation after adding time self-attention information.

[0142] 2) Spatial self-attention calculation: capturing spatial correlation between different monitoring points

[0143] Similar to time self-attention, the spatial self-attention mechanism is implemented through the following steps:

[0144] ,

[0145] ,

[0146] ,

[0147] ,

[0148] ,

[0149] where, is the spatial feature matrix, is the spatial distance matrix, introducing Geographical distance information can be integrated into attention calculation. The calculation formula is:

[0150] ,

[0151] where, is the geographical distance between monitoring points i and j, is a learnable distance decay coefficient, usually initialized to 0.1.

[0152] 3) Spatio-temporal attention fusion: integrating attention information in time and space dimensions

[0153] The invention adopts a gating mechanism to integrate time and spatial attention features:

[0154] ,

[0155] ,

[0156] wherein, is a sigmoid activation function, denotes the concatenation of time and spatial features, and are learnable parameters, denotes element-wise multiplication, is the final fused spatio-temporal feature.

[0157] 4) Multi-head attention: To improve the model's expressive power, the invention adopts a multi-head attention mechanism, which simultaneously computes h independent attentions and then merges the results:

[0158] ,

[0159] wherein, is the output of the i-th attention head, is the output mapping matrix, is the number of attention heads, which is set to 8 in the invention.

[0160] The core advantage of the spatio-temporal self-attention mechanism lies in the following aspects: first, it can capture long-term dependencies and avoid the gradient vanishing problem in traditional recurrent neural networks; second, it can automatically discover and utilize the spatial correlation between different monitoring points, effectively integrating regional settlement information; finally, through the gating fusion mechanism, it adaptively integrates time and spatial information, adjusting the importance weight of the two according to the specific scene.

[0161] In practical applications, the spatio-temporal self-attention mechanism is usually combined with the GBDT algorithm to form a two-stage prediction framework: the first stage uses the spatio-temporal self-attention mechanism to extract high-order spatio-temporal features; the second stage takes these features as the input of the GBDT model to perform the final settlement prediction. This combination takes advantage of the complementary strengths of the two algorithms: the spatio-temporal self-attention mechanism is good at capturing complex spatio-temporal dependencies, while GBDT is good at handling non-linear mappings and heterogeneous features.

[0162] Experimental results show that after introducing the spatio-temporal self-attention mechanism, the model has significantly improved in prediction accuracy and robustness: the average prediction error is further reduced by 15.3%; the recognition ability for irregular settlement patterns (such as sudden settlement, phased settlement) is significantly enhanced; the model's anti-interference ability to interference factors (such as weather changes, temporary loads) has also been significantly improved.

[0163] As shown in Figure 6 , the prediction analysis module 4 includes an input processing unit 41, a multi-model prediction unit 42, a result evaluation unit 43, and a settlement curve generation unit 44.

[0164] The input processing unit 41 is used to convert the parameters of the subgrade to be predicted into the input format required by the model. This unit first checks the input parameters to ensure data integrity and consistency; then applies the same preprocessing and normalization methods as the training data to ensure that the input features are consistent with the format expected by the model; finally, fills in the missing features with reasonable default or estimated values.

[0165] The multi-model prediction unit 42 is used to obtain multiple sets of prediction results by parallel computing of multiple prediction models. The present application uses the idea of model integration and uses different parameters and structures of GBDT models for prediction. Preferably, 5-10 differentiated models are constructed, which differ in training data subsets, feature selection, tree depth, etc., so as to capture different aspects of the data.

[0166] The result evaluation unit 43 is used to evaluate the errors of each prediction result and select the result with the smallest error as the final prediction value. The evaluation process is based on historical prediction performance and uses a weighted voting method to integrate the prediction results of each model. The weight calculation formula is:

[0167] ,

[0168] wherein, is the weight of the i-th model, is the historical error of the model on the validation set, is the total number of models. In this way, the model with better prediction performance gets higher weight, improving the overall prediction accuracy.

[0169] The settlement curve generation unit 44 is used to generate the relationship curve of subgrade settlement over time according to the final prediction value. This unit uses numerical fitting method to fit the settlement-time relationship by hyperbolic function or exponential function:

[0170] ,

[0171] or

[0172] ,

[0173] wherein, is the settlement amount at time t, is the final settlement amount, and are fitting parameters. In this way, the system can not only predict the settlement amount at a specific time point, but also predict the settlement development trend, providing more comprehensive support for engineering decision-making.

[0174] As Figure 7As shown, the feedback optimization module 5 includes an error calculation unit 51, an AI parameter correction unit 52, a historical data correction unit 53, and a model updating unit 54.

[0175] The error calculation unit 51 is used to calculate the error between the predicted value and the measured value. The present application uses two indicators, mean absolute error (MAE) and root mean square error (RMSE), to evaluate the prediction accuracy:

[0176] ,

[0177] ,

[0178] wherein, is the measured value, is the predicted value, is the sample size. When the MAE or RMSE exceeds the preset threshold (usually 10% of the measured value), the model optimization process is triggered.

[0179] The AI parameter correction unit 52 is used to automatically correct the parameters based on artificial intelligence algorithms when the error is greater than the preset threshold. This unit uses the Bayesian optimization algorithm to automatically adjust the model parameters. This algorithm establishes a probabilistic model (Gaussian process) between parameters and model performance, efficiently explores the parameter space, and finds the optimal parameter combination. Key parameters include learning rate, maximum depth of tree, minimum leaf node sample size, etc.

[0180] The historical data correction unit 53 is used to input parameters based on the error correction model of historical prediction data and measured data. This unit establishes an error compensation mechanism by analyzing the pattern of historical prediction errors and systematically adjusts new prediction results. Specifically, a mapping relationship between error and input features is established:

[0181] ,

[0182] wherein, is the prediction error, is the input feature. In this way, the system can learn and compensate for the systematic errors of the model, improving the prediction accuracy.

[0183] The model updating unit 54 is used to optimize and update the prediction model according to the correction results. This unit uses an incremental learning strategy to integrate new data and correction results into the existing model without retraining the entire model. Preferably, the model update uses a sliding window mechanism, which focuses more on recent data, allowing the model to adapt to changes in geological conditions and settlement patterns.

[0184] Specific implementation of AI parameter correction unit 52

[0185] The AI parameter correction unit 52 is specifically configured to establish a mapping relationship between the prediction parameters and the errors; identify the key parameters that have the greatest impact on the prediction results; dynamically adjust the weights of the key parameters according to the real-time settlement data; and feed back the corrected parameters to the model construction module for model optimization.

[0186] In actual application, the AI parameter correction unit 52 first determines the parameters that have the greatest impact on the prediction results through sensitivity analysis. Specifically, through the control variable method, the influence degree of each parameter on the prediction results within a certain change range is analyzed, and the influence degree is sorted. Generally, the thickness of the soft soil layer, the compression index and the load strength are the three most significant parameters affecting the settlement prediction.

[0187] The parameter correction adopts the gradient descent method to minimize the prediction error:

[0188] ,

[0189] wherein, is a model parameter vector, is a learning rate (usually set to 0.01-0.05), is a loss function, is the gradient of the loss function with respect to the parameters. To prevent overfitting, a regularization term is also introduced:

[0190] ,

[0191] wherein, is a regularization coefficient, usually set to 0.001-0.01.

[0192] To improve the parameter correction efficiency, the unit also adopts an adaptive learning rate strategy, which increases the learning rate when the parameter update direction is consistent for multiple times in a row, and decreases the learning rate when the parameter update direction frequently changes. This strategy significantly improves the convergence speed and stability of parameter optimization.

[0193] As shown in Figure 8 , the application also includes a modeling environment module 6, which is in communication connection with the model construction module 3, and is used to store multiple sets of modeling environment information, including modeling environment data, modeling environment analysis data and modeling environment judgment data; and the optimal modeling environment is selected through a similarity comparison analysis method.

[0194] The key function of the modeling environment module 6 is to select the most suitable modeling environment for the current prediction task from multiple candidate environments through a similarity comparison analysis method. The similarity comparison uses the following formula:

[0195] ,

[0196] wherein, is the similarity comparison analysis result, This is a number assigned to the modeling environment data information. For the numbering of modeling environment information, Represents modeling environment data information With modeling environment information The similarity coefficient, Indicate their correlation. The similarity weight (usually 0.6) This is the relevance weight (usually 0.4).

[0197] Similarity coefficient The calculation formula is:

[0198] ,

[0199] in, and These are modeling environment data information. The maximum and minimum values, and These are respectively modeling environment information The maximum and minimum values.

[0200] Correlation The calculation formula is:

[0201] ,

[0202] in, For modeling environmental data information The value, For modeling environmental information The value, Represents modeling environment data information The average value, Represents modeling environment information The average value.

[0203] By calculating the similarity comparison results of different modeling environments, we select... The environment with the highest value is used as the final modeling environment, providing the most suitable parameter settings and training strategies for the prediction model.

[0204] like Figure 9 As shown, the present invention also provides a method for predicting subgrade settlement in soft soil areas, comprising the following steps:

[0205] S1. Collect roadbed parameters and historical settlement data in soft soil areas;

[0206] S2. Perform outlier removal and normalization on the collected data;

[0207] S3, constructing a subgrade settlement prediction model based on a gradient boosting decision tree algorithm;

[0208] S4, performing settlement prediction according to the processed data and the constructed model;

[0209] S5, comparing the prediction result with the measured data;

[0210] S6, when the prediction error is greater than a preset threshold, correcting the model parameters based on an artificial intelligence algorithm;

[0211] S7, updating the prediction model according to the correction result;

[0212] S8, re-performing settlement prediction using the updated model until the prediction error is less than the preset threshold;

[0213] S9, outputting a final subgrade settlement prediction result, including a settlement amount and a relationship curve of settlement change over time.

[0214] In step S1, the collected soft soil area subgrade parameters include fill height, soil type, design type, etc.; and the settlement historical data include settlement observation values at different time nodes. Preferably, automatic monitoring equipment such as a precision level and a pressure sensor is used to improve the accuracy and efficiency of data collection.

[0215] In step S2, the abnormal value determination standard is 3 times the standard deviation from the mean; and the normalization adopts a maximum and minimum normalization method to map the data to the [0, 1] interval.

[0216] In step S3, the key parameters of the GBDT algorithm are set as follows: the learning rate is 0.1, the maximum depth of the tree is 5, the minimum leaf node sample number is 2% of the total samples, the maximum iteration number is 500, and the early stopping parameter is 10 (i.e., if the model performance does not improve for 10 consecutive iterations, the training is stopped). In addition, this step also applies multi-scale spatiotemporal feature extraction and fusion technology to extract richer feature representations and enhance the prediction ability of the model.

[0217] In step S4, multiple prediction models are used for parallel computing to obtain multiple sets of prediction results. Specifically, 5-10 differentiated GBDT models are constructed, which differ in training data subsets, feature selection, tree depth, etc. At the same time, a spatiotemporal self-attention mechanism is applied to enhance the modeling ability of complex spatiotemporal dependencies.

[0218] In step S5, the mean absolute error (MAE) and the root mean square error (RMSE) are used to evaluate the prediction accuracy. When the MAE or RMSE exceeds 10% of the measured value, the parameter correction process is triggered.

[0219] In step S6, the model parameters are automatically adjusted based on a Bayesian optimization algorithm. This algorithm efficiently explores the parameter space by establishing a probabilistic model (Gaussian process) between parameters and model performance to find the optimal parameter combination.

[0220] In step S7, the prediction model is updated using an incremental learning strategy, integrating new data and correction results into the existing model without the need to retrain the entire model.

[0221] In step S8, the updated model is used to re-predict the settlement, and steps S5-S7 are repeated until the prediction error is less than a preset threshold (usually 5% of the measured value).

[0222] In step S9, the final settlement prediction results are output, including predicted settlement amounts at different time points and a settlement-time curve. Preferably, the settlement curve is fitted using a hyperbolic function or an exponential function for intuitive understanding of the settlement development trend by engineers.

[0223] Through the detailed description of the above embodiments, those skilled in the art should understand that the present application is not limited to the above specific embodiments, and various modifications and improvements can be made without departing from the concept of the present application, and these modifications and improvements all fall within the protection scope of the present application.

Claims

1. A soft soil subgrade settlement prediction system, characterized by, The application relates to a soft soil area roadbed settlement prediction method and device. The method comprises the following steps: A data acquisition module is used to acquire roadbed parameter and settlement history data in a soft soil area; A data preprocessing module is in communication connection with the data acquisition module and is used to perform outlier rejection and normalization processing on the acquired data; A model construction module is in communication connection with the data preprocessing module and is used to construct a roadbed settlement prediction model based on a gradient boosting decision tree algorithm; A prediction analysis module is in communication connection with the model construction module and is used to perform settlement prediction according to the processed data and the constructed model; A feedback optimization module is in communication connection with the prediction analysis module and is used to compare the prediction result with the measured data and optimize the prediction model; The model construction module comprises a feature extraction unit, a GBDT algorithm unit and a tree structure optimization unit. The feature extraction unit adopts a multi-scale spatiotemporal feature extraction and fusion technology, which comprises the following core steps: 1) Time scale decomposition: the settlement time series is decomposed into three scale components of short-term fluctuation, medium-term trend and long-term evolution by wavelet decomposition; 2) Spatial scale feature extraction: the settlement features are extracted from three spatial scales of point, line and surface; ， wherein is the amount of settlement at position x on the profile, is the k-th B-spline basis function, is the control point; Point scale feature: the settlement data of the monitoring point and the first and second derivatives thereof are directly used; ， wherein, is the sedimentation data tensor, and are the component vectors of the rth CP decomposition, denotes the outer product operation, is the decomposition rank; Line scale feature: the settlement distribution function along the line profile is constructed by B-spline interpolation; Surface scale feature: the main mode is extracted from the two-dimensional settlement distribution by using a tensor decomposition method; 3) Spatiotemporal feature fusion: the time and space features of different scales are fused by adaptive weight; The prediction analysis module introduces a spatiotemporal self-attention mechanism, The core algorithm of the spatiotemporal self-attention mechanism is as follows: 1) Time self-attention calculation: the dependency relationship between different time points is captured; , , ， ， ， wherein, a spatial feature matrix, a spatial distance matrix, Geographical distance information can be incorporated into attention calculation, The calculation formula is: ， wherein, is the geographical distance between the monitoring points i and j, is the learnable distance decay coefficient; 2) Spatial self-attention calculation: the spatial correlation between different monitoring points is captured; The spatial self-attention mechanism is realized through the following steps:

2. The soft soil subgrade settlement prediction system according to claim 1, wherein, 3) Spatiotemporal attention fusion: the attention information of the time and space dimensions is integrated; 4) Multi-head attention: the multi-head attention mechanism is adopted, and then the results are combined. The data acquisition module comprises a roadbed parameter acquisition unit, a geological exploration data acquisition unit, a settlement history data acquisition unit and a real-time monitoring unit. The roadbed parameter acquisition unit is used to acquire the filling height, soil type and design type of the roadbed in the soft soil area; The geological exploration data acquisition unit is used to acquire the soil layer depth and soil layer longitudinal section diagram in the soft soil area; The settlement history data acquisition unit is used to acquire the measured value of the roadbed settlement at different time nodes; The real-time monitoring unit is used to perform real-time monitoring on the roadbed settlement and acquire current settlement data.

3. The soft soil subgrade settlement prediction system according to claim 1, wherein The data preprocessing module comprises: an outlier processing unit for identifying and eliminating outliers, missing values and repeated values in the data; a data conversion unit for converting the preprocessed data into an ordered list form; a normalization processing unit for mapping the data to the interval [0, 1] using the maximum and minimum normalization method; a correlation analysis unit for analyzing the linear correlation coefficient between the data, and establishing a relationship between parameters when the correlation coefficient is greater than a preset threshold.

4. The soft soil subgrade settlement prediction system according to claim 1, wherein The GBDT algorithm unit is specifically configured to: initialize model parameters, and let t=1; train a weak classifier for each sample point i wherein the weak classifier determines the optimal split point by calculating the average loss of each leaf region sample; Calculate the total error and update the tree structure complexity parameter γ and the leaf node penalty parameter λ; optimize the model through the iterative process of Taylor formula expansion; judge whether the termination condition is met, and output the final model H(x) if the termination condition is met.

5. The soft soil subgrade settlement prediction system according to claim 1, wherein, The prediction analysis module comprises: an input processing unit for converting the parameters of the roadbed to be predicted into the input format required by the model; a multi-model prediction unit for obtaining multiple sets of prediction results through parallel calculation of multiple prediction models; a result evaluation unit for evaluating the errors of each prediction result and selecting the result with the smallest error as the final prediction value; a settlement curve generation unit for generating a relationship curve of roadbed settlement change over time according to the final prediction value.

6. The soft soil subgrade settlement prediction system according to claim 1, wherein, The feedback optimization module comprises: an error calculation unit for calculating the error between the prediction value and the measured value; an AI parameter correction unit for automatically correcting the parameters based on an artificial intelligence algorithm when the error is greater than a preset threshold; a historical data correction unit for inputting parameters based on an error correction model of historical prediction data and measured data; a model updating unit for optimizing and updating the prediction model according to the correction result.

7. The soft soil subgrade settlement prediction system according to claim 6, wherein, The AI parameter correction unit is specifically configured to: establish a mapping relationship between the prediction parameters and the error; identify the key parameters that have the greatest impact on the prediction result; dynamically adjust the weight of the key parameters according to the real-time settlement data; and feed back the corrected parameters to the model construction module for model optimization.

8. The soft soil subgrade settlement prediction system according to claim 1, wherein, Further comprising a modeling environment module in communication connection with the model construction module, configured to: store multiple sets of modeling environment information, including modeling environment data, modeling environment analysis data and modeling environment judgment data; select the optimal modeling environment through a similarity comparison analysis method; wherein the similarity comparison analysis method comprises calculating the similarity coefficient and correlation of the modeling environment data information and the modeling environment information, and selecting the optimal solution based on the weighted result.

9. A method for predicting subgrade settlement in soft soil areas using the system of any one of claims 1-8. The method comprises the following steps: S1, collecting roadbed parameters and settlement history data in soft soil areas; S2, performing outlier elimination and normalization processing on the collected data; S3, constructing a roadbed settlement prediction model based on the gradient boosting decision tree algorithm; S4, predicting settlement based on the processed data and the constructed model; S5, comparing the prediction result with the measured data; S6, when the prediction error is greater than a preset threshold, correcting the model parameters based on an artificial intelligence algorithm; S7, updating the prediction model according to the correction result; S8, using the updated model to re-predict settlement until the prediction error is less than the preset threshold; S9, outputting the final roadbed settlement prediction result, including the settlement amount and the relationship curve of settlement change over time.

Images