Road collapse risk comprehensive assessment method based on multi-source data fusion

Abstract

The invention relates to a road collapse risk comprehensive assessment method based on multi-source data fusion, and the method comprises the following steps: collecting and preprocessing multi-source data related to road collapse, the multi-source data related to road collapse comprising remote sensing image data, underground water level monitoring data, geological exploration data and surface deformation monitoring data; spatio-temporal feature fusion and multi-level data fusion are carried out on the preprocessed multi-source data, the spatio-temporal feature fusion comprises spatial registration and time alignment, and the multi-level data fusion comprises low-level data fusion and high-level data fusion; and performing feature extraction on the fused multi-level data, inputting a feature extraction result into a road collapse risk assessment model constructed based on a machine learning model, outputting a road collapse risk occurrence probability, and generating a road collapse risk level according to the road collapse risk occurrence probability. Compared with the prior art, the method can comprehensively and accurately predict and evaluate the risk of road collapse.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster prevention and control, in particular to risk assessment and monitoring of road collapse, and specifically relates to a comprehensive road collapse risk assessment method based on multi-source data fusion. Background Art

[0002] Road collapse is one of the common and serious geological disasters in the process of urban development and infrastructure construction. The occurrence of road collapse will not only cause damage to traffic, buildings and underground facilities, but may even threaten human life and safety. With the acceleration of urbanization and the development and utilization of underground space, road collapse is becoming more and more common. It is particularly important to study and develop scientific assessment and early warning methods for road collapse.

[0003] At present, research on road collapse monitoring and assessment has made certain progress at home and abroad, mainly including assessment methods based on remote sensing technology, geographic information system (GIS), laser scanning technology, road geophysical exploration, and numerical simulation. These methods analyze and predict the risk of collapse areas from different perspectives. For example, remote sensing image data is used to monitor surface deformation, and GIS is combined to perform spatial analysis of road collapse areas, or high-precision laser radar is used to obtain topographic changes in collapse areas. However, due to the complex formation process of road collapse and the diverse influencing factors, including geological conditions, hydrological changes, and human activities, a single data source or a single method is difficult to comprehensively and accurately assess road collapse. In order to better deal with the risk of road collapse, in recent years, road collapse monitoring and assessment methods based on multi-source data fusion have gradually received attention. These methods can improve the accuracy and timeliness of assessment by integrating data sources from multiple sensors, such as remote sensing images, groundwater level monitoring data, and geological exploration data. For example, some studies use remote sensing and geophysical data to monitor and warn the dynamic changes of road collapse in real time; some studies also introduce artificial intelligence algorithms to extract useful features from multi-source data to achieve comprehensive evaluation and prediction of road collapse. However, the multi-source data fusion method still faces some challenges in practical applications, such as data heterogeneity, effective integration of multi-source data, and accuracy and adaptability of evaluation models. These problems limit its widespread application in road collapse monitoring.

[0004] In summary, in view of the complex formation mechanism of road collapse and its diverse influencing factors, there is an urgent need for an efficient and comprehensive road collapse assessment method to ensure the accuracy, reliability and efficiency of the assessment results and provide a scientific basis for urban planning and disaster prevention and control. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a comprehensive road collapse risk assessment method based on multi-source data fusion, to solve the problems of insufficient assessment accuracy and applicability caused by the limitations of a single data source or assessment method in the prior art, to achieve more accurate road collapse risk assessment, and to provide effective support for the prevention, control and management of road collapse.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] The present invention provides a comprehensive assessment method for road collapse risk based on multi-source data fusion, comprising the following steps:

[0008] Collect and pre-process multi-source data related to road collapse, including remote sensing image data, groundwater level monitoring data, geological exploration data and surface deformation monitoring data;

[0009] Performing spatiotemporal feature fusion and multi-level data fusion on the pre-processed multi-source data, wherein the spatiotemporal feature fusion includes spatial registration and time alignment, and the multi-level data fusion includes low-level data fusion and high-level data fusion;

[0010] Feature extraction is performed on the fused multi-level data, and the feature extraction results are input into the road collapse risk assessment model built based on the machine learning model to output the probability of road collapse risk occurrence, thereby generating the risk level of road collapse.

[0011] Furthermore, the preprocessing specifically includes the following steps:

[0012] Use the statistical method Z-score analysis to identify outliers, remove or replace the identified outliers; use interpolation or statistical methods to complete missing data;

[0013] Noise filtering through filtering technology;

[0014] Using GIS technology to convert multi-source data related to road collapse into a unified geographic coordinate system, and using a benchmark method or a registration algorithm to perform spatial registration on the remote sensing image data;

[0015] The Gauss-Krüger projection coordinate transformation method is used to transform the data into a unified reference coordinate system and perform elevation datum transformation;

[0016] Perform normalization processing;

[0017] The principal component analysis method is used for dimensionality reduction.

[0018] Furthermore, the multi-source data are spatially registered using a linear transformation model, and the multi-source data with different acquisition frequencies are resampled to achieve the time alignment.

[0019] Furthermore, the low-level data fusion is performed by a weighted average method, and the weights used in the weighted average method are set according to the reliability of the data source; and the high-level data fusion is achieved by a principal component analysis method.

[0020] Furthermore, the specific process of feature extraction for the fused multi-level data is as follows:

[0021] The mutual information method was used for feature selection, followed by further screening by recursive feature elimination.

[0022] Furthermore, the machine learning model includes a logistic regression model, a random forest model and a BP neural network model.

[0023] Furthermore, the risk levels of road collapse include low risk, medium risk and high risk, and the classification criteria of the road collapse risk levels are determined by a fuzzy logic model or a hierarchical analysis method.

[0024] Furthermore, the classification standards for road collapse risk levels are dynamically adjusted based on historical and real-time road collapse occurrences.

[0025] Furthermore, a risk map is generated according to the risk level of road collapse and visualized through GIS technology.

[0026] Furthermore, the risk level of road collapse is input into the economic loss model, the corresponding total economic loss is calculated and visualized through GIS technology. The expression of the economic loss model is as follows:

[0027]

[0028] Where L is the estimated total economic loss; C i A is the unit value of the i-th type of affected object; i is the affected area or volume of the i-th type of object; R i It is the loss ratio corresponding to the risk level, which means the expected loss ratio under a certain risk level.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. The present invention proposes a comprehensive assessment method for road collapse risk based on multi-source data fusion. First, multi-source data related to road collapse are collected and pre-processed. The multi-source data related to road collapse specifically include remote sensing image data, groundwater level monitoring data, geological exploration data and surface deformation monitoring data, which can more comprehensively reflect the potential risk of road collapse and improve the accuracy and reliability of the assessment results; then, spatiotemporal feature fusion and multi-level data fusion are performed on the pre-processed multi-source data, and the complementary information of different data sources is comprehensively utilized to further improve the accuracy and reliability of the assessment results. Specifically, spatiotemporal feature fusion includes spatial registration and time alignment, which can align the multi-source data collected at different times and spaces to ensure that each data source is consistent in time and space. Consistency, multi-level data fusion includes low-level data fusion and high-level data fusion. Low-level data fusion can retain as much information as possible in the original data, and high-level data fusion can reduce data transmission and storage requirements and improve anti-interference ability; secondly, feature extraction is performed on the fused multi-level data to obtain features that are more important for road collapse risk assessment, which are input into the road collapse risk assessment model built based on the machine learning model, and the probability of road collapse risk is output, thereby generating the risk level of road collapse; the above method has strong robustness. By organically combining multiple data sources and analysis methods, it can comprehensively and accurately predict and assess the risk of road collapse, and provide scientific decision-making support for disaster prevention and control, urban planning and mine safety management.

[0031] 2. In the present invention, the preprocessing steps specifically include data cleaning, noise filtering, spatial registration, coordinate transformation, data standardization and data dimension reduction, which can fully ensure data consistency and quality, and further improve the accuracy and reliability of road collapse risk assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of an embodiment method. DETAILED DESCRIPTION

[0033] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0034] Example:

[0035] This embodiment provides a comprehensive assessment method for road collapse risk based on multi-source data fusion. Figure 1 As shown, the following steps are included:

[0036] S1. Collect multi-source data related to road collapse.

[0037] The multi-source data related to road collapse include but are not limited to remote sensing image data, groundwater level monitoring data, geological exploration data, surface deformation monitoring data and other auxiliary data. The specific steps for collecting each data are as follows:

[0038] (1) Remote sensing image data

[0039] Synthetic Aperture Radar Interferometry (InSAR): Synthetic aperture radar technology is used to obtain surface deformation information. The surface subsidence rate is obtained through data analysis of multi-temporal images in millimeters per year, reflecting the stability of the road.

[0040] Optical remote sensing images: Collect high-resolution optical remote sensing images at different times to obtain information on surface changes. Through image change detection technology, the change range of road subsidence areas can be quantitatively analyzed.

[0041] Multispectral and hyperspectral data: Use multispectral or hyperspectral imaging data to monitor changes in surface cover characteristics and surface materials, and determine the impact of geological conditions on road collapse by analyzing information such as vegetation, water bodies, and soil.

[0042] (2) Groundwater level monitoring data

[0043] Groundwater level depth measurement: Using groundwater level monitoring wells, sensors (such as pressure sensors) are used to obtain the absolute depth of the groundwater level in meters in real time.

[0044] Groundwater level change rate: Record the groundwater level change rate (in meters per day) at different time periods to analyze the fluctuation characteristics of groundwater. Sharp fluctuations in groundwater levels may lead to unstable soil structure and cause road collapse.

[0045] Groundwater quality monitoring: Collect groundwater samples and analyze their main chemical components (such as pH value, mineral concentration, etc.) to determine the impact of water chemical properties on the soil.

[0046] (3) Geological exploration data

[0047] Geological profile structure: Obtain geological profile data through drilling and sampling analysis to determine the thickness and distribution of soil and rock layers.

[0048] Soil physical and mechanical properties: measure soil density, moisture content, porosity, shear strength, etc. Used to evaluate road bearing capacity and foundation stability.

[0049] Stratigraphic lithology and distribution: Identify different rock and soil types (such as clay, sand, etc.), and measure their thickness, physical properties and relationship with surrounding strata to accurately assess stratum stability.

[0050] (4) Surface deformation monitoring data

[0051] GNSS monitoring: Use the Global Navigation Satellite System (GNSS) to monitor the precise position changes of the earth's surface, especially the vertical settlement displacement.

[0052] Laser Ranging and Detecting (LiDAR): Use laser radar scanning technology to obtain surface elevation data and analyze small surface deformations and changes in geomorphic features.

[0053] (5) Other auxiliary data

[0054] Geophysical data: Use geophysical methods such as resistivity method and seismic wave reflection method to obtain the distribution of underground cavities and the physical properties of soil in order to determine the impact of underground cavities or abnormal structures on the surface.

[0055] Human activity data: Collect information on human activities in the target area, including construction activities, groundwater extraction, etc., especially in urban areas and areas with intensive underground space development.

[0056] Meteorological and hydrological data: Collect data such as rainfall, evaporation, and river water level changes in the target area to analyze the impact of hydrological conditions on road collapse.

[0057] S2, preprocessing the data collected in step S1, specifically including the following steps:

[0058] S21. Data cleaning.

[0059] Outlier detection and elimination: Perform outlier detection on the collected raw data, and use the statistical method Z-score analysis to identify outliers. For the detected abnormal data points, they can be eliminated or replaced to reduce the impact of data errors on model evaluation. In the Z-score method, the Z value formula of the data point is as follows:

[0060]

[0061] Among them, Z i is the Z value of the ith data, X i is the ith data point, μ is the mean of the data set, and σ is the standard deviation of the data set. If the Z value of a data point exceeds a certain threshold (such as 3 or -3), it is considered an outlier.

[0062] Missing value filling: Use interpolation methods (such as linear interpolation, polynomial interpolation, etc.) or statistical methods (such as mean and median substitution) to fill in missing data, thereby improving the integrity and consistency of the data. For time series data, use time interpolation methods (such as ARIMA model) to fill in to ensure time continuity.

[0063] The formula for linear interpolation is:

[0064]

[0065] Among them, x is the position to be interpolated, (x1, y1) and (x2, y2) are adjacent known data points.

[0066] The ARIMA time series data interpolation model is:

[0067] X t =c+φ1X t-1 +φ2X t-2 +...+φ p X t-p +ε t

[0068] Among them, X t is the value at time t, c is a constant term, φ i is the autoregressive parameter of the model, ε t is the error term.

[0069] S22, noise filtering.

[0070] Noise filtering is performed through filtering technology, specifically:

[0071] (1) Kalman filter is used to remove noise from time series data (such as remote sensing images and groundwater level monitoring data). The state equation of Kalman filter is:

[0072]

[0073] in, is the state estimate at time k, is the prior estimate, that is, the predicted value at time k, K k is the Kalman gain, z k is the observed value, H k is the observation matrix.

[0074] Kalman gain K k The calculation formula is as follows:

[0075]

[0076] Among them, P k|k-1 is the covariance matrix of the prior estimate, and R is the covariance matrix of the observation noise.

[0077] (2) Static data or geological profile data are smoothed by mean filtering or median filtering to reduce the impact of random noise. Mean filtering is suitable for removing high-frequency noise, while median filtering has a better effect on spike noise.

[0078] Mean filter:

[0079]

[0080] Among them, y[i] is the i-th data point after filtering, N is the filter window size; x[j] is the original data point.

[0081] (3) Frequency domain analysis: Perform frequency domain conversion on the data (such as Fourier transform), separate high-frequency noise and low-frequency trends, and use a low-pass filter to filter out unnecessary high-frequency components to ensure the smoothness and accuracy of the data.

[0082] S23. Spatial registration.

[0083] (1) Spatial registration of remote sensing images: For remote sensing image data from different times and sources, spatial registration is performed using reference point methods (such as ground feature points) or registration algorithms (such as image matching algorithms based on cross-correlation) to ensure spatial consistency between images and reduce evaluation errors caused by registration errors.

[0084] The spatial registration model (geological feature point matching) uses the following similarity transformation model in the registration process:

[0085] x′=a1x+a2y+t x

[0086] y′=b1x+b2y+t y

[0087] Among them, (x, y) is the original image coordinate, (x′, y′) is the target image coordinate, a1, a2, b1, b2 are rotation and scaling parameters, t x ,t y is the translation parameter.

[0088] (2) Spatial consistency processing of multi-source data: For multi-source data such as remote sensing images, geological exploration data, and groundwater level monitoring data, use geographic information system (GIS) technology to convert the data into a unified geographic coordinate system. Use methods such as the nearest neighbor method or bilinear interpolation to resample to improve the accuracy of spatial registration.

[0089] S24. Coordinate conversion.

[0090] (1) Coordinate system unification: In the case that different data sources may use different coordinate systems, the Gauss-Krüger projection coordinate transformation method is used to convert the data into a unified reference coordinate system to ensure that each data source is calculated and analyzed within a unified spatial framework. The Gauss-Krüger projection uses the following transformation formula:

[0091]

[0092] Among them, x is the horizontal coordinate of the projection, N is the radius of the ellipsoid, λ0 is the longitude of the central meridian, φ is the latitude, T is the square of the tangent of the latitude, and C is a parameter that depends on the shape of the ellipsoid.

[0093] (2) Elevation datum conversion: The surface elevation data is subjected to unified datum processing to ensure that the elevation data obtained by different measurement methods (such as LiDAR and GNSS) can be compared and integrated under the same elevation datum (such as EGM2008).

[0094] S25. Data standardization.

[0095] The multi-source data is standardized to normalize the data of different dimensions to the same scale range to eliminate the dimension difference. In a preferred embodiment, the minimum-maximum normalization is used, that is, the data is linearly converted to the interval [0,1], which is suitable for the case where all data are positive values.

[0096]

[0097] Among them, X′ is the normalized data; X is the original data; X max and X min are the minimum and maximum values ​​of the original data respectively.

[0098] S26. Data dimensionality reduction.

[0099] For high-dimensional data, principal component analysis (PCA) is used to reduce dimensionality in order to reduce redundant information, reduce computational complexity, and retain the main features of the data.

[0100] The goal of PCA is to project high-dimensional data into a low-dimensional space through linear transformation. The formula is as follows:

[0101] Y=W T X

[0102] Among them, Y is the data after dimensionality reduction; W is the eigenvector matrix, which contains the principal components of the data covariance matrix; X is the original high-dimensional data. The key step of PCA is to calculate the data covariance matrix C:

[0103]

[0104] Where n is the number of samples, X i is the i-th data point, and μ is the mean vector of the data.

[0105] Through the above data preprocessing steps, not only can the accuracy of the data be improved, but also the noise and redundant information can be effectively reduced, ensuring the data coordination and integration effect in the multi-source data fusion process, and providing a data basis for the subsequent data fusion and risk assessment model construction.

[0106] S3. Perform spatiotemporal feature fusion and multi-level data fusion on the preprocessed multi-source data, and perform feature extraction on the fused multi-level data.

[0107] Data fusion is an important step to improve the accuracy and comprehensiveness of road collapse risk assessment. By fusing multi-source data, the formation conditions and dynamic changes of road collapse can be more comprehensively understood.

[0108] S31. Space-time feature fusion.

[0109] The purpose of spatiotemporal feature fusion is to align multi-source data collected at different times and spaces to ensure the consistency of each data source in time and space, including spatial registration and temporal alignment:

[0110] (1) Spatial registration. Use the linear transformation model to perform spatial registration on the data:

[0111] x′=T·x

[0112] Among them, x′ is the vector representation of the target coordinate point, T is the linear transformation matrix including rotation, scaling and translation operations, and x is the vector representation of the original coordinate point.

[0113] (2) Time alignment. Resample data with different acquisition frequencies to achieve time alignment. The time resampling formula is:

[0114]

[0115] Among them, X′(t) is the data at the target time t, t0, t1 are the times of known data points, and X(t0), X(t1) are the corresponding known data values.

[0116] S32. Multi-level data fusion.

[0117] Multi-level data fusion is carried out step by step from low level to high level, including the fusion of observation data (low-level data fusion) and the fusion of features (high-level data fusion).

[0118] (1) Low-level data fusion through weighted average method:

[0119]

[0120] Among them, X fused is the fused data value, X i is the data value of the i-th data source, ω i The weight of the i-th data source is usually set according to the reliability of the data source.

[0121] (2) High-level data fusion through principal component analysis (PCA).

[0122] Through the above data fusion steps (S31, S32), the comprehensive utilization of multi-source data can be achieved, and the limitations of a single data source can be effectively overcome to more accurately assess the risk of road collapse.

[0123] S33. Feature extraction.

[0124] Through feature engineering, the most important features for road collapse risk assessment are extracted from the fused multi-source data to improve the accuracy and efficiency of subsequent models.

[0125] In a preferred embodiment, the mutual information method can be used for feature selection, and then further screening can be performed by recursive feature elimination. The formula of the mutual information method is:

[0126]

[0127] Where I(X;Y) is the mutual information of random variables X and Y, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y, respectively.

[0128] Recursive feature elimination uses a basic learner (linear regression or support vector machine) to train the model, obtains the weight of each feature, and repeatedly removes the feature with the smallest absolute weight until only the expected number of features remain, gradually reducing the number of features.

[0129] S4. Input the feature extraction results into the road collapse risk assessment model built based on the machine learning model, and output the probability of road collapse risk occurrence.

[0130] The steps for building, training and evaluating the road collapse risk assessment model are as follows:

[0131] S41. Use machine learning models such as logistic regression model, random forest model and BP neural network model to establish a road collapse risk assessment model.

[0132] (1) Logistic regression model

[0133] The logistic regression model is a widely used method for binary classification problems and is suitable for probabilistic assessment of high, medium, and low risks of road collapse.

[0134]

[0135] Among them, P(Y=1|X) is the probability of the target variable Y (indicating the occurrence of collapse risk) when the feature X is given, β0 is the bias term, and β i is the weight of the i-th feature, X i is the ith input feature.

[0136] Logistic regression learns parameters by maximizing the log-likelihood estimate:

[0137]

[0138] Among them, L(β) is the log-likelihood function, y (i) is the target value (0 or 1) of the i-th sample, P(Y=1|X (i) ) is the predicted probability of the i-th sample.

[0139] (2) Random Forest Risk Assessment Model

[0140] The random forest model improves the stability and accuracy of the assessment by integrating multiple decision trees, and is suitable for multi-classification of road collapse risks (low risk, medium risk, high risk). By integrating multiple decision trees, the random forest can effectively improve the stability of the model, reduce the deviation of a single decision tree, and is suitable for road collapse risk prediction after multi-source data fusion.

[0141] (3) BP neural network model

[0142] BP neural network is a nonlinear model used for comprehensive risk assessment. Through iterative training of forward propagation and back propagation, it can learn complex features and nonlinear relationships. It is suitable for complex systems such as road collapse that are affected by multiple factors.

[0143] The forward propagation formula is:

[0144] a [l] =g [l] (W [l] a [l-1] +b [l] )

[0145] Among them, a [l] is the activation value of the lth layer, g [l] is the activation function of the lth layer, W [l] The weight matrix of layer l, b [l] Bias vector for layer l.

[0146] The loss for regression problems is calculated by the mean squared error (MSE):

[0147]

[0148] Among them, J(W,b) is the loss function, y (i) is the true value of the i-th sample, The predicted value of the i-th sample, m is the total number of samples.

[0149] Backpropagation is used to calculate the gradient of the loss function with respect to the network parameters and update the parameters by gradient descent:

[0150]

[0151] Among them, α is the learning rate, is the partial derivative of the loss function with respect to the weight.

[0152] S42. Train the road collapse risk assessment model through a random deep forest model, a support vector machine or a deep neural network model.

[0153] (1) Random Forest Model

[0154] Random forest is an ensemble learning method that improves the predictive performance of the model by combining multiple decision trees. Decision tree node splitting criteria (based on Gini impurity):

[0155] Gini Impurity:

[0156]

[0157] Among them, G(D) is the Gini impurity of the node data set D, p i is the proportion of the i-th category, and m is the total number of categories.

[0158] The final prediction formula of random forest is:

[0159]

[0160] in, is the predicted value, T is the total number of trees, h t (X) is the predicted value of the tth tree.

[0161] (2) Support Vector Machine (SVM)

[0162] Support vector machines are used for classification and regression by finding an optimal hyperplane to separate different classes.

[0163] Optimal hyperplane formula:

[0164] f(x)=w T x+b

[0165] Among them, f(x) is the classification function, w is the normal vector of the hyperplane, x is the input feature vector, and b is the bias term.

[0166] Optimization goal:

[0167]

[0168] Under the constraints:

[0169] y i (w T x i+b)≥1,i=1,…,n

[0170] Among them, y i is the label of sample i, x i is the feature vector of sample i.

[0171] (3) Deep Neural Network (DNN)

[0172] Deep neural networks are used to learn nonlinear relationships from complex multi-source data.

[0173] The forward propagation formula of the neural network is:

[0174] a [l] =g [l] (W [l] a [l-1] +b [l] )

[0175] Among them, a [l] is the activation value of the lth layer, g [l] is the activation function of the lth layer, W [l] is the weight matrix of the lth layer, b [l] Bias vector for layer l.

[0176] Loss function (based on mean square error):

[0177]

[0178] Among them, J(W,b) is the loss function, y (i) is the true value of the i-th sample, The predicted value of the i-th sample, m is the total number of samples.

[0179] S43. The performance of the road collapse risk assessment model was evaluated through different data sets to ensure its effectiveness in practical applications, and the accuracy and stability of the model were improved by adjusting the model parameters.

[0180] (1) Cross-Validation

[0181] Cross-validation is used to evaluate the generalization ability of the model, usually using the k-fold cross-validation method:

[0182]

[0183] Among them, Acc cv is the average accuracy of cross validation, k is the number of folds, Acc i is the verification accuracy of the i-th fold.

[0184] (2) Hyperparameter Optimization - Grid Search

[0185] Objective function: Select the hyperparameter combination that optimizes the model performance indicators (such as accuracy, F1 score).

[0186] θ * = argmaxAcc cv (θ)

[0187] Among them, θ is the hyperparameter of the model, Θ is the candidate hyperparameter set, Acc cv (θ) Cross-validation accuracy under hyperparameter θ.

[0188] (3) Model evaluation indicators

[0189] The performance of the model is evaluated using the following metrics:

[0190] ①Accuracy:

[0191]

[0192] Among them, TP is the number of true positive examples, TN is the number of true negative examples, FP is the number of false positive examples, and FN is the number of false negative examples.

[0193] ②F1 score:

[0194]

[0195] Among them, the accuracy Recall

[0196] Through the above steps, the road collapse risk assessment model can effectively integrate multi-source data, and build a risk prediction model with high accuracy and robustness through feature selection, model training, verification and optimization. The model can accurately predict the probability of road collapse, providing an important scientific basis for geological disaster prevention and decision support.

[0197] The trained road collapse risk assessment model f() can output the probability of road collapse risk (i.e., risk assessment score R, R∈[0,1]) according to the feature extraction result X:

[0198] R = f(X)

[0199] S5. Generate the corresponding road collapse risk level according to the probability of road collapse risk, and visualize it through GIS technology.

[0200] The risk level is divided into three main categories: low risk, medium risk and high risk. According to the probability of road collapse risk, the classification criteria of road collapse risk level are determined by fuzzy logic model or analytic hierarchy process (AHP). Combined with GIS (Geographic Information System) technology, the risk level of the assessment area is visualized on the map, and a risk level distribution map is generated. The low, medium and high risk areas are marked with different colors to intuitively display the collapse risk level of different areas.

[0201] For example, the grading criteria could be:

[0202]

[0203] In a preferred embodiment, the risk level classification standard is continuously optimized and updated by tracing the source of the collapse events that have occurred. Through the above dynamic level classification mechanism and feedback loop mechanism, the road collapse risk level classification standard can be adaptively adjusted in real time, thereby achieving dynamic classification and improving the accuracy and real-time performance of risk assessment.

[0204] In a preferred embodiment, the risk level of road collapse is input into the economic loss model, the corresponding total economic loss is calculated and visualized through GIS technology, which further reflects the impact of road collapse risk on infrastructure, buildings and public services.

[0205] The expression of the economic loss model is as follows:

[0206]

[0207] Where L is the estimated total economic loss; C i A is the unit value of the i-th type of affected object; i is the affected area or volume of the i-th type of object; R i It is the loss ratio corresponding to the risk level, indicating the expected loss ratio under a certain risk level (such as 1.0 in the high-risk area, indicating complete loss).

[0208] Then, overlaying the economic impact intensity of different areas on the risk map with different colors or transparency can help relevant departments better understand the potential economic losses and formulate targeted response measures.

[0209] The above method can not only conduct a comprehensive assessment of the risk of road collapse through the comprehensive utilization of spatiotemporal data characteristics, but also classify the risks of different areas according to the assessment results, provide a scientific basis for relevant departments, support disaster prevention and control and urban planning, and is suitable for high-risk areas for road collapse such as urban areas, underground engineering construction areas, and mining areas, and has important application value.

[0210] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0211] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. A comprehensive assessment method for road collapse risk based on multi-source data fusion, characterized in that: The following steps are involved: Collect and pre-process multi-source data related to road collapse, including remote sensing image data, groundwater level monitoring data, geological exploration data and surface deformation monitoring data; Performing spatiotemporal feature fusion and multi-level data fusion on the pre-processed multi-source data, wherein the spatiotemporal feature fusion includes spatial registration and time alignment, and the multi-level data fusion includes low-level data fusion and high-level data fusion; Feature extraction is performed on the fused multi-level data, and the feature extraction results are input into the road collapse risk assessment model built based on the machine learning model to output the probability of road collapse risk occurrence, thereby generating the risk level of road collapse.

2. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The pretreatment specifically comprises the following steps: Use the statistical method Z-score analysis to identify outliers, remove or replace the identified outliers; use interpolation or statistical methods to complete missing data; Noise filtering through filtering technology; Using GIS technology to convert multi-source data related to road collapse into a unified geographic coordinate system, and using a benchmark method or a registration algorithm to perform spatial registration on the remote sensing image data; The Gauss-Krüger projection coordinate transformation method is used to transform the data into a unified reference coordinate system and perform elevation datum transformation; Perform normalization processing; The principal component analysis method is used for dimensionality reduction.

3. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The multi-source data are spatially registered using a linear transformation model, and the multi-source data with different acquisition frequencies are resampled to achieve the time alignment.

4. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The low-level data fusion is performed by a weighted average method, and the weights used in the weighted average method are set according to the reliability of the data source; the high-level data fusion is achieved by a principal component analysis method.

5. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The specific process of feature extraction for the fused multi-level data is as follows: The mutual information method was used for feature selection, followed by further screening by recursive feature elimination.

6. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The machine learning models include logistic regression models, random forest models and BP neural network models.

7. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The risk levels of road collapse include low risk, medium risk and high risk. The classification criteria of road collapse risk levels are determined by fuzzy logic model or hierarchical analysis method.

8. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: The standards for classifying road collapse risk levels are dynamically adjusted based on historical and real-time road collapse occurrences.

9. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1 is characterized in that: Generate a risk map based on the risk level of road collapse and display it visually through GIS technology.

10. The method for comprehensive assessment of road collapse risk based on multi-source data fusion according to claim 1, characterized in that: The risk level of road collapse is input into the economic loss model, and the corresponding total economic loss is calculated and visualized through GIS technology. The expression of the economic loss model is as follows: Where L is the estimated total economic loss; C i A is the unit value of the i-th type of affected object; i is the affected area or volume of the i-th type of object; R i It is the loss ratio corresponding to the risk level, indicating the expected loss ratio under a certain risk level.

Images