Machine learning driven landslide impact building vulnerability efficient probability evaluation method and product

Abstract

The invention provides a machine learning driven landslide impact building vulnerability efficient probability evaluation method and product, relates to the technical field of artificial intelligence, and uses a nonlinear Pushover to analyze and evaluate different limit state resistance of buildings with different structural forms. And building a building resistance prediction data set through multi-parameter combination simulation, and training a machine learning prediction model. A smooth particle flow and finite element coupling dynamic model is used for analyzing the impact force borne by a building under the action of different landslide impact strengths, and a mathematical calculation model of the landslide impact force is constructed based on impact force data. And establishing infilled wall damage judgment conditions, comparing the landslide impact force with the building structure resistance and the infilled wall resistance, and determining the damage state of the building. The uncertainty of landslide impact strength and building material parameters is considered, the probability vulnerability of the building is calculated, a vulnerability contour map of buildings of different structural forms is constructed, and theoretical and technical support is provided for landslide quantitative risk evaluation, disaster prevention and reduction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a machine learning-driven efficient probabilistic assessment method and product for landslide impact vulnerability of buildings. Background Art

[0002] While the SPH-FEM (Smooth Particle Flow Dynamics) coupled numerical model can perform detailed vulnerability assessments of buildings, its complex modeling and high computational cost make it difficult to effectively assess the vulnerability of groups of buildings. In practical engineering projects, landslides often affect groups of buildings, not just individual structures. Therefore, it is necessary to conduct a vulnerability analysis of groups of RC (reinforced concrete) buildings impacted by landslides. This analysis can determine the vulnerability of individual buildings within a group and rank the vulnerability of each building within the group, thereby determining prevention and control priorities or conducting further, more detailed analysis.

[0003] Due to differences in construction age, usage, and design specifications, the structural forms of RC buildings in the study area can vary significantly, such as varying storey heights and numbers of floors, and varying span lengths and numbers of spans. Developing an efficient vulnerability assessment method for group buildings, tailored to these diverse structural forms, is an urgent challenge. Summary of the Invention

[0004] The present invention provides a machine learning-driven efficient probabilistic assessment method and product for landslide impact vulnerability of buildings to at least partially address the above-mentioned problems.

[0005] A first aspect of the present invention provides a machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability, the method comprising:

[0006] Determine the ultimate resistance of buildings of different structural forms based on NSPA analysis, and construct a building structure resistance prediction data set through multi-parameter combination simulation. The building structure resistance prediction data set includes: building structure geometric parameters, building structure mechanical parameters, landslide impact height, and the corresponding building structure resistance;

[0007] Training a machine learning model based on the building structure resistance capacity prediction data set to obtain a building structure resistance capacity prediction model;

[0008] Based on the SPH-FEM coupled dynamics model, the impact force on the building structure under different landslide impact heights and landslide impact velocities was analyzed, an impact force dataset was constructed, and a mathematical calculation model for landslide impact force was constructed based on the impact force dataset;

[0009] Determine the resistance capacity of the infill wall based on the failure mode of the infill wall;

[0010] Determining the resistance of the building structure based on the random sampling values of the landslide impact height, the landslide impact velocity, the geometric parameters of the building structure, and the mechanical parameters of the building structure, and determining the landslide impact force based on the mathematical calculation model of the landslide impact force;

[0011] Compare the landslide impact force with the resistance of the building structure and the resistance of the filling wall to determine the damage status of the building;

[0012] The damage state determination method for the building is used to calculate the damage state corresponding to each random sample, the probabilistic vulnerability of the building is determined based on the Monte Carlo principle, and the vulnerability contour maps of buildings with different structural forms are constructed.

[0013] A second aspect of the present invention provides a machine learning-driven efficient probabilistic assessment device for landslide-impacted building vulnerability, the machine learning-driven efficient probabilistic assessment device for landslide-impacted building vulnerability comprising:

[0014] A dataset construction module is used to determine the ultimate resistance capacity of buildings of different structural forms based on NSPA analysis, and to construct a building structure resistance capacity prediction dataset through multi-parameter combination simulation. The building structure resistance capacity prediction dataset includes: building structure geometric parameters, building structure mechanical parameters, landslide impact height, and the corresponding building structure resistance capacity;

[0015] A training module is used to train a machine learning model based on the building structure resistance capacity prediction data set to obtain a building structure resistance capacity prediction model;

[0016] A data calculation model construction module is used to analyze the impact force on the building structure under different landslide impact heights and landslide impact velocities based on the SPH-FEM coupled dynamic model, construct an impact force data set, and construct a landslide impact force mathematical calculation model based on the impact force data set;

[0017] A first determining module is used to determine the resistance capacity of the infill wall according to the failure mode of the infill wall;

[0018] a second determination module, configured to determine the resistance of the building structure based on the randomly sampled values of the landslide impact height, the landslide impact velocity, the building structure geometric parameters, and the building structure mechanical parameters, based on the building structure resistance prediction model, and to determine the landslide impact force based on the mathematical calculation model of the landslide impact force;

[0019] The third determination module is used to compare the landslide impact force with the resistance capacity of the building structure and the resistance capacity of the filling wall to determine the damage state of the building;

[0020] The vulnerability contour map construction module uses the damage state determination method of the building to calculate the damage state corresponding to each random sample, determines the probabilistic vulnerability of the building based on the Monte Carlo principle, and constructs vulnerability contour maps of buildings with different structural forms.

[0021] The third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts as described in the first aspect of the present invention.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts as described in the first aspect of the present invention.

[0023] The fifth aspect of the present invention provides a computer program product, comprising a computer program / instruction, which is used by a processor to implement the steps of the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability as described in the first aspect of the present invention.

[0024] The present invention uses nonlinear static pushover analysis (NSPA) to rapidly evaluate the different limit state resistance capabilities of buildings with different structural forms. A building resistance prediction dataset is constructed through multi-parameter combination simulation, and a machine learning prediction model is trained based on the dataset. The SPH-FEM coupled dynamic model is used to analyze the impact force exerted on the building under different landslide impact intensities, and a mathematical calculation model for the landslide impact force is constructed based on the impact force data. According to the infill wall failure mode, a discrimination condition for infill wall damage is established. The landslide impact force is compared with the building structure resistance and the infill wall resistance to determine the damage state of the building. Taking into account the uncertainty of landslide impact intensity and building material parameters, the probabilistic vulnerability of the building is calculated, and then vulnerability contour maps of buildings with different structural forms are constructed, providing theoretical and technical support for work such as landslide quantitative risk assessment and disaster prevention and mitigation. The impact force calculation model and machine learning model constructed by the present invention are both mathematical models with very high computational efficiency and can evaluate the vulnerability of buildings with different structural forms, thereby achieving efficient probabilistic evaluation of the vulnerability of a group of buildings.

[0025] In actual landslide risk assessments, the technical solution provided by this invention can be used to individually evaluate the vulnerability of buildings within a landslide-affected area, thereby ranking their vulnerability and determining which buildings are most vulnerable to damage, allowing for the implementation of refined preventive measures. Vulnerability assessments can deeply identify weaknesses in building layout and structural design, optimizing the resulting design and enhancing its resilience to landslide impacts. In other words, vulnerability curves can help optimize the structural safety of critical buildings and provide a basis for the overall assessment and management of building damage risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the description of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 This is a flowchart of the steps of the machine learning-driven efficient probabilistic assessment method for landslide impact building vulnerability provided by the present invention;

[0028] Figure 2 Schematic diagram of a vulnerability analysis model for a group of RC buildings using the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts provided by the present invention;

[0029] Figure 3 Schematic diagram of fiber cross-sections of reinforced concrete beams and columns for the machine learning-driven efficient probabilistic assessment method for building vulnerability to landslide impacts provided by the present invention;

[0030] Figure 4 Schematic diagram of the lateral resistance capacity of RC building structures in different limit states based on the machine learning-driven efficient probabilistic evaluation method for building vulnerability to landslide impact provided by the present invention;

[0031] Figure 5 is a frequency histogram of different random variables in the input data set of the machine learning driven efficient probabilistic assessment method for landslide impact building vulnerability provided by the present invention;

[0032] Figure 6 is a frequency histogram of the lateral resistance capacity of a building structure according to the machine learning driven efficient probabilistic evaluation method for building vulnerability to landslide impact provided by the present invention;

[0033] Figure 7 This is a schematic diagram of the basic process of optimizing the hyperparameters of the machine learning model by BO-TPE in the machine learning-driven efficient probabilistic assessment method for the vulnerability of buildings to landslide impacts provided by the present invention;

[0034] Figure 8 Schematic diagram showing comparison results between predicted values and actual values of test samples by six machine learning models in the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability provided by the present invention;

[0035] Figure 9 3. It is a schematic diagram showing the importance ranking results of different characteristic variables of the prediction model of the efficient probabilistic evaluation method for landslide impact building vulnerability driven by machine learning provided by the present invention;

[0036] Figure 10 The present invention provides a machine learning-driven efficient probabilistic assessment method for landslide impact on building vulnerability, and an SPH-FEM coupled numerical model for landslide moving material impacting reinforced concrete columns;

[0037] Figure 11 90 sets of exemplary impact force data calculated by the SPH-FEM coupled numerical model of the machine learning-driven efficient probabilistic assessment method for landslide impact on buildings provided by the present invention;

[0038] Figure 12 70 sets of power law function fitting results of the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability provided by the present invention;

[0039] Figure 13 2. It is a schematic diagram showing comparison results of impact force results calculated by SPH-FEM coupled numerical simulation of the machine learning driven efficient probabilistic evaluation method for landslide impact on buildings provided by the present invention;

[0040] Figure 14 Schematic diagram showing the comparison results between a landslide maximum impact force calculation model established by the machine learning-driven efficient probabilistic assessment method for landslide impact building vulnerability provided by the present invention and a dynamic impact force calculation model fitted with measured debris flow data;

[0041] Figure 15 Schematic diagram of the relationship between the limit state and the probabilities of different damage states in the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts provided by the present invention;

[0042] Figure 16 These are the vulnerability contour maps of four exemplary RC buildings with different structural forms in the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability provided by the present invention. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] In the embodiment of the present invention, an efficient probabilistic assessment method for the vulnerability of group buildings is proposed based on NSPA analysis, machine learning, and reliability methods. Specifically, Figure 1 , which shows a flowchart of the steps of a machine learning-driven efficient probabilistic assessment method for landslide impact building vulnerability provided by an embodiment of the present invention, the method comprising:

[0045] S101, based on NSPA analysis, determine the ultimate limit state resistance capacity of building structures of different structural forms, and construct a building structure resistance capacity prediction data set through multi-parameter combination simulation. The building structure resistance capacity prediction data set includes: building structure geometric parameters, building structure mechanical parameters, landslide impact height, and corresponding building structure resistance capacity.

[0046] The embodiment of the present invention considers the building geometric parameters as random variables to construct RC buildings of different structural forms, thereby studying the vulnerability of group RC buildings. The vulnerability analysis model of group RC buildings is as follows: Figure 2 shown. Figure 2 The vulnerability analysis model of a group of RC buildings is shown, where part (a) represents the plan view of the frame structure; part (b) represents the AA section of the frame structure and the action mode of the landslide impact load.

[0047] In this embodiment of the present invention, the ranges of values for building structural geometric parameters and building structural mechanical parameters can be determined based on the actual conditions of the current landslide study area. The specific building structural geometric parameters for the RC building in the exemplary study area of the present invention are: the number of spans is 1 to 12, the number of floors is 1 to 9, the span is 5 to 9 meters with a 1-meter interval, and the floor height is 3 to 6 meters with a 0.3-meter interval. The building materials are composed of C30 grade concrete and HRB335 steel bars. The infill walls are constructed of bricks using M7.5 cement mortar.

[0048] Due to the structural symmetry of RC buildings, the central frame of the building structure can be selected to construct a two-dimensional frame model for vulnerability analysis. In order to improve the computational efficiency of numerical simulation and facilitate the construction of large-scale machine learning data sets, the embodiment of the present invention uses the central frame of the frame structure ( Figure 2 AA section) in OpenSees to establish 2D frame analysis models of various structural forms. Based on the combination of different geometric parameters, a total of 12 (number of spans) × 5 (span distances) × 9 (number of floors) × 11 (story heights) = 5940 2D frame models were considered in the vulnerability analysis of group buildings. To account for the building's deadweight and live loads, the dead and live loads on each floor were each assumed to be 4.5 kN / m. 2 and 2kN / m 2, and applied to a two-dimensional frame model. Furthermore, the numerical simulations do not consider soil-foundation-structure interactions; the building foundation is assumed to be fixed to the rigid ground. Given that the walls of RC buildings primarily serve as enclosures and partitions, for simplicity, they are not modeled in the OpenSees model. In other words, the OpenSees numerical simulations only analyze the resistance of RC building structures.

[0049] Research on the impact direction shows that the vulnerability of the forward impact scenario is greater than that of the oblique impact scenario, and the difference between the two is not much. In actual engineering, in order to ensure a reasonable safety margin, the evaluation results of the most dangerous scenario are often used as the final result. Therefore, in the vulnerability analysis of group RC buildings, only the forward impact scenario is considered, that is, the landslide impact force acts on the two-dimensional frame model of the building structure along the y direction of the plane frame. Because the infill wall is not considered, the impact force F on the two-dimensional frame can be calculated as the sum of the impact force F1 on the side column at the impact surface and the impact force F2 on the x-direction beam at the impact surface. Assuming that the impact pressure on the building is uniformly distributed along the impact surface, using the dynamic impact force calculation model, the maximum impact force F on the two-dimensional frame model is calculated as:

[0050] F=αρv 2 A (1)

[0051] In formula (1), αρv 2 is the impact pressure; ρ is the density of the moving material of the landslide; v is the impact velocity of the landslide; α is an empirical coefficient that needs to be determined through calibration; A is the impact area of the landslide, which is the sum of the impact areas A1 and A2 of F1 and F2. Among them, A1 is the impact area of F1, calculated as the product of the impact height h and the column width (0.5m); A2 is the impact area of the x-direction beam at the impact surface, calculated as the height of the beam (0.5m) multiplied by the width of the central frame, and then multiplied by the number of beams covered by the impact height (the value of the impact height divided by the floor height is rounded down). Among them, F1 is evenly distributed along the impact height h, while F2 acts on the two-dimensional frame model in the form of a concentrated load, see Figure 2 .

[0052] In the probabilistic quantitative assessment method for RC building vulnerability proposed in an embodiment of the present invention, RC buildings are classified into different damage states based on multiple limit states. The key to assessing the vulnerability of buildings to landslide impacts lies in determining the damage state of the building under the impact of a landslide. According to relevant research describing different building damage states, minor damage to a building is classified based on infill wall damage, while moderate to complete damage requires classification based on the extent of structural damage. Therefore, once the landslide impact load distribution pattern is determined, the lateral resistance of the building structure under this load distribution pattern can be further calculated through NSPA analysis. By comparing the lateral resistance of the building structure with the landslide impact force, the damage state of the building structure is determined, thereby classifying the building from moderate to complete damage. Finally, combined with the criteria for determining infill wall damage, the final damage state of the building is determined.

[0053] Opensees has been widely used in the vulnerability assessment of RC building structures because it can effectively describe the nonlinear characteristics of both the overall and local components of a building structure. In Opensees, distributed plastic elements (including force-based nonlinear beam-column elements and displacement-based nonlinear beam-column elements) are commonly used to model reinforced concrete beams and columns. Of these, the computational accuracy of displacement-based nonlinear beam-column elements is linked to the number of element divisions, so the optimal element division must be determined before finite element modeling, which increases the complexity of finite element modeling. Force-based nonlinear beam-column elements (nonlinerBeamColumn elements) not only allow a single element to be used to model a component, but also ensure the computational accuracy of structural finite element analysis. Therefore, in an embodiment of the present invention, force-based nonlinear beam-column elements are used to perform finite element modeling of reinforced concrete beams and columns. The force-based nonlinear beam-column element determines the section resistance and section stiffness matrices of each element's control section and calculates the resistance and tangent stiffness matrices of the entire element using the Gauss-Lobatto integration method along the length of the beam. Therefore, determining a reasonable number of Gauss-Lobatto integration points is key to ensuring the computational accuracy of the force-based nonlinear beam-column element. Generally speaking, the more Gauss-Lobatto integration points, the more accurate the calculation results. However, after reaching a certain number of integration points, further additions will not improve the calculation accuracy and will instead lead to poor convergence of the finite element analysis. Therefore, this paper selects the nonlinerBeamColumn element with four integration points to simulate beam-column components.

[0054] The beam-column section is simulated using fiber units, that is, the beam-column section is divided into multiple small fiber units including steel fibers and concrete fibers, and then the mechanical properties of the corresponding fibers are expressed by the stress-strain relationship of each material. Figure 3 As shown, Figure 3 The fiber sections of reinforced concrete beams and columns are shown, where part (a) represents the fiber section of the column; part (b) represents the fiber section of the beam. Taking into account the computational efficiency and convergence of the numerical model, the Concrete01 model based on the uniaxial Kent-Scott-Park concrete constitutive relationship in OpenSees is used to simulate the stress-strain behavior of concrete. The Concrete01 material model has four parameters related to concrete properties: peak strength, peak strain, residual strength, and residual strain. Because of the use of fiber sections, the concrete can be divided into whether it is constrained or not on the fiber section, and the reinforcing effect of the stirrup constraint on the axial compressive strength of concrete can be simulated by assigning different mechanical parameters. Therefore, in the numerical model, there are a total of eight parameters of concrete, namely the peak strength f of the core concrete, cp,core , peak strain ε cp,core , residual strength f cr,core and residual strain ε cr,core and the peak strength of the outer concrete f cp,cover , peak strain ε cp,cover , residual strength f cr,cover and residual strain ε cr,cover The uniaxial Giuffre-Menegotto-Pinto material constitutive relation with isotropic strain hardening, that is, the Steel02 material model in OpenSees, is used to simulate the stress-strain behavior of steel bars. The characteristics of this material model are that the simulation results are highly consistent with the results of repeated loading tests, the influence of the strain hardening phenomenon of steel bars on the constitutive relation of steel bars can be considered, and the calculation efficiency is high. The Steel02 material model has six parameters: yield strength f r , elastic modulus E r , strain hardening rate, and three shape parameters (R0, CR1, and CR2) that control the transition of the constitutive curve from the elastic to the plastic phase.

[0055] According to relevant research, among the above parameters of steel bars and concrete, ε cp,cover 、f cr,cover, strain hardening rate, R0, CR1 and CR2 are considered as deterministic parameters, which are 0.002, 0, 0.01, 20, 0.925 and 0.15 respectively, and the other parameters are considered as random variables obeying the log-normal distribution. Referring to relevant research and concrete structure design specifications, the statistical information of random variables in HRB335 steel bars and C30 concrete materials are listed in Table 1. In addition, the geometric information of the span, number of spans, storey height and number of floors of the building structure are considered as random variables obeying discrete uniform distribution to simulate the lateral resistance of different building structures. In Pushover analysis, the landslide impact force is applied to the structure along the impact height in a certain distribution pattern, and then the load is applied to control the gradually monotonically increasing lateral force. When the structure reaches the bearing capacity limit, the load is the lateral resistance of the building structure in the ultimate state. Therefore, when using Pushover analysis to determine the lateral resistance of the building, it is only necessary to determine the distribution of the landslide load along the impact surface and an arbitrary initial load value, without the need to input the accurate landslide impact load value. According to Figure 2 As shown in the schematic diagram of the impact load distribution, the distribution of the landslide load along the impact surface is determined solely by the landslide impact height h, since it is considered uniformly distributed. Therefore, the landslide impact intensity h is also considered a random variable to investigate the lateral resistance of buildings under different impact heights. Referring to the inversion simulation results of a landslide movement process at a certain location, the impact height in the accumulation zone was less than 30 meters throughout the entire movement process. Therefore, the impact height was considered a uniformly distributed random variable ranging from 0 to 30 meters to construct a machine learning prediction model that is valid for impact heights within this range. Therefore, a total of 13 random variables were considered in the evaluation of the lateral resistance of RC building structures. The specific statistical information is summarized in Table 1.

[0056] Table 1 Statistics of random variables

[0057]

[0058] In order to determine the lateral resistance of RC building structures in different limit states, to judge the damage of building structures in combination with the magnitude of landslide impact force, and to establish the basis for dividing RC buildings from moderate damage to complete damage, it is first necessary to use NSPA analysis to obtain the Pushover curve of the building. After the two-dimensional frame finite element model of the RC structure is established, the dead load and live load are applied to the building finite element model for initial gravity analysis, and then the load is applied along the impact height. Figure 2The lateral load distribution shown in FIG. 1 is subjected to NSPA analysis to obtain the Pushover curve of the building structure. The NSPA analysis adopts a displacement loading control mode, in which the horizontal displacement of the top displacement node is selected as the control point. A total of 3000 steps are analyzed, and the displacement increment is 0.1 mm per step. The Pushover curve usually represents the relationship between the lateral force of the building structure and the maximum inter-story displacement angle β, and is used to evaluate the ability of the building structure to resist lateral impact. Therefore, in the embodiment of the present invention, three limit states are defined using the Pushover curve to reflect the performance state of the building structure, which are respectively recorded as SLS1, SLS2 and SLS3, see Figure 4 , Figure 4 The figure shows the lateral resistance of RC building structures in different limit states. According to relevant research, the three limit states are defined as follows: SLS1 represents the end of the linear portion of the Pushover curve, which is the first time that the global performance of the building structure undergoes a significant change; SLS2 is the structural yield point determined by the Equvialent Elasto-Plastic Energy Method, which is the point at which the structural yield point is obtained. Figure 4 The area of S1 is equal to the area of S2; SLS3 is at the peak point of the Pushover curve, which means that the structure has reached its bearing capacity limit and may collapse. The lateral force corresponding to each limit state of the building structure is the ability of the building structure to resist the lateral impact of the landslide, R i (i = 1, 2 and 3). Because the three determined limit state points reflect the ability of RC building structures to resist lateral impact when different degrees of damage occur, they can serve as an important basis for judging the damage state of RC buildings.

[0059] First, in order to provide a data set to train a machine learning model that replaces the numerical model for efficient prediction of the lateral resistance of building structures, a large number of NSPA analyses are required to calculate the lateral resistance (R1, R2, and R3) of different building structures under different parameter combinations. Since in the group RC building vulnerability analysis model, it is necessary to consider building structures with different spans, number of spans, number of floors, and floor heights (a total of 5940 structural forms), and different structural forms require the re-establishment of its finite element model. Since thousands of calculations are required, it is impossible to use traditional manual methods for modeling, calculation, and data extraction. To this end, an embodiment of the present invention develops an automatic modeling and calculation program for the OpenSees frame structure model based on the Python language, including geometric modeling of beam-column structures, assigning boundary conditions, beam-column section division, material definition and assignment, frame structure gravity analysis, pushover analysis, extraction of lateral forces and maximum inter-story displacement angles of the structure, drawing pushover curves, defining limit state points based on the pushover curves, and outputting the lateral resistance of the building structure.

[0060] Secondly, the MCS method was used to sample the probability distribution of the random variables (Table 1) involved in the lateral resistance analysis of group RC building structures 100,000 times to obtain the input database for NSPA analysis. Because it is a random sampling, there may be samples in the database where the landslide impact height is greater than the building height. In this case, the building is considered to be completely buried, and the building is directly defined as a completely damaged state, without the need for Pushover analysis to determine the lateral resistance of the building structure to further judge its damage state based on the landslide impact force. Therefore, these samples where the landslide impact height is greater than the building height were eliminated from the 100,000 sampled input data. After elimination, a total of 65,892 samples were retained to form the input data set, and their frequency histograms are shown in Figure 2. Figure 5 , Figure 5 shows the frequency histogram of different random variables in the input dataset.

[0061] Then, these 65,892 input data were assigned to the Python automated analysis program in sequence to calculate the lateral resistance of the building structure (R1, R2 and R3) corresponding to the 65,892 samples. This means that 65,892 NSPA numerical simulations were required. The frequency histogram of the lateral resistance of the building structure is shown in Figure 6 , Figure 6 A frequency histogram of the lateral resistance of RC building structures at different limit states is shown, with statistical information in Table 2. It can be seen that the mean values of R2 and R3 are relatively close. This is primarily due to the fact that after yielding, the increase in lateral force rapidly increases structural deformation, leading to a rapid peak load-bearing capacity. Finally, the input dataset was fused with the output dataset of RC building structure lateral resistance to form a training dataset with a sample size of 65,892, which was used for training and optimizing the machine learning model.

[0062] Table 2 Statistical information on lateral resistance of RC building structures in different limit states

[0063] Limit state Mean (kN) Standard deviation (kN) Maximum value (kN) Minimum value (kN) <![CDATA[SLS1:R1]]> 561.31 380.77 3606.413 11.92 <![CDATA[SLS2:R2]]> 1255.95 773.69 6381.104 31.14 <![CDATA[SLS3:R3]]> 1408.51 886.87 7657.021 39.93

[0064] Therefore, the embodiment of the present invention proposes that step S101 specifically includes:

[0065] S1011: Establish a two-dimensional frame finite element model of the building structure. Apply dead load and live load to the finite element model to perform an initial gravity analysis. Apply lateral load distribution along the impact height to perform NSPA analysis to obtain the pushover curve of the building structure.

[0066] S1012: Use the Pushover curve to determine three limit states to reflect the performance status of the building structure. The three limit states include SLS1, SLS2, and SLS3. SLS1 represents the end of the linear portion of the Pushover curve, indicating the first significant change in the global performance of the building structure. SLS2 is the structural yield point determined by the equal energy method. SLS3 is located at the peak point of the Pushover curve, indicating that the structure has reached its bearing capacity limit.

[0067] S1013, determine the lateral force corresponding to each limit state as the building structure's ability to resist the lateral impact of the landslide;

[0068] S1014, using the MCS method to perform multiple sampling on the probability distribution of random variables involved in the lateral resistance analysis of RC building structures to obtain the input database for NSPA analysis;

[0069] The random variables include: yield strength and elastic modulus of steel bars, peak strength, peak strain, residual strength and residual strain of core concrete, and peak strength and residual strain of outer concrete; number of spans, span length, number of floors, floor height, and landslide impact height.

[0070] S1015, assigning the input database to the Python automated analysis program in sequence, and calculating the lateral resistance of the building structure corresponding to each input data: R1, R2 and R3;

[0071] S1016, finally, the input data set is fused with the building structure lateral resistance capacity output data set to form a building structure resistance capacity prediction data set.

[0072] S102: Training a machine learning model based on the building structure resistance capacity prediction data set to obtain a building structure resistance capacity prediction model.

[0073] In an embodiment of the present invention, for the dataset constructed using the above method, six machine learning models (DT, RF, MLP, XGBoost, GBM, and SVM) are proposed to train a prediction model for the lateral resistance of RC building structures, and the optimal machine learning model is selected by comparing their prediction accuracy. However, the accuracy of the prediction of the machine learning model depends not only on the model itself, but also on the selection of hyperparameters. Hyperparameters refer to parameters that need to be manually set before model training, which control the model structure and learning process. Currently, grid search and random search are commonly used for hyperparameter optimization of machine learning models due to their simple principles and easy implementation. However, the main limitation of grid search and random search is that each evaluation in the iteration is independent of the previous evaluation, wasting a lot of time evaluating poorly performing areas in the search space. Therefore, this optimization method has high computational complexity, and when the search space is large and there are many hyperparameters, it takes a long time to determine the global optimal hyperparameters.

[0074] Compared to grid search and random search, Bayesian optimization (BO) can overcome these limitations and quickly obtain optimal hyperparameters. Bayesian optimization is an iterative hyperparameter optimization algorithm that uses previously evaluated objective function results to determine the next set of hyperparameters to evaluate, then iterates until convergence is achieved. For regression and classification problems, the root mean square error (RMSE) and error rate can be used, respectively, to construct the objective function for hyperparameter optimization. The goal of Bayesian optimization is to find the optimal hyperparameters that minimize the RMSE or error rate. Bayesian optimization algorithms have two core components: a surrogate model and an acquisition function. Because machine learning is generally a black-box model, lacking an explicit objective function and requiring high computational cost, a surrogate model is needed to approximate the true objective function, thereby more quickly estimating the output of the objective function at unexplored points. Common surrogate models used in Bayesian optimization include Gaussian processes (GPs), random forests (RFs), and tree-structured Parzen estimators (TPEs). Therefore, there are three main types of BO algorithms based on proxy models: BO-GP, BO-RF and BO-TPE. Because BO-TPE performs well in processing high-dimensional hyperparameter spaces, and can process discrete and continuous hyperparameters with low time complexity, the embodiment of the present invention selects BO-TPE for Bayesian hyperparameter optimization. The acquisition function makes a trade-off between developing areas near known points and exploring areas with high uncertainty based on the prediction of the proxy model to determine the most valuable evaluation points for the next iterative update of the proxy model. The BO-TPE method generally uses the expected improvement (EI) acquisition function to select the most valuable evaluation points. See the steps for using BO-TPE to optimize machine learning model hyperparameters. Figure 7 , Figure 7 The basic process of BO-TPE optimization of machine learning model hyperparameters is shown.

[0075] 80% of the training data set was randomly selected for model training (52,714 training data), and the remaining 20% (13,178 test data) was used for model testing to evaluate the accuracy of the training model. The lateral resistance prediction model of RC building structures is a multi-input (13 random variables)-multi-output (R1, R2 and R3) regression problem, and three regression models need to be constructed at the same time. Because R1, R2 and R3 are all obtained based on a Pushover curve, there must be a certain correlation between them. In order to consider this correlation to improve the prediction performance of the model, the regression chain model is used to connect the three regression models in series, and the previously predicted target variable is used as the feature of the subsequent prediction to gradually predict the target variable, that is, the prediction result R1 of the first regression model is used as the input feature to predict R2, and the prediction results R1 and R2 of the first and second regression models are used as input features to predict R3. According to the calculation process of Bayesian optimization ( Figure 5-7 ) It can be seen that before Bayesian optimization, it is necessary to determine the objective function to be optimized and the prior range of the hyperparameters to be optimized. The prediction of lateral resistance capability in the embodiment of the present invention is a regression problem, so the RMSE of the test data is selected to construct the optimization objective function, that is, to minimize the RMSE value of the model prediction. With reference to relevant research, the prior ranges of the hyperparameters of six machine learning models are summarized in Table 3. Secondly, based on the prior information of the hyperparameters and the training data, the BO-TPE optimization method is used to optimize the hyperparameters of the six machine learning models, and the prediction models with the best performance are obtained respectively. The BO-TPE optimization was iterated 200 times in total, and the optimal hyperparameters of the six different machine learning models are also listed in Table 3.

[0076] Table 3. Hyperparameter prior ranges and optimal parameters of machine learning models

[0077]

[0078]

[0079] Figure 8 A diagram showing the comparison of the predicted values and actual values of the test samples by six machine learning models is shown, and the coefficient of certainty R is used. 2 The prediction accuracy of the six machine learning models for the resistance capacity of three different limit states of building structures was evaluated. Table 4 summarizes the R 2 and RMSE. The results show that regardless of R 2 Still, RMSE shows that the SVM and DTR models have the lowest prediction accuracy for building lateral resistance among the six machine learning models, while the GBM model has the highest accuracy. 2The values for the GBM model are 0.970, 0.985, and 0.982, respectively. This indicates that the GBM model has high prediction accuracy for this dataset and can accurately predict the lateral resistance of building structures under landslide impact. Therefore, the GBM model is ultimately selected for the prediction of the lateral resistance of building structures in this embodiment of the present invention.

[0080] Table 4 Prediction accuracy of six different machine learning models

[0081]

[0082]

[0083] After the GBM model training is completed, the importance of the input parameters can be ranked by counting the number of times each feature appears as a split node in the decision tree. Figure 9 , which shows a schematic diagram of the importance ranking results of different feature variables to the prediction model. Features with a large number of splits can usually divide the data better, which means that these features can provide useful information at different nodes and help the model fit the data better, that is, they are more important to the prediction model. According to the importance ranking of the R1, R2 and R3 prediction models, it can be found that the landslide impact height is the most important for the three prediction models, indicating that the landslide impact height is a key factor affecting the lateral resistance of the building structure. Therefore, when conducting building vulnerability analysis, it is necessary to conduct more detailed research on the determination of the landslide impact height. In addition, it can be found that the prediction results of the previous regression model in the regression chain model are also very important to the next regression model, which shows that there is a strong correlation between the three regression models, and it is necessary to use the regression chain model to consider this problem.

[0084] S103, based on the SPH-FEM coupled dynamic model, analyzing the magnitude of the impact force on the building structure under different landslide impact heights and landslide impact velocities, constructing an impact force data set, and constructing a landslide impact force mathematical calculation model based on the impact force data set.

[0085] The embodiment of the present invention uses the impact force calculation formula of formula (1) to achieve efficient evaluation of landslide impact intensity. However, the empirical coefficient α of formula (1) needs to be determined by calibration. In some studies, by conducting multiple sets of indoor chute physical tests, the impact forces under different impact heights and speeds are obtained, and then the empirical coefficient α is calibrated using the test data. However, for simplicity, indoor chute tests generally use scaled model tests, which have scale effects and are difficult to simulate the impact force of landslides on buildings in real cases. The embodiment of the present invention constructs an analytical calculation model for the maximum impact force based on the SPH-FEM dynamic coupling numerical model. The SPH-FEM dynamic coupling numerical model can calculate the full-time impact force of the landslide on the building through the penalty function method. Therefore, after verifying the reliability of the impact force calculation of the SPH-FEM dynamic coupling numerical model using the physical model test, the SPH-FEM coupling method can be used to simulate the unscaled numerical model with the landslide in the study area as the background, and calculate the maximum impact force under different combinations of moving material density, impact speed and impact height, thereby calibrating the empirical coefficient α. In this way, not only the impact of scale effect can be reduced, but also the consistency of maximum impact force when NSPA numerical simulation and SPH-FEM dynamic coupling numerical simulation are used for building vulnerability assessment can be guaranteed.

[0086] In the vulnerability assessment of buildings, the impact force of landslides acts directly on reinforced concrete columns, so it is necessary to focus on analyzing the impact pressure of landslide materials on reinforced concrete columns. To this end, an SPH-FEM coupled numerical model of landslide materials impacting reinforced concrete columns under different impact height and impact velocity combinations was constructed (e.g. Figure 10 As shown). The height of the reinforced concrete column is considered to be 3.3m. The width of the landslide moving material is 1m to cover the width of the column (0.5m). The height is considered as a variable to calculate the impact force under different impact heights h, using v 2The landslide material was modeled using SPH particles with a spacing of 0.08 m. The reinforced concrete column was modeled using eight-node solid hexahedral elements with a mesh size of 0.1 m. The constitutive models and material parameters of the reinforced concrete column and landslide material were consistent with the actual mechanical parameters of landslide soil and building materials at a certain location. Deterministic parameters used deterministic values, while random variables used mean values. Similarly, the INIVEL / TRA keyword in OpenRadioss was used to assign an initial velocity (i.e., impact velocity v) to the SPH particles along the impact direction, enabling them to initiate motion and impact the reinforced concrete column. Furthermore, the INTER / TYPE7 model in OpenRadioss was used to couple SPH with FEM to calculate the impact force of the landslide material on the reinforced concrete column. To prevent the impact force from causing severe deformation of the column and resulting in non-convergence, the reinforced concrete column was considered a non-deformable body and fixed to the rigid ground. Since the only concern was the impact force of the landslide material on the reinforced concrete column, considering it as a non-deformable body was feasible.

[0087] According to relevant research, through a power law relationship, the empirical coefficient α can be calculated from the Froude number (Fr) as:

[0088]

[0089] Where a1 and a2 are the two coefficients of the power law function; Fr can be calculated from the landslide impact velocity and impact height. To calculate the empirical coefficient α, two power-law coefficients, a1 and a2, must be determined. If multiple sets of α values for different Fr values are available, the two power-law coefficients can be determined through data fitting, thereby calibrating the empirical coefficient α.

[0090] Because the impact height, impact velocity and density of the landslide moving material are related to the impact force calculation, they are considered as variables and the impact force under different parameter combinations is obtained by the SPH-FEM coupled numerical model. According to the maximum impact force F calculated by numerical simulation, the empirical coefficients α and Fr under different parameter combinations are back-calculated, and the power law coefficients a1 and a2 are determined by data fitting. Among them, the empirical coefficient α=F / (ρv 2 A), A is the impact area, which is calculated as the landslide impact height h multiplied by the column width (0.5m). For this purpose, the embodiment of the present invention uses GF-deviation at 1600kg / m of landslide material density ρ 3 ~1900kg / m 3The impact height h ranged from 0m to 3.3m, and the impact velocity v ranged from 0m / s to 10m / s. 90 uniform samplings were performed to obtain 90 sets of sample data under different combinations of density, velocity, and height. Secondly, the 90 sets of sample data were sequentially substituted into the established SPH-FEM coupled numerical model to obtain the corresponding maximum impact force ( Figure 11 , which shows 90 sets of exemplary impact force data calculated by the SPH-FEM coupled numerical model. Among the 90 sets of data obtained, 70 sets of samples were randomly selected to further calculate the empirical coefficient α and Froude number Fr (partial data are shown in Table 5) for fitting the power law coefficients a1 and a2. Figure 12 The power law function fitting results of 70 exemplary sets of α and Fr data are shown. The two power law coefficients a1 and a2 are 9.175 and -0.843 respectively. The R 2 The maximum impact force of the landslide on the building structure is calculated using the analytical formula:

[0091] F=9.175Fr -0.843 ρv 2 A (3)

[0092] Table 5 Partial fitting data of exemplary embodiment of impact force formula

[0093]

[0094]

[0095] In order to verify the accuracy of the impact force calculation, the remaining 20 groups of samples were substituted into formula (3), and the corresponding impact force was calculated using the impact force formula. The results were compared with the impact force results calculated by SPH-FEM coupling numerical simulation. Figure 13 The results show that the coefficient of certainty (R 2 ) is 0.995, indicating that the impact force calculation formula has good approximate accuracy for complex SPH-FEM coupled numerical simulations. However, compared with complex SPH-FEM coupled numerical simulations, the impact force formula is a simple analytical expression with simple and fast calculations, which can achieve efficient prediction of landslide impact intensity in group building vulnerability assessments.

[0096] Due to the sudden nature of landslides, the impact forces they exert on structures are rarely monitored, making it difficult to verify the reliability of models using field data. Related research summarized 218 impact force datasets obtained through physical model tests or field measurements and used them to develop a formula for calculating the dynamic impact force of debris flows. Figure 14The maximum impact force calculation model for landslides established in the embodiment of the present invention and the maximum impact force calculation model for debris flows fitted using measured debris flow data were compared. It can be found that the overall trends of the two models are consistent, but the α value of the impact force calculation model constructed in the embodiment of the present invention is larger at the same Fr value, which means that landslides have greater impact force than debris flows at the same Fr value. Due to the differences in material and movement characteristics between debris flows and landslides, it is inevitable that there are certain differences in the calculation of impact force between the two models, but the consistency of the overall trend shows that the maximum impact force calculation model constructed in the embodiment of the present invention conforms to the variation law of the measured impact force, and also proves the reliability of the impact force calculation model to a certain extent.

[0097] Therefore, the embodiment of the present invention proposes that step S103 includes:

[0098] S1031, performing multiple uniform samplings within a preset range of landslide material density ρ, a preset range of landslide impact height h, and a preset range of landslide impact velocity v, to obtain multiple sets of sample data under different combinations of density, velocity, and height; sequentially substituting the multiple sets of sample data into the established SPH-FEM coupled numerical model to obtain the landslide impact force corresponding to each set of sample data

[0099] S1032, randomly select multiple groups of samples to calculate the empirical coefficient α and the Froude number Fr; which are used to fit the power law coefficients a1 and a2;

[0100] S1033, determining a mathematical calculation model for landslide impact force based on the fitting results of a1 and a2 and an analytical formula for the maximum impact force of the landslide on the building structure, wherein the analytical formula is: F = αρv 2 A; Where ρ is the density of the moving material in the landslide; v is the impact velocity of the landslide; and A is the impact area of the landslide.

[0101] In the embodiment of the present invention, the preset range of the landslide material density ρ, the preset range of the landslide impact height h, and the preset range of the landslide impact velocity v are determined by the actual conditions of the current landslide study area.

[0102] The main disadvantage of using the SPH-FEM coupled numerical model to calculate impact force is the high computational cost. However, approximating the impact force calculated by the SPH-FEM coupled numerical model through the analytical calculation formula of landslide impact force can effectively avoid this disadvantage and use it for the efficient calculation of landslide impact intensity in the vulnerability assessment of group buildings.

[0103] S104: Determine the resistance capacity of the infill wall according to the failure mode of the infill wall.

[0104] In the NSPA analysis of RC buildings, infilled walls are not considered. Therefore, the limit state of the building structure (i.e., the beam-column frame structure) is defined according to the Pushover curve. According to the description of the building damage state, slight damage is described as only infilled wall damage without structural damage. Structural damage occurs when the building is damaged more severely than slightly. Therefore, the three limit states determined by the NSPA analysis of the building structure are used as the basis for dividing the RC building from moderate damage to complete damage, while whether the infilled wall is damaged is used as the basis for dividing the undamaged and slightly damaged states. Finally, after obtaining the lateral resistance capacities R1, R2, and R3 of different limit states of the building structure and the maximum landslide impact force F, according to the limit state of the building structure and whether the infilled wall is damaged, the damage states of RC buildings in the vulnerability assessment of buildings are divided as follows: If the building structure is undamaged (F < R1) and the infilled wall is undamaged, the building is in an undamaged state; if the building structure is undamaged (F < R1) and the infilled wall is damaged, the building is in a slightly damaged state; if the global performance of the building structure changes significantly for the first time but does not reach the structural yield point (R1 ≤ F < R2), the building is in a moderately damaged state; if the building structure reaches the structural yield point but does not reach the peak bearing capacity (R2 ≤ F < R3), the building is in a severely damaged state; if the building structure reaches the peak bearing capacity (F ≥ R3), the building is in a completely damaged state. In addition, when the landslide impact height h is greater than the building height H building it is considered that the building is completely buried and cannot be repaired, and it is directly defined as a completely damaged state.

[0105] In the above damage state discrimination criteria, it is also necessary to establish the judgment conditions for the infilled wall to be damaged. According to relevant research, under lateral loads, the infilled wall mainly undergoes flexural failure. Therefore, whether the wall undergoes flexural failure can be used as the judgment condition for the infilled wall to be damaged. According to relevant research, the maximum bending moment M b received by a unit-width wall can be expressed as a function of the impact velocity v and the impact height h:

[0106]

[0107] In formula (4), L is the height of the infilled wall; α is the empirical coefficient in the impact force calculation model; ρ is the density of the landslide moving material. It can be seen from formula (5-4) that the bending moment received by the wall is positively correlated with the impact velocity and the impact height, and negatively correlated with the height of the wall. According to the code for design of masonry structures, the maximum flexural capacity M c of the masonry infilled wall is estimated as:

[0108] M c = f t W (5)

[0109] In formula (5), f t is the maximum tensile strength of the masonry; W is the cross-sectional resistance moment, which is related to the width and thickness of the wall. The thickness of the filling wall in the embodiment of the present invention is considered to be 240mm, and M7.5 cement mortar is used, and its maximum bending resistance M c Using formula (5-5), it is estimated to be 1344N·m. When calculating the maximum bending moment M b Less than the maximum bending capacity M c When , the infill wall is not damaged, otherwise, it is damaged. Finally, the damage state judgment criteria for efficient building vulnerability evaluation are summarized in Table 6.

[0110] Table 6 Judgment criteria for different damage states of RC buildings for efficient building vulnerability evaluation

[0111]

[0112]

[0113] Therefore, the embodiment of the present invention proposes that step S104 includes:

[0114] S1041: Use the following formula to determine the maximum bending moment M per unit width of the infill wall based on the impact velocity v and impact height h. b :

[0115] S1042, where L is the height of the filling wall; α is the empirical coefficient in the impact force calculation model; ρ is the density of the landslide moving material;

[0116] S1043, according to the design code for masonry structures, the maximum bending capacity M of a masonry infill wall is calculated using the following formula: c :M c =f t W; where f t is the maximum tensile strength of masonry; W is the section resistance moment, which is related to the width and thickness of the wall;

[0117] S1044, the calculated maximum bending moment M b Less than the maximum bending capacity M c In the case of the infill wall is not damaged; in the calculated maximum bending moment M b Not less than the maximum bending capacity M c In this case, it is determined that the infill wall is damaged.

[0118] S105, based on the random sampling values of the landslide impact height, the landslide impact velocity, the building structure geometric parameters and the building structure mechanical parameters, based on the building structure resistance prediction model, determining the building structure resistance, and determining the landslide impact force based on the mathematical calculation model of the landslide impact force.

[0119] In the embodiments of the present invention, in the actual application process, first, the structural forms of each building in the RC building complex can be determined, such as the span and number of spans along the potential impact direction, as well as the storey height and number of storeys. Then, the random variables of the building materials and landslide impact intensity in the study area are determined, and their statistical information such as the mean value, standard deviation, and distribution type is determined. The random variables related to the properties of building materials include the peak strength, peak strain, residual strength, and residual strain of the core concrete, as well as the peak strength and residual strain of the outer concrete. According to the landslide impact force calculation model, the landslide impact intensity is related to the density of moving materials, impact velocity, and impact height. Therefore, these three parameters are considered as random variables related to the landslide impact intensity.

[0120] Use MCS to sample the probability density function of the random variables to obtain multiple random samples. Secondly, input the structural form parameters of a specific building, the random samples related to building materials, and the random samples of impact height into the trained GBM machine learning model to obtain the resistance capabilities (R1, R2, and R3) of different ultimate states of the building structure corresponding to each sample. At the same time, substitute the random samples of the density of moving materials, impact velocity, and impact height into the impact force calculation model to obtain the landslide impact force corresponding to each sample. According to the magnitude of the landslide impact force on the building structure and the resistance capabilities of different ultimate states of the building structure, combine Table 6 to judge the damage state of the building. Among them, when F < R1, it is necessary to further calculate the maximum bending moment M of the infill wall based on Equation (4) using the random samples of the landslide impact intensity b , and judge whether the infill wall is damaged according to the maximum flexural capacity M c to determine whether the building is in a lightly damaged state or an undamaged state.

[0121] S106. Compare the landslide impact force with the resistance capabilities of the building structure and the infill wall to determine the damage state of the building.

[0122] In the embodiments of the present invention, the step S106 includes:

[0123] S1061. When F < R1 and M b < M c , determine that the damage state of the building is undamaged; when F < R1 and M b ≥ M c , determine that the damage state of the building is lightly damaged; when R1 ≤ F < R2, determine that the damage state of the building is moderately damaged; when R2 ≤ F < R3, determine that the damage state of the building is severely damaged; when F ≥ R3 or h > H building , determine that the damage state of the building is completely damaged; H building represents the building height;

[0124] S1062: Determine the corresponding damage state of the building based on the relationship between the calculated landslide impact force F and the building structure resistance capacity (R1, R2, and R3), and the resistance capacity of the infill wall. The damage state includes no damage, slight damage, moderate damage, severe damage, or complete damage.

[0125] S107, using the damage state determination method for the building to calculate the damage state corresponding to each random sample, determining the probabilistic vulnerability of the building based on the Monte Carlo principle, and constructing vulnerability contour maps of buildings with different structural forms.

[0126] In the embodiments of the present invention, the concept of probabilistic analysis is introduced, and the uncertainty parameters involved in the landslide impact intensity and the building's resistance capacity are considered as random variables that obey a certain probability distribution (such as normal or lognormal random variables). Therefore, within the framework of probabilistic analysis, the evaluation results of the building's damage state are not fixed. Instead, each damage state has a possibility, and the probabilities of being in different damage states vary, but their sum is 1. Based on the concept of probability theory, considering the probability of a building being in different damage states and the loss ratio corresponding to each damage state, the expected loss level V of the building (i.e., the probabilistic vulnerability proposed in the embodiments of the present invention) can be calculated as:

[0127]

[0128] In formula (6), n = 1, 2, 3, 4, and 5; lr n is the loss ratio of the nth damage state; represents the probability that the building is in the nth damaged state.

[0129] According to the relationship between the limit state and the probability of the building being in different damage states (such as Figure 15 As shown, it shows the relationship between the limit state and the probability of different damage states), the probability of different damage states The exceedance probability of different limit states can be calculated as:

[0130]

[0131] In formula (7), The probability of exceeding LS1 in the damaged state; The probability of exceeding LS2 in the damaged state; Indicates the probability of the damage state exceeding LS3, represents the probability of the damage state exceeding LS4. Therefore, the key to the probability evaluation of building vulnerability is to estimate the exceedance probability of different limit states. In order to solve the exceedance probability, it is first necessary to define the function function G for each limit state. i (θ)(i=1, 2, 3 and 4):

[0132] G i (θ)=sign[D(θ)-D tv,i ] (8)

[0133] In formula (8), θ is the random variable vector involved in evaluating the building damage state using the numerical simulation method (including key parameters such as landslide impact velocity, impact height, landslide moving material density, and elastic modulus of reinforced concrete materials); D(θ) is the building damage state calculated by the physical mechanics numerical model. For the convenience of calculation and explanation, the five damage states of no damage, slight damage, moderate damage, severe damage, and complete damage are recorded as 1, 2, 3, 4, and 5 respectively. D tv,i Indicates LS i The threshold value of LS1 is tv,1 =1, D of LS2 tv,2 =2, LS3's D tv,3 =3, LS4's D tv,4 =4. sign[·] is G i (θ), if D(θ)>D tv,i , G i (θ)=1, otherwise G i (θ)=0. Therefore, based on the calculation principle of MCS, each LS i The exceedance probability can be estimated as:

[0134]

[0135] In formula (9), T is the number of samples of MCS, θ j is the jth sample of the simulation.

[0136] Therefore, in the embodiment of the present invention, step S107 includes the following sub-steps:

[0137] S1071, determine the exceedance probability of different limit states of the building based on the following formula: G i (θ) represents the performance function of each limit state; G i (θ)=sign[D(θ)-D tv,i ]; θ represents the random variable vector; T is the number of samples of MCS, θ j is the jth sample of the simulation; D tv,i Indicates LS i The threshold value of LS1 is tv,1 =1, D of LS2 tv,2 =2, LS3's D tv,3 =3, LS4's D tv,4 =4; sign[·] is G i(θ), if D(θ)>D tv,i , G i (θ)=1, otherwise G i (θ) = 0; the four limit states LS1, LS2, LS3, and LS4 divide the damage state of the building into five levels: no damage, slight damage, moderate damage, severe damage, and complete damage;

[0138] S1072, calculate the probability of the building being in different damage states based on the following formula:

[0139]

[0140] The probability of exceeding LS1 in the damaged state; The probability of exceeding LS2 in the damaged state; Indicates the probability of the damage state exceeding LS3, Indicates the probability of the damage state exceeding LS4; DS1 means no damage, DS2 means slight damage, DS3 means moderate damage, DS4 means severe damage, and DS5 means complete damage;

[0141] S1073, calculate the vulnerability of the building based on the following formula:

[0142] Where n = 1, 2, 3, 4 and 5; lr n is the loss ratio of the nth damage state; represents the probability that the building is in the nth damaged state;

[0143] The loss ratio is calculated as the ratio of the cost of repairing a building to the cost of rebuilding it. Based on relevant research, the loss ratios for no damage, minor damage, and moderate damage are 0.0, 0.1, and 0.4, respectively. Severe damage is the same as complete damage, meaning the building is completely unusable and cannot be repaired, so the loss ratio for both is 1.0.

[0144] The vulnerability of buildings with different structural forms at different impact velocities and impact heights is calculated in turn, and a vulnerability contour map of each building in the group of buildings is constructed.

[0145] Figure 16Vulnerability contour maps were plotted for four exemplary RC buildings with different structural forms. (a) represents a two-story, two-span RC building; (b) a four-story, four-span RC building; (c) a six-story, six-span RC building; and (d) an eight-story, eight-span RC building. The vulnerability contour maps show similar trends, but the vulnerability values for different impact velocity and impact height combinations vary. Overall, vulnerability decreases with increasing spans and floors. For the two-story, two-span building, since it is only 6.6 meters tall, when the impact height exceeds 6.6 meters, the building is completely buried and considered completely destroyed, resulting in a vulnerability of 1.00. Furthermore, it is observed that the vulnerability curves exhibit a non-smooth transition when the impact height is a multiple of the floor height (3.3 meters). This is primarily due to the increased impact force on the beams as the impact height reaches the top of each floor, resulting in a sudden increase in the impact force of the landslide on the building structure, which in turn causes a significant change in vulnerability. Building vulnerability contour maps reflect the vulnerability of RC buildings under different impact height and velocity combinations, and can guide landslide risk assessment and disaster prevention and mitigation efforts. While vulnerability curves for four different building structures are shown here as examples, the method proposed in this embodiment can efficiently construct vulnerability contour maps for any of the 5,940 considered RC building structures. These vulnerability contour maps are essentially based on Opensees calculations, and therefore still take into account the actual physical and mechanical properties of the buildings.

[0146] To facilitate the application of efficient probabilistic vulnerability assessment methods for RC buildings in engineering practice, this embodiment of the present invention has initially developed a practical program for probabilistic vulnerability assessment of RC buildings based on the Python language, along with a simple software interface. The program includes four input modules (building geometry parameters, building material parameters, landslide impact intensity parameters, and reliability analysis parameters), two calculation commands (vulnerability assessment and vulnerability map drawing), and an output window. Users can obtain statistical characteristics of RC building geometry and material parameters through field surveys, regulatory references, and literature research. The statistical characteristics of landslide impact intensity can be determined using techniques such as numerical simulation of landslide motion. After obtaining these input parameters, they are sequentially entered into the program. Clicking the "Vulnerability Assessment" button will determine the building's vulnerability under the current landslide impact intensity, with the results displayed in the output window. Furthermore, clicking the "Draw Vulnerability Map" command creates a vulnerability contour map for the currently evaluated building. When drawing the vulnerability curve, the program defaults to using the landslide impact height h in the range of 0m to 14m and the landslide impact velocity v in the range of 0m / s to 20m / s. There is no need to input additional landslide impact intensity parameters.

[0147] By designing a practical software program, engineers only need to obtain the input parameters for building vulnerability assessment, eliminating the need for complex numerical modeling and reliability analysis. This makes it more user-friendly in practical applications (especially in the vulnerability assessment of group buildings). This is because OpenSees considers the physical and mechanical properties of buildings when analyzing their lateral resistance, and the training data used by the machine learning model is actually obtained based on OpenSees. Therefore, using the constructed machine learning model to assess building vulnerability still inherently considers the physical and mechanical properties of buildings. In other words, while the practical software program designed in this embodiment of the present invention offers simple input and fast calculations, the resulting vulnerability assessment results still take into account the physical and mechanical processes of landslide impacts on buildings. However, this practical program also has certain limitations. For example, in the vulnerability assessment method for RC buildings impacted by landslides, the pushover curve obtained by the NSPA method reflects the global performance of the RC building and cannot intuitively reflect the local failure process of beams and columns like the SPH-FEM coupled numerical simulation. It is precisely because the NSPA analysis does not fully consider the different failure mechanisms of buildings that the vulnerability assessment accuracy of the practical software program is limited. However, given current theoretical and technical capabilities, the NSPA method is feasible for rapid, unrefined assessment of the vulnerability of clusters of RC buildings to landslides. If localized building failure and the resulting progressive collapse are considered for a more accurate assessment of building vulnerability, the probabilistic assessment of clustered buildings can be used to rank the buildings by vulnerability, identifying those most vulnerable to damage. Further analysis can then be performed using more refined assessment methods.

[0148] The efficient building vulnerability assessment method proposed in this embodiment of the present invention has an average computation time of only 47 seconds, which is 383 times more efficient than previous research results. Due to its remarkable computational efficiency, it can be used to evaluate the vulnerability of a large number of buildings with different structural forms, thereby achieving efficient probabilistic assessment of the vulnerability of a group of RC buildings to landslide impacts.

[0149] Based on the same inventive concept, the present invention also provides a machine learning-driven efficient probabilistic assessment device for landslide-impacted building vulnerability, the machine learning-driven efficient probabilistic assessment device for landslide-impacted building vulnerability comprising:

[0150] A dataset construction module is used to determine the ultimate resistance capacity of buildings of different structural forms based on NSPA analysis, and to construct a building structure resistance capacity prediction dataset through multi-parameter combination simulation. The building structure resistance capacity prediction dataset includes: building structure geometric parameters, building structure mechanical parameters, landslide impact height, and the corresponding building structure resistance capacity;

[0151] A training module is used to train a machine learning model based on the building structure resistance capacity prediction data set to obtain a building structure resistance capacity prediction model;

[0152] A data calculation model construction module is used to analyze the impact force on the building structure under different landslide impact heights and landslide impact velocities based on the SPH-FEM coupled dynamic model, construct an impact force data set, and construct a landslide impact force mathematical calculation model based on the impact force data set;

[0153] A first determining module is used to determine the resistance capacity of the infill wall according to the failure mode of the infill wall;

[0154] a second determination module, configured to determine the resistance of the building structure based on the randomly sampled values of the landslide impact height, the landslide impact velocity, the building structure geometric parameters, and the building structure mechanical parameters, based on the building structure resistance prediction model, and to determine the landslide impact force based on the mathematical calculation model of the landslide impact force;

[0155] The third determination module is used to compare the landslide impact force with the resistance capacity of the building structure and the resistance capacity of the filling wall to determine the damage state of the building;

[0156] The vulnerability contour map construction module uses the damage state determination method of the building to calculate the damage state corresponding to each random sample, determines the probabilistic vulnerability of the building based on the Monte Carlo principle, and constructs vulnerability contour maps of buildings with different structural forms.

[0157] Based on the same inventive concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When executed, the processor implements the steps in the machine learning-driven efficient probabilistic assessment method for landslide-impacted building vulnerability as described in any of the above embodiments.

[0158] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability described in any of the above embodiments.

[0159] Based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the machine learning-driven efficient probabilistic assessment method for landslide-impact building vulnerability described in any of the above embodiments.

[0160] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0161] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0162] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (apparatus), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0163] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0164] These computer program instructions can also be loaded onto a computer or other programmable terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0165] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0166] Finally, it should be noted that, in the present invention, relational terms such as first and second, etc., are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises", or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article, or terminal device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0167] The above is a detailed introduction to the machine learning-driven efficient probabilistic assessment method for landslide-impacted building vulnerability provided by the present invention. Specific examples are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A machine learning-driven efficient probabilistic assessment method for landslide-induced building vulnerability, characterized by: The method comprises: Determine the ultimate resistance of buildings of different structural forms based on NSPA analysis, and construct a building structure resistance prediction data set through multi-parameter combination simulation. The building structure resistance prediction data set includes: building structure geometric parameters, building structure mechanical parameters, landslide impact height, and the corresponding building structure resistance; Training a machine learning model based on the building structure resistance capacity prediction data set to obtain a building structure resistance capacity prediction model; Based on the SPH-FEM coupled dynamics model, the impact force on the building structure under different landslide impact heights and landslide impact velocities was analyzed, an impact force dataset was constructed, and a mathematical calculation model for landslide impact force was constructed based on the impact force dataset; Determine the resistance capacity of the infill wall based on the failure mode of the infill wall; Determining the resistance of the building structure based on the random sampling values of the landslide impact height, the landslide impact velocity, the geometric parameters of the building structure, and the mechanical parameters of the building structure, and determining the landslide impact force based on the mathematical calculation model of the landslide impact force; Compare the landslide impact force with the resistance of the building structure and the resistance of the filling wall to determine the damage status of the building; The damage state determination method for the building is used to calculate the damage state corresponding to each random sample, the probabilistic vulnerability of the building is determined based on the Monte Carlo principle, and the vulnerability contour maps of buildings with different structural forms are constructed.

2. The machine learning driven efficient probabilistic assessment method for landslide impact building vulnerability according to claim 1 is characterized in that: Based on NSPA analysis, the ultimate resistance capacity of buildings with different structural forms is determined. A building structure resistance capacity prediction dataset is constructed through multi-parameter combination simulation, including: A two-dimensional frame finite element model of the building structure was established. Dead loads and live loads were applied to the finite element model to perform an initial gravity analysis. Lateral load distribution was applied along the impact height to perform an NSPA analysis to obtain the pushover curve of the building structure. The Pushover curve is used to determine three limit states to reflect the performance status of the building structure. The three limit states include SLS1, SLS2, and SLS3. SLS1 represents the end of the linear portion of the Pushover curve, indicating the first significant change in the global performance of the building structure. SLS2 is the structural yield point determined by the equal energy method. SLS3 is located at the peak point of the Pushover curve, indicating that the structure has reached its bearing capacity limit. The lateral force corresponding to each limit state is determined as the ability of the building structure to resist the lateral impact of the landslide; The MCS method is used to perform multiple sampling on the probability distribution of random variables involved in the lateral resistance analysis of RC building structures to obtain the input database for NSPA analysis. The input database is assigned to the Python automated analysis program in turn, and the lateral resistance of the building structure corresponding to each input data is calculated: R1, R2 and R3; Finally, the input dataset is fused with the output dataset of building structure lateral resistance capacity to form a building structure resistance capacity prediction dataset.

3. The machine learning driven efficient probabilistic assessment method for landslide impact building vulnerability according to claim 1, characterized in that: Based on the SPH-FEM coupled dynamic model, analyze the magnitude of the impact force on the building structure under different landslide impact heights and landslide impact velocities, construct an impact force data set, and construct a mathematical calculation model for the landslide impact force based on the impact force data set, including: Within the preset ranges of the landslide material density ρ, the landslide impact height h, and the landslide impact velocity v, perform multiple uniform samplings to obtain multiple groups of sample data under different combinations of density, velocity, and height; substitute the multiple groups of sample data into the established SPH-FEM coupled numerical model in sequence to obtain the landslide impact force corresponding to each group of sample data; Randomly select multiple groups of samples to calculate the empirical coefficient α and the Froude number Fr; used to fit the power-law coefficients a1 and a2; Based on the fitting results of a1 and a2 and the analytical formula for the maximum impact force of the landslide on the building structure, determine the mathematical calculation model for the landslide impact force, and the analytical formula is: F=αρv 2 A; Where ρ is the density of the moving material in the landslide; v is the impact velocity of the landslide; and A is the impact area of the landslide.

4. The machine learning driven efficient probabilistic assessment method for landslide impact building vulnerability according to claim 3, characterized in that: According to the failure mode of the infill wall, determine the resistance of the infill wall, including: The maximum bending moment M per unit width of the infill wall is determined based on the impact velocity v and impact height h using the following formula: b : Where, L is the height of the infill wall; α is the empirical coefficient in the impact force calculation model; ρ is the density of the landslide moving material; According to the masonry structure design code, the maximum bending capacity M of the masonry infill wall is calculated using the following formula: c : M c =f t W; Among them, f t is the maximum tensile strength of masonry; W is the section resistance moment, which is related to the width and thickness of the wall; The calculated maximum bending moment M b Less than the maximum bending capacity M c In the case of the infill wall is not damaged; in the calculated maximum bending moment M b Not less than the maximum bending capacity M c In this case, it is determined that the infill wall is damaged.

5. The machine learning driven efficient probabilistic assessment method for landslide impact on buildings according to claim 4 is characterized in that: Compare the landslide impact force with the resistance of the building structure and the resistance of the infill wall to determine the damage state of the building, including: F < R1 and M b <M c In the case of, determine that the damage state of the building is undamaged; F < R1 and M b ≥ M c In the case of, determine that the damage state of the building is slight damage; When R1 ≤ F < R2, determine that the damage state of the building is medium damage; When R2 ≤ F < R3, determine that the damage state of the building is severe damage; F≥R3 or h>H building In the case of , the damage state of the building is determined to be total damage; According to the relationship between the calculated landslide impact force F and the resistance of the building structure: R1, R2 and R3, and the resistance of the infill wall, determine the corresponding damage state of the building, and the damage states include: no damage, slight damage, medium damage, severe damage or complete damage.

6. The machine learning driven efficient probabilistic assessment method for landslide impact building vulnerability according to claim 1, characterized in that: Use the method for determining the damage state of the building to calculate the damage state corresponding to each random sample, determine the probabilistic vulnerability of the building based on the Monte Carlo principle, and construct the vulnerability contour map of buildings with different structural forms, including: Determine the exceedance probability of different limit states of the building based on the following formula: G i (θ) represents the performance function of each limit state; G i (θ)=sign[D(θ)-D tv,i ]; θ represents the random variable vector; T is the number of samples of MCS, θ j is the jth sample of the simulation; D tv,i Indicates LS i The threshold value of LS1 is tv,1 =1, D of LS2 tv,2 =2, LS3's D tv,3 =3, LS4's D tv,4 =4; sign[·] is G i (θ), if D(θ)>D tv,i , G i (θ)=1, otherwise G i (θ) = 0; the four limit states LS1, LS2, LS3, and LS4 divide the damage state of the building into five levels: no damage, slight damage, moderate damage, severe damage, and complete damage; Calculate the probability of the building being in different damage states based on the following formula: The probability of exceeding LS1 in the damaged state; The probability of exceeding LS2 in the damaged state; Indicates the probability of the damage state exceeding LS3, Indicates the probability of the damage state exceeding LS4; DS1 means no damage, DS2 means slight damage, DS3 means moderate damage, DS4 means severe damage, and DS5 means complete damage; Calculate the vulnerability of the building based on the following formula: Where n = 1, 2, 3, 4 and 5; lr n is the loss ratio of the nth damage state; represents the probability that the building is in the nth damaged state; Calculate the vulnerability of buildings with different structural forms at different impact velocities and impact heights in sequence, and construct the vulnerability contour map of each building in the group of buildings.

7. A machine learning-driven efficient probabilistic assessment device for landslide impact on building vulnerability, characterized by: The machine learning-driven high-efficiency probabilistic evaluation device for the vulnerability of landslide-impacted buildings includes: A data set construction module for determining the limit state resistance of building structures with different structural forms based on NSPA analysis, and constructing a prediction data set for the resistance of building structures through multi-parameter combination simulation. The prediction data set for the resistance of building structures includes: geometric parameters of the building structure, mechanical parameters of the building structure, landslide impact height, and the corresponding resistance of the building structure; A training module for training a machine learning model based on the prediction data set for the resistance of building structures to obtain a prediction model for the resistance of building structures; A data calculation model construction module is used to analyze the impact force on the building structure under different landslide impact heights and landslide impact velocities based on the SPH-FEM coupled dynamic model, construct an impact force data set, and construct a landslide impact force mathematical calculation model based on the impact force data set; A first determining module is used to determine the resistance capacity of the infill wall according to the failure mode of the infill wall; a second determination module, configured to determine the resistance of the building structure based on the randomly sampled values of the landslide impact height, the landslide impact velocity, the building structure geometric parameters, and the building structure mechanical parameters, based on the building structure resistance prediction model, and to determine the landslide impact force based on the mathematical calculation model of the landslide impact force; The third determination module is used to compare the landslide impact force with the resistance capacity of the building structure and the resistance capacity of the filling wall to determine the damage state of the building; The vulnerability contour map construction module is used to calculate the damage state corresponding to each random sample using the damage state determination method of the building, determine the probabilistic vulnerability of the building based on the Monte Carlo principle, and construct vulnerability contour maps of buildings with different structural forms.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts is implemented as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the machine learning-driven efficient probabilistic assessment method for vulnerability of buildings to landslide impacts is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the machine learning-driven efficient probabilistic assessment method for landslide-impacted building vulnerability described in any one of claims 1 to 6 are implemented.

Images