CN-121118679-B - Method for predicting post-earthquake node damage of assembled steel structure building based on two-step method

CN121118679BCN 121118679 BCN121118679 BCN 121118679BCN-121118679-B

Abstract

The invention discloses a two-step method-based method for predicting post-earthquake node damage of an assembled steel structure building, which relates to the technical field of post-earthquake node damage prediction and comprises the steps of obtaining structural response data of the assembled steel structure building under the action of an earthquake, wherein the structural response data comprise an earthquake level, a time stamp and ground acceleration; based on the magnitude, the time stamp and the ground acceleration, adopting RIME-LSTM-self-attention algorithm to predict building bottom acceleration, adopting ELM algorithm to predict target node acceleration based on the magnitude, the time stamp, the ground acceleration and the building bottom acceleration, determining Park-Ang damage index parameters, calculating Park-Ang damage index by combining the target node acceleration, and judging the state of the target node according to the Park-Ang damage index. According to the method, the bottom acceleration of the assembled steel structure building is predicted firstly by a two-step prediction method, then the node acceleration is predicted, and finally the damage index is calculated, so that the post-earthquake node damage prediction of the assembled steel structure building is realized.

Inventors

WANG SHUNGUO
DING ZUDE
WANG CHANGYU
REN ZHIHUA
CHEN YUSHENG
LI RUNTAO

Assignees

昆明理工大学

Dates

Publication Date: 20260512
Application Date: 20250918

Claims (9)

1. The method for predicting the damage of the post-earthquake node of the assembled steel structure building based on the two-step method is characterized by comprising the following steps of: Simulating the actions of different earthquake magnitudes and earthquake waves through a vibration table test, and obtaining structural response data of the assembled steel structure building under the earthquake action, wherein the response data comprise the earthquake magnitudes, time stamps and ground acceleration; Based on the magnitude, the time stamp and the ground acceleration, adopting RIME-LSTM-Self-attention algorithm to predict the building bottom acceleration, wherein the RIME-LSTM-Self-attention algorithm is a hybrid model combining an optimization algorithm, a deep learning component and machine learning, super-parameters are optimized through RIME algorithm, LSTM layer captures time sequence dependency, and Self-attention mechanism enhances key characteristics and provides model interpretability; Predicting the acceleration of a target node by adopting an ELM algorithm based on the magnitude of vibration, the time stamp, the ground acceleration and the building bottom acceleration; Determining Park-Ang damage index parameters, and calculating Park-Ang damage indexes by combining the target node acceleration, wherein the Park-Ang damage index parameters comprise yield strength, shear strength, cross-sectional area, layer height and node quality, and the Park-Ang damage indexes are used for quantifying the damage degree of the target node; judging the state of a target node according to the Park-Ang damage index; Wherein, determining Park-Ang damage index parameters, and calculating Park-Ang damage index in combination with the target node acceleration comprises: Calculating plastic bending moment according to yield strength and plastic section modulus of steel : In the formula (I), in the formula (II), And Respectively the yield strength and the plastic section modulus of the steel; based on the plastic bending moment Calculating limit lateral load of lateral force resisting component by layer height : In the formula (I), in the formula (II), The total layer height of the assembled steel structure building is; according to the limit side load of the side force resisting component Deformation limit value under static force action of component rigidity calculation : In the formula (I), in the formula (II), Is the rigidity of the component; And (3) quantifying plastic deformation energy by integrating the product of the acceleration and the speed of the target node: In the formula (I), in the formula (II), Indicating hysteresis energy dissipation, i.e. plastic deformation energy; the node quality is taken as the distribution quality of the target node; And Respectively represent Acceleration and velocity of the time node, wherein, ; According to the plastic deformation energy Deformation limit under static force Calculating a Park-Ang damage index of the target node: In the formula (I), in the formula (II), Representing a target node Park-Ang damage index; indicating that under the action of the earthquake, Maximum deformation value before time; Indicating the shear strength of the node point, , Is a cross-sectional area; Representing the energy weighting coefficients.
2. The method for predicting post-earthquake node damage of an assembled steel structure building based on a two-step method according to claim 1, wherein the method further comprises constructing an LSTM-self-attention prediction network model before predicting building bottom acceleration by adopting RIME-LSTM-self-attention algorithm based on the magnitude, the time stamp and the ground acceleration, and comprises the following steps: Obtaining structural response data of the assembled steel structure building under the action of different earthquake magnitudes and earthquake waves and corresponding building bottom acceleration through a vibration table test, and constructing a first data set containing the earthquake magnitudes, the time stamp, the ground acceleration and the building bottom acceleration; Inputting the first data set into a bidirectional LSTM network, respectively generating a forward hidden state sequence and a reverse hidden state sequence through a forward propagation layer and a reverse propagation layer, and splicing to form a bidirectional fusion characteristic tensor; inputting the bidirectional fusion feature tensor into a self-attention module, and mapping the input features into query vectors, key vectors and value vectors through a linear transformation layer to form a multi-head attention calculation space; calculating a similarity matrix between a query vector and a key vector by utilizing a scaling dot product algorithm, and generating an attention weight matrix by Softmax normalization, wherein the weight matrix represents global dependency strength among different time step characteristics; Applying a bottleneck compression layer to the globally enhanced feature representation, reducing the channel dimension to 1/4 of the original input by adopting a1×1 convolution kernel, and inhibiting redundant information through layer normalization; and inputting the compressed characteristics into a fully-connected regression layer, and outputting the predicted value of the building bottom acceleration.
3. The two-step method based post-earthquake node damage prediction method for an assembled steel structure building according to claim 2, wherein after constructing the LSTM-self-attention prediction network model, the method further comprises: And utilizing self-adaptive LSTM super-parameter configuration of an ice optimization algorithm, and jointly optimizing the number of LSTM neurons and the initial learning rate in the LSTM-self-attention prediction network model through a multi-objective fitness function to construct a RIME-LSTM-self-attention prediction model.
4. The two-step method based method for predicting post-earthquake node damage of an assembled steel structure building according to claim 1, wherein the method further comprises constructing an ELM prediction network before predicting the target node acceleration by adopting an ELM algorithm based on the magnitude of earthquake, the time stamp, the ground acceleration and the building bottom acceleration, and comprises: Constructing a second data set containing magnitude, time stamp, ground acceleration, building bottom acceleration and acceleration corresponding to each monitoring node; independently constructing a single hidden layer feedforward neural network for each monitoring node, wherein the single hidden layer feedforward neural network comprises an input layer, a hidden layer and an output layer, wherein the input layer is used for receiving original characteristic data acquired by a sensor, and the dimension is consistent with the number of physical parameters; Randomly generating initial parameters from uniform distribution or normal distribution, and locking a connection weight matrix and a bias vector from an input layer to a hidden layer by adopting the initial parameters, wherein the connection weight matrix and the bias vector cannot be updated in the training process; The ELM prediction network is independently deployed to each monitoring node, receives the real-time data stream and synchronously outputs the target node acceleration prediction value.
5. The two-step method based post-earthquake node damage prediction method for fabricated steel structure building according to claim 4, wherein after constructing ELM prediction network, the method further comprises: and utilizing the adaptive ELM super-parameter configuration of the frost optimization algorithm to jointly optimize the number of ELM neurons in the ELM prediction network through a multi-objective fitness function.
6. The two-step method based post-earthquake node damage prediction method for an assembled steel structure building according to claim 1, wherein the judging of the target node state according to the Park-Ang damage index comprises: and if the Park-Ang damage index of the target node is greater than 1, the target node is considered to be damaged.
7. A two-step method-based post-earthquake node damage prediction device for an assembled steel structure building, based on the two-step method-based post-earthquake node damage prediction method for an assembled steel structure building according to any one of claims 1 to 6, comprising: The data acquisition module is used for simulating the effects of different earthquake magnitudes and earthquake waves through a vibration table test and acquiring structural response data of the assembled steel structure building under the earthquake effect; the building bottom acceleration prediction module is used for predicting the building bottom acceleration by adopting RIME-LSTM-self-attention algorithm based on the magnitude, the time stamp and the ground acceleration; The target node acceleration prediction module is used for predicting the target node acceleration by adopting an ELM algorithm based on the magnitude of vibration, the time stamp, the ground acceleration and the building bottom acceleration; the node damage estimation module is used for determining Park-Ang damage index parameters, calculating Park-Ang damage indexes by combining the target node acceleration, and judging the state of the target node according to the Park-Ang damage indexes.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor performs the steps of the two-step method of predicting post-earthquake node damage of an assembled steel structure building as claimed in any one of claims 1 to 6.
9. A storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the two-step method of predicting post-earthquake node damage of an assembled steel structure building as claimed in any one of claims 1 to 6.

Description

Method for predicting post-earthquake node damage of assembled steel structure building based on two-step method Technical Field The invention relates to the technical field of post-earthquake node damage prediction, in particular to a two-step method-based method for predicting post-earthquake node damage of an assembled steel structure building. Background In the field of structural seismic damage evaluation, the dynamic transmission relation between the ground acceleration and the building node acceleration is a core physical mechanism for evaluating damage. In the prior art, structural response data of a specific building under the action of a simulated earthquake is obtained through a vibration table test, a mapping relation between earthquake motion input (namely base acceleration) and building node acceleration response is established, and based on the mapping relation, a Park-Ang damage model is combined to quantitatively evaluate the node damage degree, or a large-scale earthquake-structural response data set is generated through finite element simulation, and a direct prediction model from earthquake motion time (such as peak acceleration, peak Ground Acceleration and PGA) to node damage is established through a machine learning algorithm. In addition, the data driving method establishes an end-to-end damage classification model by fusing the seismic intensity index (such as CAV, cumulative Absolute Velocity, cumulative absolute velocity; ASI, acceleration Spectrum Intensity, acceleration spectrum intensity) with the structural response characteristics (interlayer displacement angle, strain energy density). The method takes ground movement characteristics as input, takes node dynamic response as an intermediate bridge, and finally correlates to damage state output. However, despite the advances made by existing methods in the area of structural seismic damage assessment, there are still a number of limitations: Firstly, the data acquisition cost is high and the diversity is insufficient, namely, the vibrating table test is limited by the test scale and cost, only a single building type can be covered generally, the response characteristics of different structures under the action of earthquake are difficult to comprehensively reflect, while the finite element simulation can simulate various structure types, but the high-performance computing resources are relied on, the modeling process is time-consuming and labor-consuming, and the generated data set still has larger limitation on the scale and diversity; Secondly, the prediction precision is limited, and the prediction precision of the data driving model is difficult to break through a physical simulation boundary, wherein the prediction precision of the data driving model depends on the accuracy of finite element simulation or test data to a great extent, however, due to a plurality of uncertainties and simplifying assumptions in the physical simulation process, a certain error exists between simulation data and real structure response, and the prediction precision of the data driving model is further limited, so that the inherent precision boundary of the physical simulation is difficult to break through; Moreover, the generalization capability is insufficient, and the method is difficult to adapt to the change of structural parameters, wherein the traditional method is generally customized for specific structural types in the modeling process, is difficult to adapt to the change of parameters such as the connection form of the assembled steel structure nodes, the size of the components and the like, and needs to perform modeling and training again when the structural parameters change; In addition, the characteristic utilization is insufficient, the traditional research is that a plurality of direct mapping earthquake vibrations are directly mapped to damage results, intermediate physical processes such as substrate acceleration transmission, node response evolution and the like are not effectively decoupled, and the direct modeling mode can simplify the evaluation flow, but neglects a complex dynamic response mechanism of a structure under the action of an earthquake, so that the reliability and the interpretation of the evaluation results are affected to a certain extent. Disclosure of Invention In order to solve the technical problems, the invention provides a two-step method-based method for predicting post-earthquake node damage of an assembled steel structure building, so as to realize the prediction of the post-earthquake node damage of the assembled steel structure building. According to one aspect of the invention, a method for predicting post-earthquake node damage of an assembled steel structure building based on a two-step method is provided, the method comprises the steps of simulating different vibration levels and earthquake wave actions through a vibration table test, obtaining structural response data of the assembled steel s