Construction method of human body earthquake response database based on multi-scenario shaking table test

Abstract

The application discloses a kind of based on multi-scenario shaking table test human earthquake response database construction method, through multi-scenario shaking table test shaking table simulates different grades and frequency seismic waves;In the multi-scenario shaking table test shaking table multidimensional physiological and biochemical index monitoring to human, collects experimental data and constructs human earthquake response graph sample set;The human earthquake response graph sample set is subjected to sample enhancement;After enhancement, human earthquake response graph sample is subjected to graph neural network training and obtains earthquake response neural network model;Through earthquake response neural network model, new earthquake response graph sample is generated by generative adversarial network model and is added in human earthquake response database.The application utilizes graph neural network model, has good adaptive performance for human earthquake response interpretation, can better extract the key information of human earthquake response, so good effect can be achieved.

Description

Technical Field

[0001] This invention relates to the field of seismic response database technology, and in particular to a method for constructing a human seismic response database based on multi-scenario shaking table tests. Background Technology

[0002] Earthquakes are a common natural disaster with immense destructive power over human society and the environment. During an earthquake, the human body is affected by seismic waves and exhibits certain reactions. However, current research on human responses to earthquakes primarily relies on finite element models and theoretical analysis, lacking direct experimental data. Therefore, obtaining data on human responses to earthquakes through actual experiments is of great significance for earthquake engineering and ergonomics. Natural earthquake motion is significantly irregular due to the combined influence of factors such as the hypocenter, propagation path, and local site conditions; in fact, almost no two records of strong natural earthquakes are exactly alike. Furthermore, the impact of earthquake motion on human behavior is influenced by different individuals and their different states. In other words, even with the same earthquake record, the characteristics and patterns of human earthquake behavior can vary significantly among different individuals, and even among the same individual in different states.

[0003] With the continuous development of social economic construction and science and technology, people are paying more and more attention to the impact of earthquakes. On the one hand, it is necessary to carry out corresponding scientific research to analyze the characteristics and laws of human earthquake behavior and provide corresponding suggestions for dealing with the adverse results caused by human earthquake behavior during real earthquakes. Therefore, a method for constructing a human earthquake response database based on multi-scenario shaking table tests is needed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing a human seismic response database based on multi-scenario shaking table tests.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0006] This invention includes the following steps:

[0007] A simulates seismic waves of different levels and frequencies using a multi-scenario shaking table test.

[0008] B uses the multi-scenario shaking table test to monitor the human body's physiological and biochemical indicators in multiple dimensions, and collects experimental data to construct a sample set of human body seismic response maps;

[0009] C performs sample augmentation on the aforementioned human seismic response map sample set;

[0010] D. A graph neural network model for earthquake response neural network is obtained by training the enhanced human seismic response map samples.

[0011] E uses a seismic response neural network model and a generative adversarial network model to generate new seismic response map samples and add them to the human seismic response database.

[0012] Furthermore, the method for monitoring multi-dimensional physiological and biochemical indicators of the human body using the multi-scenario vibration table test vibration table includes real-time acquisition of experimental data and status information of the human body. The acquisition unit is installed on or near the human body. The experimental data is the rate of change of heart rate intervals. The status information includes physiological indicators, behavioral performance, psychological indicators, environmental factors, and physical injury status, which are used to establish correlation analysis between the experimental data and the status information.

[0013] Furthermore, the experimental data include, but are not limited to, electrocardiogram, blood pressure, blood oxygen saturation, and adrenaline.

[0014] Furthermore, the status information is obtained through non-contact monitoring methods, including data on the motion trajectory and dynamic indicators of key human nodes, as well as video data of the demonstration process recorded by a video monitoring system.

[0015] Furthermore, the sample enhancement method in step C includes:

[0016] a. Randomly select human earthquake response samples from the constructed sample set;

[0017] b. Delete and add nodes and lines to the extracted sample maps to enhance the samples and expand the human seismic response map sample set;

[0018] Furthermore, the graph neural network training method in step D includes constructing a graph neural network model and training the model on human seismic response map samples. The graph neural network model includes multiple graph convolutional layers, graph attention layers, and fully connected layers. The model is trained using a training set and validated and adjusted using a validation set. Backpropagation and an optimizer are used to update the model's weights and parameters.

[0019] Furthermore, the dataset is divided into training and testing sets according to an 8-2 split, i.e., 80% is the training set and 20% is the testing set.

[0020] Furthermore, the method for sample augmentation of the human seismic response map sample set includes using a neural network model for incremental learning, inputting new data into the model, and allowing the model to automatically learn and update weight parameters. The calculation formula for the weight parameters includes:

[0021]

[0022]

[0023]

[0024] The experimental data for the s-th data point is b. s The g-th state data of the s-th data is c. gs Experimental data b s State data c gs The single-state weights are Data object c p In experimental data b s The equivalence class is The a-th related state is b a Related state b a State data c gs The weight is State value c gs The weight is ω(c) gs ), state value c gs and state value c ga The number of co-occurrences is The number of relevant states is d. Randomly initialize cluster centers, average the weights of the corresponding state values, and calculate the distance between the data object and the cluster.

[0025]

[0026] The t-th state value of the s-th data is c. st The state value of the t-th data point is c. at State value c st and state value c at The distance between clusters is d(c st ,c at ), state value c st The weight is ω(c) st ), state value c at The weight is ω(c) at ),

[0027] Furthermore, the input of the graph neural network includes two main parts: a feature matrix X and an adjacency matrix A. The feature matrix X is an N×F matrix, where N is the number of nodes in the graph and F is the number of input features for each node. The features of each node in the feature matrix X are represented as an F-dimensional vector, including the clustering coefficient and in-degree of the node. The clustering coefficient can measure the importance of a node among its neighboring nodes, while the in-degree and out-degree measure the tightness of the connection between the node and other nodes.

[0028] The beneficial effects of this invention are:

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

[0030] This invention utilizes a graph neural network model, which has excellent adaptability to interpreting human seismic responses and can better extract key information about human seismic responses, thus achieving good results. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the method for constructing a human seismic response database based on multi-scenario shaking table tests according to the present invention.

[0032] Figure 2 This is a schematic diagram illustrating the construction method of the human seismic response database based on multi-scenario shaking table tests according to the present invention. Detailed Implementation

[0033] The present invention will be further described below with reference to specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0034] This invention includes the following steps:

[0035] like Figure 1 and 2 As shown, A simulates seismic waves of different levels and frequencies using a multi-scenario shaking table test. The test objectives and the required seismic wave levels and frequency ranges are clearly defined. A suitable shaking table is selected, and necessary settings and adjustments are made to ensure that it can accurately simulate the levels and frequencies of seismic waves. Human models or subjects are prepared, and their physical condition and posture are ensured to be suitable for the test.

[0036] An earthquake intensity scale is a table developed based on parameters such as earthquake intensity, duration, and focal depth. It is used to assess the extent of damage caused by an earthquake to buildings, facilities, and the environment. Actual earthquake experience refers to the direct feelings and descriptions of people who have experienced an earthquake, including physical reactions, psychological states, and behavioral responses.

[0037] Case 1 under Intensity Level 1 (Low Intensity):

[0038] Peak acceleration: 0.3g

[0039] Frequency range: 1-2Hz

[0040] Duration: 30 seconds

[0041] In this set of experiments, subjects exhibited slight swaying and wobbling, but their body posture remained stable. The movement trajectory showed a small range and amplitude. In terms of physiological responses, subjects experienced mild shortness of breath and increased heart rate, but no other significant changes.

[0042] Case 2 under intensity level 3 (moderate intensity):

[0043] Peak acceleration: 1.0g

[0044] Frequency range: 2-3Hz

[0045] Duration: 30 seconds

[0046] In this set of experiments, the subjects exhibited more noticeable swaying and swinging, with some changes in body posture. The range and amplitude of their movement trajectories increased compared to before. Physiologically, the subjects experienced significant shortness of breath, increased heart rate, and a certain degree of elevated blood pressure.

[0047] Case 3 under intensity level 5 (high intensity):

[0048] Peak acceleration: 2.5g

[0049] Frequency range: 3-5Hz

[0050] Duration: 30 seconds

[0051] In this experiment, subjects exhibited strong shaking and swaying, with significant impact on their body posture. The range and amplitude of their movement trajectories increased significantly. Physiologically, subjects experienced severe shortness of breath, increased heart rate, and a significant increase in blood pressure. Some subjects exhibited panic reactions, requiring emergency measures to ensure their safety.

[0052] B uses the multi-scenario shaking table test to monitor the human body's physiological and biochemical indicators in multiple dimensions, and collects experimental data to construct a sample set of human body seismic response maps;

[0053] The experimental data collection unit comprises multiple devices installed on or around the human body to monitor and collect vital data. Specifically, in addition to a heart rate monitor, the unit includes a pulse oximeter, blood pressure monitor, thermometer, a network camera capturing facial and body movements, and surface myoelectric sensors measuring muscle movements. Each device can be installed individually or in multiple units. For example, installing multiple pulse oximeters at different locations on the body improves measurement accuracy. In addition, microphones and pedometers that collect human voice data can be set as status data. The experimental data is human seismic dynamics data, which can provide researchers with data support for analyzing the characteristics and laws of human seismic response from a kinematic perspective. It can vividly record human behavior and reaction state during the shaking table test, providing data support for scientific research analysis and popular science demonstration. The status data can provide data support for studying human seismic response and perception from a physiological perspective. Combined with the analysis results of human motion capture data, the influence of vibration state on physiological indicators such as heart rate and blood oxygen can be studied. Screen recording and image capture of the collected data can produce animated videos of the model composed of key human nodes during the shaking table test, as well as trajectory images of some key nodes. Based on the human shaking table test video monitoring data, the vibration response process of the human shaking table test can be recorded, and detailed records of human vibration response can also be provided.

[0054] C performs sample augmentation on the aforementioned human seismic response map sample set;

[0055] D. A graph neural network model for earthquake response neural network is obtained by training the enhanced human seismic response map samples.

[0056] E uses a seismic response neural network model and a generative adversarial network model to generate new seismic response map samples and add them to the human seismic response database.

[0057] In this embodiment, the method for monitoring multi-dimensional physiological and biochemical indicators of the human body using the multi-scenario vibration table test vibration table includes real-time acquisition of experimental data and status information of the human body. The acquisition unit is installed on or near the human body. The experimental data is the rate of change of heart rate intervals. The status information includes physiological indicators, behavioral performance, psychological indicators, environmental factors, and physical injury status, which are used to establish correlation analysis between the experimental data and the status information.

[0058] In this embodiment, the experimental data includes, but is not limited to, electrocardiogram, blood pressure, blood oxygen saturation, and adrenaline.

[0059] In this embodiment, the status information is obtained by non-contact monitoring methods, including data on the motion trajectory and dynamic indicators of key human nodes, and video data of the demonstration process recorded by a video monitoring system.

[0060] In this embodiment, the sample enhancement method in step C includes:

[0061] a. Randomly select human earthquake response samples from the constructed sample set;

[0062] b. Delete and add nodes and lines to the extracted sample maps to enhance the samples and expand the human seismic response map sample set;

[0063] In this embodiment, the graph neural network training method in step D includes constructing a graph neural network model and training the model on human seismic response map samples. The graph neural network model includes multiple graph convolutional layers, graph attention layers, and fully connected layers. The model is trained using a training set and validated and adjusted using a validation set. Backpropagation and an optimizer are used to update the model's weights and parameters.

[0064] In this implementation example, the dataset is divided into training and test sets according to an 8-2 split, that is, 80% is the training set and 20% is the test set.

[0065] In this embodiment, the method for sample augmentation of the human seismic response map sample set includes incremental learning using a neural network model, inputting new data into the model, and allowing the model to automatically learn and update weight parameters. The calculation formula for the weight parameters includes:

[0066]

[0067]

[0068]

[0069] The experimental data for the s-th data point is b. s The g-th state data of the s-th data is c. gs Experimental data b s State data c gs The single-state weights are Data object c p In experimental data b s The equivalence class is The a-th related state is b a Related state b a State data c gs The weight is State value cgs The weight is ω(c) gs ), state value c gs and state value c ga The number of co-occurrences is The number of relevant states is d. Randomly initialize cluster centers, average the weights of the corresponding state values, and calculate the distance between the data object and the cluster.

[0070]

[0071] The t-th state value of the s-th data is c. st The state value of the t-th data point is c. at State value c st and state value c at The distance between clusters is d(c st ,c at ), state value c st The weight is ω(c) st ), state value c at The weight is ω(c) at ),

[0072] In this embodiment, the input of the graph neural network includes two main parts: a feature matrix X and an adjacency matrix A. The feature matrix X is an N×F matrix, where N is the number of nodes in the graph and F is the number of input features for each node. The features of each node in the feature matrix X are represented as an F-dimensional vector, including the clustering coefficient and in-degree of the node. The clustering coefficient can measure the importance of a node among its neighboring nodes, while the in-degree and out-degree measure the tightness of the connection between the node and other nodes.

[0073] This invention utilizes non-contact monitoring methods to obtain relevant data on the movement trajectory and dynamic indicators of key human nodes, video data of the demonstration process recorded by a video monitoring system, and data on relevant physiological indicators of the human body monitored by a pulse oximeter and a heart rate monitor.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a human seismic response database based on multi-scenario shaking table tests, characterized in that, Includes the following steps: A simulates seismic waves of different levels and frequencies using a multi-scenario shaking table test. B uses the multi-scenario shaking table test to monitor the human body's physiological and biochemical indicators in multiple dimensions, and collects experimental data to construct a sample set of human body seismic response maps; C performs sample augmentation on the aforementioned human seismic response map sample set; D. A graph neural network model for earthquake response neural network is obtained by training the enhanced human seismic response map samples. E uses a seismic response neural network model and a generative adversarial network model to generate new seismic response map samples and add them to the human seismic response database. The graph neural network training method in step D includes constructing a graph neural network model, training the graph neural network model on human seismic response map samples using the graph neural network model, wherein the graph neural network model includes multiple graph convolutional layers, graph attention layers, and fully connected layers, training the graph neural network model using a training set, validating and adjusting the model using a validation set, and updating the model's weights and parameters using a backpropagation algorithm and an optimizer; The method for enhancing the human seismic response map sample set includes using a neural network model for incremental learning. New data is input into the model, allowing it to automatically learn and update weight parameters. The calculation formula for the weight parameters includes: ； ； ； The experimental data for the s-th data point are: The g-th state data of the s-th data is Experimental data Status data The single-state weights are Data objects In experimental data The equivalence class is The a-th related state is Related status Status data The weight is State value The weight is State value and state value The number of co-occurrences is Given d related states, randomly initialize cluster centers, average weights and combine the weights of corresponding state values, and calculate the distance between data objects and clusters: ； The t-th state value of the s-th data is The t-th state value of the a-th data is State value and state value The distance between clusters is State value The weight is State value The weight is .

2. The method for constructing a human seismic response database based on multi-scenario shaking table tests according to claim 1, characterized in that, The method for monitoring multi-dimensional physiological and biochemical indicators of the human body using the multi-scenario vibration table test includes real-time acquisition of experimental data and status information of the human body. The acquisition unit is installed on the human body. The experimental data is the rate of change of heart rate intervals. The status information includes physiological indicators, behavioral performance, psychological indicators, environmental factors, and physical injury status, which are used to establish correlation analysis between the experimental data and the status information.

3. The method for constructing a human seismic response database based on multi-scenario shaking table tests according to claim 2, characterized in that... The experimental data included electrocardiogram, blood pressure, blood oxygen saturation, and adrenaline.

4. The method for constructing a human seismic response database based on multi-scenario shaking table tests according to claim 2, characterized in that... The status information is obtained through non-contact monitoring methods, including data on the movement trajectory and dynamic indicators of key human nodes, and video data of the demonstration process recorded by a video monitoring system.

5. The method for constructing a human seismic response database based on multi-scenario shaking table tests according to claim 1, characterized in that... The sample enhancement method in step C includes: a. Randomly select human earthquake response samples from the constructed sample set; b. Delete and add nodes and lines to the extracted sample maps to enhance the samples and expand the human seismic response map sample set.

6. The method for constructing a human seismic response database based on multi-scenario shaking table tests according to claim 1, characterized in that, The human body earthquake response map sample set is divided into training and test sets according to an 8-2 ratio, that is, 80% is the training set and 20% is the test set.

Images