Generative adversarial network-driven intelligent seismic isolation support design system and method

Abstract

The invention discloses an intelligent seismic isolation support design system and method driven by a generative adversarial network, and relates to the technical field of seismic isolation support design, and the method comprises the following steps: obtaining first people flow data and first vibration data of a transfer station, and load data of a subway; according to the first people flow data, outputting second people flow data representing people flow loads through a generative adversarial network; and generating second vibration data representing vibration prediction through a pre-constructed response generation model according to the second people flow data, the load data and the first vibration data. According to the scheme, the vibration difference is decomposed into quantitative adjustment requirements of specific physical attributes such as rigidity and damping through multi-dimensional difference analysis and physical attribution mapping, so that design parameter optimization of the shock insulation support has definite physical significance and operability; according to the technical scheme, the problems of single vibration difference analysis dimension and fuzzy attribution in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of seismic isolation bearing design technology, specifically to a generative adversarial network-driven intelligent seismic isolation bearing design system and method. Background Technology

[0002] With the continuous expansion of urban rail transit networks and the increasing travel demands of passengers, subway transfer nodes have become one of the areas with the highest population density and the most complex structural loads in urban transportation systems. In order to ensure the long-term safety and operational stability of subway transfer node structures, seismic isolation technology has been widely applied to the structural design and safety assurance of such complex nodes.

[0003] However, in the process of implementing the technical solution of the invention in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0004] Existing technologies acquire data on pedestrian density, subway load, and structural vibration through historical or real-time monitoring. They then use statistical methods or finite element simulation to model and predict structural responses under different working conditions. Subsequently, design parameters for seismic isolation bearings are obtained based on empirical formulas or safety specifications, demonstrating a certain degree of practicality. However, after predicting the structural response, existing technologies do not conduct in-depth analysis of the differences between predicted vibration data and actual monitored vibration data. This results in the inability to accurately identify the specific deviation characteristics of different vibration responses. Furthermore, the lack of quantification of these specific deviation characteristics prevents the conversion of vibration response differences into specific adjustment data, thus limiting the accuracy of seismic isolation bearing design parameters and affecting the overall seismic isolation effect and structural safety. Summary of the Invention

[0005] The purpose of this invention is to provide a generative adversarial network-driven intelligent seismic isolation bearing design system and method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, this invention discloses a generative adversarial network-driven intelligent seismic isolation bearing design method, applied to the design of intelligent seismic isolation bearings at subway transfer points, comprising the following steps:

[0008] Acquire the first passenger flow data, the first vibration data, and the subway load data at the transfer point;

[0009] Based on the first pedestrian flow data, a second pedestrian flow data representing the pedestrian flow load is output through a generative adversarial network;

[0010] Based on the second pedestrian flow data, the load data, and the first vibration data, a second vibration data characterizing vibration prediction is generated using a pre-built response generation model.

[0011] A difference analysis is performed on the first vibration data and the second vibration data to form difference features. Feature attribution processing is then performed on the difference features to obtain vibration adjustment data.

[0012] The first pedestrian flow data and the second pedestrian flow data are subjected to nonlinear processing, and the nonlinear processing result and the vibration adjustment data are coupled and analyzed to generate joint anomaly data;

[0013] The differential features are weighted and fused with the joint anomaly data to obtain a comprehensive early warning index;

[0014] The comprehensive early warning index, the second pedestrian flow data, the load data, and the second vibration data are input into the pre-constructed seismic isolation parameter adjustment mapping model, and the seismic isolation bearing design parameters are output.

[0015] Secondly, this invention discloses a generative adversarial network-driven intelligent seismic isolation bearing design system, comprising:

[0016] The data acquisition module is used to acquire the first pedestrian flow data, the first vibration data, and the subway load data at the transfer point;

[0017] The first pedestrian flow data processing module is used to output second pedestrian flow data representing pedestrian flow load by generating an adversarial network based on the first pedestrian flow data.

[0018] The second vibration data generation module is used to generate second vibration data characterizing vibration prediction based on the second pedestrian flow data, the load data, and the first vibration data, through a pre-built response generation model.

[0019] The vibration adjustment data generation module is used to perform difference analysis on the first vibration data and the second vibration data to form difference features, and to perform feature attribution processing on the difference features to obtain vibration adjustment data.

[0020] The joint anomaly analysis module is used to perform nonlinear processing on the first pedestrian flow data and the second pedestrian flow data, and to perform coupled analysis on the nonlinear processing results and the vibration adjustment data to generate joint anomaly data;

[0021] The comprehensive early warning index calculation module is used to weight and fuse the differential features with the joint abnormal data to obtain the comprehensive early warning index;

[0022] The seismic isolation bearing design module is used to input the comprehensive early warning index, the second pedestrian flow data, the load data, and the second vibration data into a pre-constructed seismic isolation parameter adjustment mapping model, and output the seismic isolation bearing design parameters.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] 1. This solution decomposes vibration differences into quantitative adjustment requirements for specific physical properties such as stiffness and damping through multidimensional difference analysis and physical attribution mapping, making the optimization of seismic isolation bearing design parameters have clear physical meaning and operability. Through the above technical solution, this application solves the problems of single dimension and ambiguous attribution in the existing technology of vibration difference analysis, improves the accuracy and interpretability of seismic isolation bearing design parameter adjustment, and effectively optimizes the dynamic response matching capability of the seismic isolation system.

[0025] 2. This application, through nonlinear processing and multidimensional correlation analysis, can more accurately identify the complex relationship between abnormal pedestrian flow and vibration adjustment. Through the above technical solution, this application solves the problem of anomaly detection error caused by ignoring nonlinear correlation and time matching deviation in the prior art. By improving the accuracy of anomaly identification through spatiotemporal coupling analysis, it provides more reliable anomaly data input for the optimization of seismic isolation bearing design parameters, thereby improving the refinement level of structural safety assessment.

[0026] 3. This scheme, through time-frequency decomposition and multi-channel processing mechanisms, can accurately capture the coupling relationship between pedestrian flow, load, and vibration at specific frequencies. At the same time, through a cross-channel error feedback mechanism, it avoids mutual interference between prediction results of different frequency bands. Through the above technical solutions, this application solves the problem of vibration prediction models ignoring the correlation of frequency domain characteristics in existing technologies. By combining independent optimization of each channel with cross-channel collaborative adjustment, it significantly improves the prediction accuracy of structural vibration response under complex working conditions, providing a reliable data foundation for subsequent seismic isolation parameter optimization. Attached Figure Description

[0027] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0028] Figure 1 This is a flowchart illustrating the steps of the intelligent seismic isolation bearing design method driven by generative adversarial networks according to the present invention.

[0029] Figure 2 A schematic diagram of the process for generating a composite anomaly risk mapping matrix provided by the present invention;

[0030] Figure 3 A schematic diagram of the process for generating a comprehensive early warning index provided by the present invention;

[0031] Figure 4 A flowchart illustrating the output design parameters of the intermediate seismic isolation bearing provided by this invention;

[0032] Figure 5 A schematic diagram of the module functions of the intelligent seismic isolation bearing design system driven by generative adversarial networks provided by the present invention. Detailed Implementation

[0033] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0034] Application Overview:

[0035] In existing technologies, subway transfer nodes, as the areas with the densest passenger flow and the most complex structural loads in urban rail transit networks, have long relied on historical monitoring data and finite element simulation for seismic isolation design. Traditional methods predict structural response through statistical models or empirical formulas, but they have significant drawbacks: the differences between predicted vibration data and actual monitoring data are not analyzed in depth, making it impossible to identify specific deviation characteristics; the lack of quantitative processing of differences makes it difficult to convert them into effective adjustment parameters, directly affecting the accuracy of seismic isolation bearing design and causing structural safety hazards.

[0036] To address the aforementioned issues and the lack of vibration response difference analysis, a dynamic difference feature extraction mechanism is introduced. By comparing the multidimensional features of predicted vibration and actual monitoring data, it is found that time-frequency domain differences and phase drift have a significant impact on structural response. To solve the problem of difference quantification, a feature mapping method based on physical attribution is proposed to transform mathematical differences into engineering-interpretable adjustment parameters. For the nonlinear relationship between pedestrian load and vibration response, a coupling analysis mechanism is designed to reveal potential anomaly correlations. Finally, a comprehensive early warning system is constructed to achieve dynamic optimization of seismic isolation parameters.

[0037] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0038] Example 1:

[0039] Please see Figure 1 A generative adversarial network-driven intelligent seismic isolation bearing design method is applied to the design of intelligent seismic isolation bearings at subway transfer points, including the following steps:

[0040] Acquire the first passenger flow data, the first vibration data, and the subway load data at the transfer point;

[0041] Based on the first pedestrian flow data, a second pedestrian flow data representing the pedestrian flow load is output through a generative adversarial network;

[0042] Based on the second pedestrian flow data, load data, and first vibration data, a second vibration data characterizing vibration prediction is generated using a pre-built response generation model.

[0043] A difference analysis is performed on the first vibration data and the second vibration data to form difference features. Feature attribution processing is then performed on the difference features to obtain vibration adjustment data.

[0044] Nonlinear processing is performed on the first and second pedestrian flow data, and the nonlinear processing results and vibration adjustment data are coupled and analyzed to generate joint anomaly data.

[0045] By weighting and fusing the differential characteristics with the joint anomaly data, a comprehensive early warning index is obtained;

[0046] The comprehensive early warning index, second pedestrian flow data, load data, and second vibration data are input into the pre-constructed seismic isolation parameter adjustment mapping model, and the seismic isolation bearing design parameters are output.

[0047] Among them, the first flow data refers to the structured dataset that is collected in real time by physical sensing devices and preprocessed in the subway transfer area, and is used to characterize the flow characteristics of the area.

[0048] The first vibration data refers to the normalized structural vibration response data collected for subway transfer points, specifically characterizing the structural dynamic response characteristics of the transfer point under the combined action of pedestrian disturbance and train load in the actual operating environment.

[0049] Load data refers to the mechanical data, after normalization, that reflects the forces exerted on the structural system by the subway during its operation at transfer points;

[0050] Generative adversarial networks (GANs) are deep learning models consisting of two neural networks: a generator and a discriminator.

[0051] Secondary pedestrian flow data refers to composite pedestrian load characteristic data that is simulated or predicted after being processed by a generative adversarial network and then normalized, based on the primary pedestrian flow data at transfer points.

[0052] A response generation model is a model that, under specific data input conditions, outputs vibration response data with physical or statistical significance based on training results from historical samples or generative adversarial data.

[0053] The second vibration data refers to the predicted or simulated values ​​of the structural vibration response generated by a neural network prediction model based on the first vibration data that has been collected and the associated second pedestrian flow data and load data, and after normalization.

[0054] Difference analysis refers to the process of identifying the dynamic deviation and change patterns between the first vibration data collected on-site at the transfer point and the second vibration data generated by the generative adversarial network through quantitative comparison and analysis of multi-dimensional, multi-scale, and multi-physical quantity characteristics, and then extracting the core difference features that reflect vibration anomalies or disturbances.

[0055] The difference features refer to the set of statistical and structural features obtained by multi-dimensional comparison between the first vibration data and the second vibration data collected at the same subway transfer point at different times or under different conditions.

[0056] Feature attribution processing refers to the process of quantitatively analyzing and decomposing the contribution or influence of each component element in the difference features based on the difference features between the first vibration data and the second vibration data, using data-driven algorithms.

[0057] Vibration adjustment data refers to the correction parameters and state description information of structural vibration response extracted by the feature attribution method based on the difference analysis results between the first vibration data and the second vibration data at the transfer point.

[0058] Nonlinear processing refers to the methods, models, or algorithms used in data analysis and processing that do not follow simple linear superposition or linear mapping rules, but instead reveal the complex relationships and dynamic characteristics hidden between data through complex nonlinear transformations, interactions, multidimensional coupling, and other mechanisms.

[0059] Nonlinear processing results refer to the comprehensive output data obtained by constructing a nonlinear interaction mapping model for the first flow of people and the second flow of people, and by adopting a multi-level, multi-scale, and multi-variable coupling mechanism to explore the nonlinear relationship and complex dynamic laws between the data.

[0060] Coupling analysis refers to the process of jointly calculating and mapping various data processing results based on the nonlinear and multidimensional correlation between multi-source heterogeneous data, using mathematical modeling and statistical learning methods, to reveal their inherent dynamic interaction mechanism and structural characteristics, and output comprehensive feature data with a high degree of information fusion.

[0061] Joint anomaly data refers to the set of anomaly features obtained by coupling analysis of nonlinearly processed pedestrian flow data and vibration adjustment data, followed by normalization.

[0062] Weighted fusion refers to the process of combining two or more sets of feature data from different sources and of different types by assigning a weight coefficient to each dataset and combining them linearly or nonlinearly according to a certain fusion strategy, ultimately outputting a fused comprehensive index (such as a comprehensive early warning index).

[0063] The comprehensive early warning index is a multi-dimensional structural safety anomaly comprehensive assessment index obtained by differential analysis, nonlinear coupling, weighted fusion, and normalization processing based on the first flow of people data, the second flow of people data, the first vibration data, the second vibration data, and load data.

[0064] The seismic isolation parameter adjustment mapping model refers to a high-dimensional mapping relationship model based on multi-source complex data input and constructed through a multi-layer nonlinear deep learning network structure.

[0065] Seismic isolation bearing design parameters refer to the set of key technical indicators and parameters used to guide and determine the structural design and adjustment of intelligent seismic isolation bearings at subway transfer points.

[0066] In this application, the ordinal numbers such as "first" and "second" mentioned are used to distinguish multiple objects, and are not used to limit the order, timing, priority or importance of multiple objects; for example, the preset abnormal threshold and the adjusted abnormal threshold are only used to distinguish different abnormal thresholds, and do not indicate the difference in priority or importance of the two abnormal thresholds.

[0067] This solution enhances data authenticity through generative adversarial networks and improves vibration prediction accuracy by combining time-frequency analysis. The physical attribution mechanism established by this method transforms mathematical differences into operable engineering parameters. Through the above technical solutions, the problem of analyzing the difference between vibration prediction and measured data is effectively solved, and abstract mathematical differences are transformed into specific engineering adjustment parameters. The established composite early warning system can accurately identify the correlation between abnormal pedestrian flow and structural response, and improve the pertinence of seismic isolation parameter design.

[0068] The above describes the complete scheme of the intelligent seismic isolation bearing design method driven by generative adversarial networks. The following section introduces the acquisition of the first passenger flow data, the first vibration data, and the subway load data at transfer points, specifically including:

[0069] Raw pedestrian flow data at transfer points is collected through a ground pressure sensor array; the raw pedestrian flow data includes, but is not limited to, ground pressure, passenger trajectory, and dwell time.

[0070] The first vibration data of the transfer point is collected by an array of accelerometers; the first vibration data includes, but is not limited to, structural foundation vibration response data, structural strain response changes, etc.

[0071] Raw load data of the subway is collected using trackside strain gauges; the raw load data includes, but is not limited to, the subway's dynamic load, load, static load, and operating status.

[0072] Preprocessing operations, including but not limited to denoising and outlier removal, are performed on the original pedestrian flow data and original load data. The preprocessed original pedestrian flow data and original load data are then segmented at multiple scales to obtain the first pedestrian flow data and the original load sub-data.

[0073] Based on the original payload sub-data, the output payload data is generated through a generative adversarial network.

[0074] Among them, the ground pressure sensing array refers to an array-type data acquisition device composed of multiple pressure sensors, which can be implemented using distributed piezoelectric sensors or fiber optic pressure sensors.

[0075] Raw pedestrian flow data refers to a multi-dimensional, multi-modal raw observation dataset of crowd flow and behavior, which is directly collected at subway transfer points by sensors and other equipment without any algorithm processing or only basic preprocessing (such as noise reduction and anomaly removal).

[0076] An accelerometer sensor array refers to a vibration monitoring device composed of multiple triaxial accelerometers, which can be implemented using MEMS accelerometers or piezoelectric accelerometers.

[0077] Trackside strain gauges are dynamic strain measurement devices installed near subway tracks, which can be implemented using resistive strain gauges or fiber optic grating sensors.

[0078] Raw load data refers to the basic data collected directly from various sensors in the subway transfer area, reflecting the operating status of subway vehicles and the dynamic and static load information acting on the structure.

[0079] Preprocessing operations include denoising and outlier removal, which can be achieved using wavelet thresholding denoising algorithms and the 3σ criterion based on statistical principles.

[0080] Multi-scale segmentation refers to segmenting data according to time intervals or train operation status, which can be achieved by using a sliding window algorithm or an event-triggered segmentation method.

[0081] Raw load sub-data refers to the raw load data collected at transfer points by sensors such as trackside strain gauges. After preprocessing such as noise reduction and outlier removal, the overall load data is divided into several segments of local load data according to the different operating states of the train (such as acceleration, constant speed, braking, stopping, etc.).

[0082] The following section provides a detailed explanation of the multi-scale segmentation of the preprocessed raw pedestrian flow data and raw load data:

[0083] Among them, the preprocessed raw pedestrian flow data is divided according to time intervals to obtain the first pedestrian flow data;

[0084] The preprocessed raw load data is segmented according to the train operating status to obtain raw load sub-data.

[0085] The following section details how to output payload data using a generative adversarial network based on the original payload sub-data:

[0086] Specifically, the first pedestrian flow data and the original load sub-data are asymmetrically mapped to form a pedestrian flow-load data set;

[0087] The pedestrian-load data set is subjected to feature standardization and dimension unification to obtain a joint feature data pair of pedestrian-load.

[0088] The combined feature data of pedestrian flow and load is input into a pre-constructed nonlinear interactive mapping network, and the composite load data is output.

[0089] The nonlinear interactive mapping network includes an input layer, a multi-layer nonlinear transformation module, an interactive attention mechanism, and a mapping output layer.

[0090] The input layer receives joint feature data pairs after feature standardization and dimensional unification.

[0091] The multi-layer nonlinear transformation module consists of multi-layer deep neural networks, combining convolutional neural network and recurrent neural network modules; the convolutional neural network is responsible for spatial feature extraction, and the recurrent neural network module is responsible for time series feature capture.

[0092] Interactive attention mechanisms are used to weightedly fuse spatial features and time-series features to obtain fused features;

[0093] The mapping output layer is used to map the fused features extracted by deep learning into composite payload data.

[0094] Based on the composite payload data, a generative adversarial network (GAN) is used to output payload data; the GAN consists of a generator and a discriminator.

[0095] The generator employs a multi-layer convolutional neural network and a recurrent neural network structure.

[0096] The discriminator employs a multi-scale discriminant structure;

[0097] Input the composite load data into the generator and output the composite load disturbance data;

[0098] The composite load disturbance data is input into the discriminator, and the reasonableness score of the composite load disturbance data is calculated. The specific calculation formula is as follows:

[0099]

[0100] In the formula, This represents the reasonableness score of the discriminator for the composite load disturbance data, with a value range of [0,1]. This represents the first flow of people output by the generator. and original payload sub-data The composite load perturbation data after cross-fusion This represents the set of parameters for the discriminator network. This represents the discriminator neural network mapping function;

[0101] Determine whether the rationality score is greater than the preset score threshold; otherwise, output the corresponding original load sub-data as load data.

[0102] The weighting adjustment factor is then calculated based on the rationality score, and the specific calculation formula is as follows:

[0103]

[0104] In the formula, Indicates the first flow of people data and original payload sub-data Weight correction factor when performing asymmetric mapping. This represents the learning rate hyperparameter. The original payload sub-data is corrected according to the weight correction factor, and the corrected original payload sub-data is used as the payload data and output.

[0105] The above describes the acquisition of initial pedestrian flow data, initial vibration data, and subway load data at transfer points. The following describes how, based on the initial pedestrian flow data, a generative adversarial network (GAN) is used to output second pedestrian flow data representing the pedestrian flow load. Specifically, this includes:

[0106] Asymmetric mapping is performed on the first pedestrian flow data and the original load sub-data to form a pedestrian flow-load data set;

[0107] The pedestrian-load data set is subjected to feature standardization and dimension unification to obtain a joint feature data pair of pedestrian-load.

[0108] The combined feature data of pedestrian flow and load is input into a pre-constructed nonlinear interactive mapping network, and the composite load data is output.

[0109] Based on the composite payload data, a second person flow data is output through a generative adversarial network.

[0110] The method of obtaining weight correction factors from composite load data using generative adversarial networks has already been introduced above, and will not be elaborated further here;

[0111] The first flow data is corrected based on the weighting correction factor to obtain the second flow data.

[0112] The above technical solutions can effectively simulate and supplement the implicit patterns of pedestrian flow changes and load fluctuations in the actual operation of transfer points, realize high-fidelity reconstruction and enhancement of pedestrian load data, provide richer and more accurate input information for subsequent load prediction and seismic isolation bearing design, and improve the overall system's response capability and early warning accuracy to complex dynamic environments.

[0113] The above describes how, based on the first pedestrian flow data, a second pedestrian flow data model representing the pedestrian flow load is output through a generative adversarial network. The following describes how, based on the second pedestrian flow data, load data, and the first vibration data, a second vibration data model representing vibration prediction is generated using a pre-built response generation model. Specifically, this includes:

[0114] Construct a response generation model; the response generation model includes an input layer, a multi-channel sub-network layer, and an output layer;

[0115] The second pedestrian flow data, load data, and first vibration data are received through the input layer and then subjected to time-frequency decomposition processing to obtain a time-frequency feature subset.

[0116] The time-frequency feature subset is input into the multi-channel sub-network layer, the structural vibration response data of the time-frequency feature subset under that channel is output, and the error value between the structural vibration response data and the first vibration data is calculated.

[0117] Adjust the parameters of the channel sub-network based on the error value;

[0118] Based on the parameters of the adjusted channel subnetwork, the intermediate structure vibration response data of the time-frequency feature subset under that channel is output. The intermediate structure vibration response data of all channels are combined to obtain the second vibration data, which is then output through the output layer.

[0119] Among them, time-frequency decomposition refers to the process of converting time series data into time-frequency domain features, which can be achieved by a combination of short-time Fourier transform and wavelet transform.

[0120] The time-frequency feature subset refers to the set of multi-dimensional features within a specific frequency range and corresponding time period extracted from multi-source input data (second pedestrian flow data, load data, and first vibration data) through joint time-frequency analysis methods.

[0121] Structural vibration response data refers to the collection of dynamic vibration characteristic information of a structure (such as a building or structure supported by intelligent seismic isolation bearings) under the action of external excitations (such as pedestrian flow, load, environmental vibration, etc.).

[0122] Error value refers to the difference between the predicted vibration response and the actual vibration data. Specifically, it can be calculated by combining the mean square error with the summation of cross-channel differences.

[0123] Intermediate structural vibration response data refers to the time-frequency domain data reflecting the structural vibration response characteristics output by each frequency channel after calculation by the subnetwork under the current parameter state within the multi-channel subnetwork layer. This data has not yet been finally integrated and output as complete "second vibration data".

[0124] The above content will be explained in detail below:

[0125] Specifically, short-time Fourier transform and wavelet transform are applied to the second pedestrian flow data, load data, and first vibration data respectively to obtain pedestrian flow features, load features, and vibration features corresponding to the frequency and time dimensions. The pedestrian flow features, load features, and vibration features of the same frequency are combined to obtain a time-frequency feature subset.

[0126] The time-frequency feature subset is input into the multi-channel sub-network layer. Each channel corresponds to a time-frequency feature subset at a certain frequency. The structural vibration response data of the time-frequency feature subset at that channel is output, and the error value between the structural vibration response data and the first vibration data is calculated. The specific calculation formula is as follows:

[0127]

[0128] In the formula, Indicates the first Channel error value, Indicates the first Structural vibration response data generated by the channel, Indicates the first The first vibration data at the corresponding frequency of the channel. This represents the weighting adjustment coefficient. Indicates the first The summation of the differences in the responses of one channel to other channels. Indicates the first Structural vibration response data generated by the channel, This indicates the calculation of mean square error;

[0129] The parameters of this channel sub-network are adjusted based on the error value, and the specific calculation formula is as follows:

[0130]

[0131] In the formula, Indicates the first New parameters for the channel subnetwork, Indicates the first The old parameters of the channel subnetwork, Indicates the learning rate. Indicates the cross-feedback adjustment parameter. Indicates the error value for the first Gradients of the channel subnetwork parameters Indicates the error values ​​of other channels relative to the first... The influence of gradients on the parameters of the channel subnetwork;

[0132] Based on the parameters of the adjusted channel subnetwork, the intermediate structure vibration response data of the time-frequency feature subset under that channel is output. The intermediate structure vibration response data of all channels are combined to obtain the second vibration data, which is then output through the output layer.

[0133] This solution, through time-frequency decomposition and multi-channel processing mechanisms, can accurately capture the coupling relationship between pedestrian flow, load, and vibration at specific frequencies. Simultaneously, through a cross-channel error feedback mechanism, it avoids mutual interference between prediction results across different frequency bands, significantly improving the accuracy of vibration prediction. This technical solution effectively solves the problem of existing vibration prediction models neglecting frequency domain feature correlation, achieving precise matching of multi-source data in the time-frequency domain. By combining independent optimization of individual channels with cross-channel collaborative adjustment, it significantly improves the prediction accuracy of structural vibration response under complex working conditions, providing a reliable data foundation for subsequent seismic isolation parameter optimization.

[0134] The above describes how, based on the second pedestrian flow data, load data, and first vibration data, a pre-built response generation model is used to generate second vibration data characterizing vibration prediction. The following describes the difference analysis of the first and second vibration data to form difference features, and the feature attribution processing of these difference features to obtain vibration adjustment data, specifically including:

[0135] Perform difference analysis on the first vibration data and the second vibration data, including but not limited to time domain difference analysis, frequency domain difference analysis, phase drift analysis, etc.

[0136] The difference analysis results are normalized, and the normalized difference analysis results are then combined with multidimensional features to obtain the difference features.

[0137] The differential features are subjected to feature attribution processing, which divides and maps the differential features to different physical attribution categories, including but not limited to structural stiffness, damping, mass and frequency offset, thereby forming vibration adjustment data.

[0138] Among them, multidimensional feature concatenation refers to integrating the normalized time-domain, frequency-domain, and phase difference features into a unified feature vector according to the time series. Specifically, it can be implemented by tensor concatenation or feature fusion network to construct high-dimensional difference representation.

[0139] The feature attribution process is explained in detail below, and the workflow is as follows:

[0140] Based on the change type of each difference analysis result within the difference feature and the corresponding preset discrimination threshold, the difference features are classified, and then the corresponding physical property anomalies are determined. For example, displacement difference exceeding the threshold is attributed to stiffness anomaly, and frequency shift corresponds to dynamic characteristic change.

[0141] By calling up historical first vibration data, historical second vibration data, and physical attribution results, a difference analysis is performed on historical first vibration data and historical second vibration data to obtain historical difference characteristics;

[0142] Based on historical differences and physical attribution results, a mapping relationship between differences and attribution vectors is established through linear regression.

[0143] Based on the difference characteristics, the attribution vector is obtained through the mapping relationship between the difference characteristics and the attribution vector;

[0144] The attribution vector is analyzed to extract the direction and amplitude of physical property change corresponding to each attribution component, thereby forming the corresponding dynamic vibration adjustment factor, including but not limited to stiffness adjustment factor, damping adjustment factor, frequency adjustment factor, etc.

[0145] By combining the above dynamic vibration adjustment factors, vibration adjustment data is obtained.

[0146] To better understand the above content, an application scenario example is given below:

[0147] Taking a subway transfer point in the city as an example, intelligent seismic isolation bearings were deployed to reduce structural vibrations caused by pedestrian flow and train operation;

[0148] The system uses a high-precision sensor array (model AX-9000, sampling rate 1000Hz, bandwidth 0.5Hz~300Hz) to collect first vibration data, first pedestrian flow data, and load data in real time, and calculates second vibration data.

[0149] Synchronization Protocol: The IEEE 1588 Precision Time Protocol (PTP) is used to achieve hardware-level time synchronization of multi-sensor data, with time error controlled within ±500μs;

[0150] Difference analysis:

[0151] Time-domain analysis: Calculate the difference in peak acceleration, root mean square (RMS) value, and signal envelope between the first and second vibration data.

[0152] Frequency domain analysis: Perform Fast Fourier Transform (FFT) on the first and second vibration data to compare the differences in peak main frequency and spectral energy distribution;

[0153] Phase drift analysis: The instantaneous phase is extracted using Hilbert transform, and the phase drift of the first and second vibration data in the key frequency range is calculated;

[0154] Feature normalization and splicing:

[0155] Z-score normalization is applied to time-domain differences (peak difference, RMS difference), frequency-domain differences (dominant frequency offset, band energy difference), and phase drift.

[0156] The normalized differential features are concatenated in chronological order to form a multidimensional differential feature matrix. The time window is set to 5 seconds and the window sliding step is 1 second.

[0157] Feature attribution processing:

[0158] Threshold setting:

[0159] The peak acceleration difference threshold is 0.15 m / s².

[0160] The main frequency offset threshold is ±3Hz;

[0161] The phase drift threshold is ±10°;

[0162] Anomaly detection:

[0163] When the peak difference exceeds the threshold, it is attributed to abnormal stiffness changes;

[0164] The frequency shift is attributed to damping or mass change;

[0165] Vibration signal phase drift correlates with dynamic frequency response deviation;

[0166] Historical data citation: retrieve the first and second vibration data collected by sensors at the same location within the past 12 months, and calculate the historical difference characteristics;

[0167] Mapping relationship established:

[0168] A multiple linear regression model is used, with historical difference characteristics as input and attribution vector (including stiffness adjustment factor, damping adjustment factor, and frequency adjustment factor) as output.

[0169] Based on the attribution vector, the current differential characteristics are mapped to obtain the dynamic vibration adjustment factor;

[0170] Application of dynamic vibration adjustment factor:

[0171] Example of stiffness adjustment factor: 1.12 (12% increase compared to the reference stiffness);

[0172] Example of damping adjustment factor: 0.95 (5% reduction compared to the reference damping);

[0173] Example of frequency adjustment factor: main frequency offset -2.3Hz;

[0174] Based on the aforementioned dynamic vibration adjustment factor, vibration adjustment data is obtained.

[0175] This solution decomposes vibration differences into quantitative adjustment requirements through multidimensional difference analysis and physical attribution mapping. Through the above technical solution, this application can transform the difference between vibration prediction data and actual monitoring data into specific physical attribute adjustment requirements, solving the problems of single dimension and ambiguous attribution in the existing technology, and improving the accuracy and interpretability of the adjustment of seismic isolation bearing design parameters.

[0176] The above describes the difference analysis of the first and second vibration data to form difference features. Feature attribution processing is then performed on these difference features to obtain vibration adjustment data. The following describes the nonlinear processing of the first and second pedestrian flow data, followed by a coupled analysis of the nonlinear processing results and vibration adjustment data to generate joint anomaly data. Specifically, this includes:

[0177] Feature extraction is performed on the first and second flow data, including but not limited to density peak features, stagnation period features, and flow change features;

[0178] The feature extraction results are combined to obtain abnormal behavior data;

[0179] Time-point matching is performed on abnormal behavior data and vibration adjustment data to generate behavior-vibration data pairs;

[0180] Spatial mapping of behavior-vibration data pairs yields joint anomaly data.

[0181] Among them, abnormal behavior data refers to the set of characteristic data that reflects potential safety hazards, disturbance risks or unexpected flow patterns in the flow of people, which is obtained by extracting and combining features from the first flow data and the second flow data.

[0182] Time point matching refers to aligning data with different time resolutions using interpolation methods, specifically linear interpolation or cubic spline interpolation.

[0183] Spatial mapping refers to projecting high-dimensional feature vectors onto a low-dimensional space. This can be achieved using principal component analysis or an autoencoder to extract key coupling features from the data.

[0184] The following details the spatial mapping of behavior-vibration data pairs to obtain joint anomaly data:

[0185] Multidimensional correlation analysis was used to calculate the correlation intensity index of each behavior-vibration data pair, and the correlation intensity indexes were combined to generate a correlation intensity matrix.

[0186] Assign weights to the matrix elements within the correlation strength matrix;

[0187] The matrix elements are weighted and fused according to the time series to extract the core coupling feature vector;

[0188] Spatial mapping of the core coupling feature vectors yields joint anomaly data.

[0189] This application, through nonlinear processing and multidimensional correlation analysis, can more accurately identify the complex relationship between abnormal pedestrian flow and vibration adjustment. Through the above technical solution, this application can solve the problem of anomaly detection error caused by ignoring nonlinear correlation and time matching deviation in the prior art. By improving the accuracy of anomaly identification through spatiotemporal coupling analysis, it can provide more reliable anomaly data input for the optimization of seismic isolation bearing design parameters, thereby improving the level of refinement of structural safety assessment.

[0190] The above describes the nonlinear processing of the first and second pedestrian flow data, followed by coupled analysis of the nonlinear processing results and vibration adjustment data to generate joint anomaly data. The following section describes the generation of a composite anomaly risk mapping matrix; please refer to the documentation. Figure 2 , Figure 2 This is a flowchart illustrating the process of generating a composite anomaly risk mapping matrix provided in an embodiment of this application. Generating the composite anomaly risk mapping matrix specifically includes:

[0191] Non-windowed time-series accumulation of abnormal behavior data is used to form an anomaly intensity curve;

[0192] Preset a time delay interval, and calculate the time delay correlation coefficient between the anomaly intensity curve and the load data within the time delay interval;

[0193] Arrange all the time delay correlation coefficients corresponding to the time delay interval in order to obtain the time delay correlation matrix;

[0194] A composite anomaly risk mapping matrix is ​​obtained by performing nonlinear cross mapping on the time-delay correlation matrix and joint anomaly data.

[0195] Among them, non-windowed time series accumulation refers to the accumulation processing of abnormal behavior data without relying on a fixed time window. Specifically, it can be achieved by using a Gaussian kernel function to perform smooth weighted accumulation of time series data.

[0196] An anomaly intensity curve is a time-series curve that reflects the characteristics of anomaly accumulation and mutation over time, formed based on the acquired abnormal behavior data and through a specific mathematical smoothing and accumulation process.

[0197] The time lag interval refers to a set of time intervals preset for lag alignment when analyzing the correlation between abnormal behavior data and load data. It is usually represented as a set of discrete time steps or continuous time windows.

[0198] The time-delay correlation coefficient measures the degree of correlation between the anomaly intensity curve and the load data under different time delays. Specifically, it can be calculated using the ratio of covariance to standard deviation.

[0199] The time-delay correlation matrix is ​​a two-dimensional matrix formed by arranging a set of time-delay correlation coefficients calculated based on the anomaly intensity curve and load data under different time lag relationships, in lag order.

[0200] Nonlinear cross mapping refers to combining the time-delay correlation matrix with joint anomaly data using a nonlinear function. Specifically, a composite transformation can be performed using the hyperbolic tangent function and the natural logarithm function.

[0201] The composite anomaly risk mapping matrix refers to a multidimensional data matrix that comprehensively characterizes the evolution trend and potential risk intensity of structural anomalies under multiple time delays and multiple factors, based on the cross-time lag correlation and nonlinear cross-mapping relationship between anomaly behavior data and load data.

[0202] The above content will be explained in detail below:

[0203] The abnormal behavior data is accumulated over time without windowing to form an anomaly intensity curve, and the specific calculation formula is as follows:

[0204]

[0205] In the formula, Indicates time The abnormal intensity smoothing value, Indicates time Abnormal behavior data, The standard deviation of the Gaussian kernel function is represented by... Represents the natural exponential function;

[0206] A preset time delay interval is used to calculate the time delay correlation coefficient between the anomaly intensity curve and the load data within the time delay interval. The specific calculation formula is as follows:

[0207]

[0208] In the formula, This represents the time-delay correlation coefficient, ranging from [-1, 1]. This represents the average value of the anomaly intensity curve. Indicates the load data at time... The value, This represents the average value of the load data. Represents time-delay variables;

[0209] Arrange all the time delay correlation coefficients corresponding to the time delay interval in order to obtain the time delay correlation matrix;

[0210] A composite anomaly risk mapping matrix is ​​obtained by performing a nonlinear cross-mapping on the time-delay correlation matrix and the joint anomaly data. The specific calculation formula is as follows:

[0211]

[0212] In the formula, This represents the composite anomaly risk mapping matrix. Represents the time-delay correlation matrix. Indicates combined abnormal data. Represents the hyperbolic tangent function. Represents the natural logarithm function. This represents the matrix magnification adjustment factor. This indicates an increased risk adjustment factor.

[0213] The above describes the generation of a composite anomaly risk mapping matrix. The following describes the generation of a comprehensive early warning index; please refer to that description. Figure 3 , Figure 3 This is a schematic diagram of the process for generating a comprehensive early warning index provided in an embodiment of this application. Generating the comprehensive early warning index specifically includes:

[0214] Normalize the differential features and joint outlier data separately;

[0215] The normalized differential characteristics and joint abnormal data are weighted and fused to obtain the early warning index;

[0216] Input the differential characteristics, time-lag correlation matrix, and composite anomaly risk mapping matrix into the pre-constructed structural health status evolution model, and output health status data;

[0217] Based on health status data, the early warning index is dynamically weighted to obtain a comprehensive early warning index.

[0218] Normalization refers to the process of converting feature data of different dimensions or magnitudes into a unified standard range, which can be achieved by using maximum-minimum scaling or Z-score normalization.

[0219] The early warning index is a numerical indicator used to characterize the potential health risk level of a current structural system, obtained by weighted fusion of differential characteristics and joint abnormal data.

[0220] Structural health status evolution model refers to a mathematical model that assesses changes in structural status through time series analysis and correlation analysis. Specifically, it can be constructed using hidden Markov models or long short-term memory networks.

[0221] Health status data refers to a set of multi-dimensional and multi-factor comprehensive indicators used to characterize the current health status of a structure, which is obtained by comprehensively calculating input data (difference characteristics, time-lag correlation matrix, composite abnormal risk mapping matrix) and structural health status evolution model.

[0222] Dynamic weighting refers to dynamically adjusting the weighting coefficients of the early warning index based on real-time changes in structural health status data. This makes the weighting process not only dependent on fixed weights, but also adaptively adjusting the weight distribution in combination with the current structural status, so as to achieve a risk early warning index output that is more in line with the actual evolution of the status.

[0223] The following details how to input differential characteristics, time-lag correlation matrices, and composite anomaly risk mapping matrices into a pre-constructed structural health status evolution model to output health status data:

[0224] The structural health status evolution model includes a time-series evolution unit, a correlation analysis unit, and a health projection unit;

[0225] The temporal evolution unit receives differential characteristics, time-delay correlation matrix, and composite anomaly risk mapping matrix;

[0226] Sliding window analysis is performed on the differential characteristics to extract statistical features (mean, variance, etc.).

[0227] Perform fluctuation analysis on statistical characteristics to identify abnormal sliding windows;

[0228] The time evolution trend parameters are calculated by weighted cumulative calculation;

[0229] The correlation analysis unit calculates the correlation strength of the differential features, time-lag correlation matrix, and composite anomaly risk mapping matrix pairwise, thereby forming a correlation strength matrix;

[0230] The health projection unit predefines a multidimensional set of health status indicators, performs factor decomposition on the correlation strength matrix, and then maps it to the time evolution trend parameters to obtain the health status factor mapping.

[0231] The predefined multidimensional health status index set is weighted and calculated based on the health status factor mapping, and the weighted calculation results are normalized in multiple dimensions to obtain and output the health status data.

[0232] This scheme achieves dynamic optimization of the early warning index through a structural health status evolution model. It effectively captures the temporal evolution of structural status by combining sliding window analysis and correlation strength matrix, and accurately identifies the core factors affecting health status using factor decomposition technology. Through the above technical solutions, this application can adjust the weight of the early warning index according to the dynamic evolution of structural health status, effectively integrate time-domain statistical characteristics and frequency-domain correlation characteristics, accurately identify the composite risks caused by sudden changes in pedestrian load and abnormal vibration response, and provide a comprehensive evaluation index for seismic isolation bearing design that takes into account both real-time monitoring data and long-term evolution trends, thereby improving the accuracy of seismic isolation parameter adjustment and structural safety assurance capabilities.

[0233] The above describes the generation of the comprehensive early warning index. The following describes the output of the intermediate seismic isolation bearing design parameters. Please refer to this section. Figure 4 , Figure 4 This is a flowchart illustrating the output of intermediate seismic isolation bearing design parameters provided in an embodiment of this application. The output of intermediate seismic isolation bearing design parameters specifically includes:

[0234] Construct a seismic isolation parameter adjustment mapping model;

[0235] Feature extraction was performed on the comprehensive early warning index, second pedestrian flow data, load data, second vibration data, and composite abnormal risk mapping matrix, respectively.

[0236] The feature extraction results are normalized and input into the seismic isolation parameter adjustment mapping model to output the original seismic isolation bearing design parameters;

[0237] Determine whether the original seismic isolation bearing design parameters are greater than the preset safety threshold. If so, adjust the seismic isolation parameters and adjust the mapping model, and re-output the original seismic isolation bearing design parameters until the safety threshold is met. Use the original seismic isolation bearing design parameters that meet the safety threshold as the intermediate seismic isolation bearing design parameters.

[0238] Among them, the original seismic isolation bearing design parameters refer to the set of preliminary seismic isolation bearing design parameters that are directly calculated and output by the seismic isolation parameter adjustment mapping model under the current input data and have not yet been compared and corrected with the safety threshold.

[0239] The safety threshold refers to the minimum performance boundary value required to ensure both seismic isolation effect and structural safety for a designed seismic isolation bearing under a specific operating environment.

[0240] The design parameters of the intermediate isolation bearing refer to the original design parameters of the isolation bearing, which are derived from the comprehensive early warning index, second pedestrian flow data, load data, second vibration data and composite abnormal risk mapping matrix, and are determined by the isolation parameter adjustment mapping model and constrained by safety threshold.

[0241] The following details the normalization process for the feature extraction results, the input of seismic isolation parameters to adjust the mapping model, and the output of the original seismic isolation bearing design parameters:

[0242] The seismic isolation parameter adjustment mapping model includes an input layer, a feature extraction layer, a cross-coupling layer, and an output layer.

[0243] The input layer receives the comprehensive early warning index, second pedestrian flow data, load data, second vibration data, and composite abnormal risk mapping matrix;

[0244] The feature extraction layer performs multi-scale feature extraction on the received data.

[0245] The cross-coupling layer constructs a multi-dimensional mapping space through a multi-layer attention mechanism, deeply integrates different data dimensions, and dynamically adjusts the contribution of different input data to the original seismic isolation bearing design parameters.

[0246] The output layer outputs the original seismic isolation bearing design parameters based on the deep fusion results.

[0247] This solution achieves collaborative analysis of pedestrian flow, load, vibration, and risk data by constructing a multi-level feature extraction and dynamic coupling mechanism. By setting safety thresholds and feedback adjustment mechanisms, it ensures that the output parameters meet engineering specifications. Through the above technical solution, the problems of single-dimensional calculation of seismic isolation parameters and lack of dynamic adjustment capability in existing technologies can be solved. At the same time, through multi-level feature fusion and safety verification mechanisms, the failure of design parameters due to data anomalies or model deviations can be avoided.

[0248] The above describes the design parameters for the output intermediate seismic isolation bearing. The following describes the specific design parameters for the output seismic isolation bearing:

[0249] Based on the design parameters of the intermediate seismic isolation bearing, the pedestrian flow-load-vibration conditions at the transfer point are reconstructed through simulation to obtain simulated pedestrian flow data, simulated vibration data, and simulated load data.

[0250] The simulated pedestrian flow data, simulated vibration data, and simulated load data were compared with the second pedestrian flow data, second vibration data, and load data respectively to obtain the risk difference data.

[0251] Each risk difference data is checked to see if it exceeds the preset abnormal threshold. If so, the seismic isolation parameters are adjusted to regulate the mapping model and the original seismic isolation bearing design parameters are re-output.

[0252] Otherwise, a multi-time series data matrix is ​​constructed based on the simulated pedestrian flow data, simulated vibration data, simulated load data, second pedestrian flow data, second vibration data, and load data;

[0253] Data evolution analysis is performed on the multi-time series data matrix to obtain the mapping relationship of seismic isolation parameter adjustment, and then the design parameters of seismic isolation bearings are obtained.

[0254] Among them, simulated pedestrian flow data refers to spatiotemporal sequence data that reflects the flow status of pedestrians in the transfer point area, generated based on the design parameters of the intermediate seismic support through a multi-physics simulation model and a dynamic crowd behavior simulation system.

[0255] Simulated vibration data refers to the dataset of time-series vibration signals obtained by simulating the vibration response of the transfer point structure under actual working conditions based on the design parameters of the intermediate seismic isolation bearing, using a physical model and a multi-field coupled simulation platform, and through digital twin technology.

[0256] Simulation load data refers to the set of time-series data of structural stress state under different pedestrian flow and vibration conditions in the transfer point area, which is obtained by simulation calculation based on the design parameters of the intermediate seismic isolation bearing and constructed by the digital simulation model.

[0257] Risk difference data refers to the set of numerical differences between various key indicators after quantitative comparison of pedestrian flow data, vibration data, and load data from different sources (simulation and actual monitoring) under the same spatiotemporal transfer point environment;

[0258] An anomaly threshold is a numerical limit used to determine whether the difference between simulated data and actual measured data exceeds the normal fluctuation range of the system.

[0259] A multi-time series data matrix refers to a matrix data structure formed by arranging multi-dimensional time series data from different sources and of different types synchronously along a time axis and then integrating them in a structured manner.

[0260] Data evolution analysis refers to the process of deeply mining and quantitatively describing the dynamic change trends, pattern shifts, structural reorganizations, and potential coupling relationships of input simulation and measured datasets in the spatiotemporal dimensions based on multi-time series and multi-source heterogeneous data matrices, using nonlinear, multi-dimensional cross-correlation and time series dynamic modeling techniques.

[0261] The mapping relationship for seismic isolation parameter adjustment refers to the mathematical or statistical model established between multi-source, multi-time-series simulation data and measured data.

[0262] The following details the data evolution analysis of the multi-time series data matrix to obtain the mapping relationship for seismic isolation parameter adjustment, and thus obtain the design parameters of the seismic isolation bearings:

[0263] Temporal feature extraction algorithms (such as sliding window statistics, Fourier transform, and wavelet transform) are used to mine local and global features of multi-time series data matrices to obtain frequency domain features, time domain fluctuation features, and trend features, forming a feature matrix;

[0264] Nonlinear time series analysis methods (such as multivariate mutual information, Granger causality detection, and dynamic time warping DTW) are used to detect complex coupling relationships between different data within the feature matrix and identify key influencing factors.

[0265] Recurrent neural networks, long short-term memory networks, or time-series graph neural networks are used to model the feature matrix, obtain the mapping relationship of seismic isolation parameter adjustment, and capture the dynamic evolution law of multi-time series data.

[0266] During the training process of the mapping relationship for seismic isolation parameter adjustment, a time-dependent regularization term is introduced to constrain the model's effective capture of long-term dependencies and prevent overfitting and information loss.

[0267] The output of the seismic isolation parameter adjustment mapping relationship is the seismic isolation bearing design parameter, that is, inputting the multi-time series data characteristics at the current moment and outputting the corresponding seismic isolation bearing design parameter.

[0268] This solution ensures that the design parameters meet dynamic safety requirements by judging the difference threshold and iteratively adjusting it. Through the above technical solution, this application can effectively verify the reliability of intermediate design parameters under simulated working conditions, avoid structural safety hazards caused by prediction deviations, and capture the correlation between dynamic load and vibration response through multi-time series data evolution analysis, so that the seismic isolation bearing design parameters can meet both real-time and long-term stability requirements, significantly improving the self-adaptive capability and structural safety level of the seismic isolation system.

[0269] Example 2:

[0270] Please see Figure 5 A generative adversarial network-driven intelligent seismic isolation bearing design system includes:

[0271] The data acquisition module is used to acquire the first pedestrian flow data, the first vibration data, and the subway load data at the transfer point;

[0272] The first pedestrian flow data processing module is used to output second pedestrian flow data representing pedestrian flow load by generating an adversarial network based on the first pedestrian flow data.

[0273] The second vibration data generation module is used to generate second vibration data characterizing vibration prediction based on the second pedestrian flow data, load data, and first vibration data through a pre-built response generation model.

[0274] The vibration adjustment data generation module is used to perform difference analysis on the first vibration data and the second vibration data to form difference features, and to perform feature attribution processing on the difference features to obtain vibration adjustment data.

[0275] The joint anomaly analysis module is used to perform nonlinear processing on the first and second pedestrian flow data, and to perform coupled analysis on the nonlinear processing results and vibration adjustment data to generate joint anomaly data.

[0276] The comprehensive early warning index calculation module is used to weight and fuse differential characteristics with joint abnormal data to obtain a comprehensive early warning index;

[0277] The seismic isolation bearing design module is used to input the comprehensive early warning index, second pedestrian flow data, load data, and second vibration data into the pre-constructed seismic isolation parameter adjustment mapping model, and output the seismic isolation bearing design parameters.

[0278] This embodiment has the same technical effects as Embodiment 1.

[0279] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0280] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A generative adversarial network-driven intelligent seismic isolation bearing design method, applied to the design of intelligent seismic isolation bearings at subway transfer points, characterized in that... Includes the following steps: Acquire the first passenger flow data, the first vibration data, and the subway load data at the transfer point; Based on the first pedestrian flow data, a second pedestrian flow data representing the pedestrian flow load is output through a generative adversarial network; Based on the second pedestrian flow data, the load data, and the first vibration data, a second vibration data characterizing vibration prediction is generated using a pre-built response generation model. A difference analysis is performed on the first vibration data and the second vibration data to form difference features. Feature attribution processing is then performed on the difference features to obtain vibration adjustment data. The first pedestrian flow data and the second pedestrian flow data are subjected to nonlinear processing, and the nonlinear processing result and the vibration adjustment data are coupled and analyzed to generate joint anomaly data; The differential features are weighted and fused with the joint anomaly data to obtain a comprehensive early warning index; The comprehensive early warning index, the second pedestrian flow data, the load data, and the second vibration data are input into the pre-constructed seismic isolation parameter adjustment mapping model, and the seismic isolation bearing design parameters are output.

2. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 1, characterized in that: Acquiring the first passenger flow data, the first vibration data, and the subway load data at the transfer point specifically includes: Raw pedestrian flow data at transfer points is collected using a ground pressure sensor array. The first vibration data at the transfer point was collected using an array of accelerometer sensors. Raw load data of the subway was collected using trackside strain gauges; The raw pedestrian flow data and raw load data are preprocessed respectively; The preprocessed raw pedestrian flow data and raw load data are segmented at multiple scales to obtain the first pedestrian flow data and the raw load sub-data. Based on the original payload sub-data, the adversarial network outputs payload data.

3. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 1, characterized in that: Based on the second pedestrian flow data, the load data, and the first vibration data, the second vibration data characterizing vibration prediction is generated using a pre-built response generation model, specifically including: Construct a response generation model; the response generation model includes an input layer, a multi-channel sub-network layer, and an output layer; The second pedestrian flow data, the load data, and the first vibration data are received through the input layer, and time-frequency decomposition processing is performed to obtain a time-frequency feature subset. The time-frequency feature subset is input into a multi-channel sub-network layer, and the structural vibration response data of the time-frequency feature subset under that channel is output. The error value between the structural vibration response data and the first vibration data is calculated. Adjust the parameters of the channel sub-network based on the error value; Based on the parameters of the adjusted channel subnetwork, the intermediate structure vibration response data of the time-frequency feature subset under that channel is output. The intermediate structure vibration response data of all channels are combined to obtain the second vibration data, which is then output through the output layer.

4. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 1, characterized in that: A difference analysis is performed on the first vibration data and the second vibration data to form difference features. Feature attribution processing is then performed on the difference features to obtain vibration adjustment data, specifically including: A difference analysis is performed on the first vibration data and the second vibration data, including time-domain difference analysis and frequency-domain difference analysis; The difference analysis results are normalized, and the normalized difference analysis results are then combined with multidimensional features to obtain the difference features. The differential features are subjected to feature attribution processing, and the differential features are divided and mapped to different physical attribution categories, thereby forming vibration adjustment data.

5. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 1, characterized in that: The first and second pedestrian flow data undergo nonlinear processing, and the nonlinear processing results and the vibration adjustment data are coupled for analysis to generate joint anomaly data. Specifically, this includes: Feature extraction is performed on the first pedestrian flow data and the second pedestrian flow data, including density peak features and stagnation period features; The feature extraction results are combined to obtain abnormal behavior data; The abnormal behavior data and the vibration adjustment data are matched at specific times to generate behavior-vibration data pairs. Spatial mapping is performed on the behavior-vibration data pairs to obtain joint anomaly data.

6. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 5, characterized in that: After generating the joint anomaly data, the method further includes performing spatiotemporal aggregation analysis on the anomaly behavior data, specifically including: The abnormal behavior data is accumulated over time in a non-windowed manner to form an abnormality intensity curve; A preset time delay interval is used to calculate the time delay correlation coefficient between the anomaly intensity curve and the load data within the time delay interval. Arrange all the time delay correlation coefficients corresponding to the time delay interval in order to obtain the time delay correlation matrix; A composite anomaly risk mapping matrix is ​​obtained by performing a nonlinear cross-mapping on the time-delay correlation matrix and the joint anomaly data.

7. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 6, characterized in that: The comprehensive early warning index is obtained by weighted fusion of the aforementioned differential features and the joint anomaly data, specifically including: The differential features and the joint anomaly data are respectively normalized; The normalized difference features are weighted and fused with the joint anomaly data to obtain the early warning index. The differential features, the time-lag correlation matrix, and the composite anomaly risk mapping matrix are input into a pre-constructed structural health status evolution model, and health status data are output. Based on the health status data, the early warning index is dynamically weighted to obtain a comprehensive early warning index.

8. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 7, characterized in that: The input of the comprehensive early warning index, second pedestrian flow data, load data, and second vibration data into the pre-constructed seismic isolation parameter adjustment mapping model specifically includes: Construct a seismic isolation parameter adjustment mapping model; Feature extraction is performed on the comprehensive early warning index, the second pedestrian flow data, the load data, the second vibration data, and the composite abnormal risk mapping matrix, respectively. The feature extraction results are normalized and input into the seismic isolation parameter adjustment mapping model to output the original seismic isolation bearing design parameters; Determine whether the original seismic isolation bearing design parameters are greater than the preset safety threshold. If so, adjust the seismic isolation parameter adjustment mapping model and re-output the original seismic isolation bearing design parameters until the safety threshold is met. Use the original seismic isolation bearing design parameters that meet the safety threshold as the intermediate seismic isolation bearing design parameters.

9. The intelligent seismic isolation bearing design method driven by generative adversarial networks according to claim 8, characterized in that: The specific design parameters for output seismic isolation bearings include: Based on the design parameters of the intermediate seismic isolation bearing, the pedestrian flow-load-vibration conditions at the transfer point are reconstructed through simulation to obtain simulated pedestrian flow data, simulated vibration data, and simulated load data. The simulated pedestrian flow data, simulated vibration data, and simulated load data were compared with the second pedestrian flow data, second vibration data, and load data to obtain the risk difference data. Each risk difference data is checked to see if it exceeds the preset abnormal threshold. If so, the seismic isolation parameters are adjusted to regulate the mapping model and the original seismic isolation bearing design parameters are re-output. Otherwise, a multi-time series data matrix is ​​constructed based on the simulated pedestrian flow data, simulated vibration data, simulated load data, second pedestrian flow data, second vibration data, and load data; Data evolution analysis is performed on the multi-time series data matrix to obtain the mapping relationship of seismic isolation parameter adjustment, and then the design parameters of seismic isolation bearings are obtained.

10. A generative adversarial network-driven intelligent seismic isolation bearing design system, characterized in that, include: The data acquisition module is used to acquire the first pedestrian flow data, the first vibration data, and the subway load data at the transfer point; The first pedestrian flow data processing module is used to output second pedestrian flow data representing pedestrian flow load by generating an adversarial network based on the first pedestrian flow data. The second vibration data generation module is used to generate second vibration data characterizing vibration prediction based on the second pedestrian flow data, the load data, and the first vibration data, through a pre-built response generation model. The vibration adjustment data generation module is used to perform difference analysis on the first vibration data and the second vibration data to form difference features, and to perform feature attribution processing on the difference features to obtain vibration adjustment data. The joint anomaly analysis module is used to perform nonlinear processing on the first pedestrian flow data and the second pedestrian flow data, and to perform coupled analysis on the nonlinear processing results and the vibration adjustment data to generate joint anomaly data; The comprehensive early warning index calculation module is used to weight and fuse the differential features with the joint abnormal data to obtain the comprehensive early warning index; The seismic isolation bearing design module is used to input the comprehensive early warning index, the second pedestrian flow data, the load data, and the second vibration data into a pre-constructed seismic isolation parameter adjustment mapping model, and output the seismic isolation bearing design parameters.

Images