Data center seismic isolation performance detection method based on AI intelligence

Abstract

The invention discloses a data center seismic isolation performance detection method based on AI intelligence, and particularly relates to the technical field of seismic isolation performance detection. Earthquake vibration characteristic data are collected in real time on a data center building and equipment, earthquake input characteristics are extracted in combination with a historical earthquake wave database, a seismic isolation performance prediction model is constructed and trained, a foundation shear wave velocity gradient abnormal value and a seismic isolation layer dynamic displacement response deviation ratio are further extracted, a deviation index is comprehensively calculated, and a seismic isolation performance prediction model is constructed and trained. And classification processing is carried out according to the deviation level, especially under the condition of a middle deviation level, foundation physical parameters are dynamically updated, and the model is optimized, so that the accuracy of the prediction model and the self-adaptive capability of the system are remarkably improved, and the shock isolation safety and the continuous operation capability of the data center in a strong shock environment are effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic isolation performance detection, and particularly relates to an AI intelligent-based seismic isolation performance detection method for data centers. Background Art

[0002] The seismic isolation performance detection of a data center refers to testing and evaluating the seismic isolation systems (such as seismic isolation devices, seismic isolation layers, etc.) of the data center building and its key equipment through professional technical means, verifying whether the shock absorption and seismic isolation effects under seismic and other vibration environments meet the design requirements and safety standards, ensuring that the data center can maintain structural safety and stable operation of equipment during an earthquake, so as to guarantee the continuity of data storage and services.

[0003] The existing technologies have the following deficiencies: In the existing technologies, when training a DNN seismic isolation performance prediction model based on simulation modeling, systematic deviations are often caused by the inconsistency between the actual engineering boundary conditions and the simulation assumptions. For example, in a coastal data center, due to uneven settlement of reclamation in the foundation, but it was misassumed as a homogeneous elastic body during simulation, resulting in the model seriously underestimating the displacement amplification effect at the bottom in the scenario of a large earthquake. Eventually, the seismic isolation device failed prematurely during a strong earthquake, causing equipment tilt and system failures, revealing the significant impact of simulation boundary condition deviations on the accuracy and safety of seismic isolation performance prediction. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI intelligent-based seismic isolation performance detection method for data centers to solve the deficiencies in the background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An AI intelligent-based seismic isolation performance detection method for data centers, including: Deploy multi-dimensional Internet of Things sensors on the data center building structure and equipment, collect real-time equipment vibration characteristic data during seismic events, and extract seismic wave characteristic parameters through a seismic database containing historical seismic wave records; Based on the collected equipment vibration characteristic data and seismic wave characteristic data, construct and train a deep neural network model to learn the non-linear mapping relationship between seismic input and structural response, and form a seismic isolation performance prediction model; For the actually monitored equipment vibration characteristic data and seismic wave characteristic data, respectively extract the abnormal value of the foundation shear wave velocity gradient and the deviation rate of the dynamic displacement response of the seismic isolation layer, and comprehensively calculate the comprehensive deviation index based on a preset weight; According to the comprehensive deviation index, divide the systematic deviation degree of the seismic isolation performance prediction model into high deviation, medium deviation, and low deviation levels, and perform corresponding processing respectively; For the case of medium deviation level, dynamically update the parameters of the foundation physical model and retrain or optimize the seismic isolation performance prediction model.

[0006] Preferably, the multi-dimensional Internet of Things sensor includes an acceleration sensor, a displacement sensor, an inclination sensor, a multi-axis gyroscope, and an environmental sensor.

[0007] Preferably, the method for extracting the abnormal value of the foundation shear wave velocity gradient is as follows: input the discrete data point set of the measured foundation shear wave velocity with depth: ; where is the depth, is the corresponding shear wave velocity; the data set of shear wave velocity with depth obtained by simulation modeling: ; N is the total number of discrete data points of the measured foundation shear wave velocity with depth, is the total number of shear wave velocity data with depth obtained by simulation modeling; is the j-th depth point in the simulation model, which refers to the depth position of a certain layer in the foundation profile set during the simulation modeling process, is the shear wave velocity of the simulation model at the depth indicating the value of the foundation shear wave velocity set at this depth position in the simulation modeling; Perform one-dimensional discrete wavelet transform on the shear wave velocity data, decompose it to multiple scales, only retain the low-frequency approximation components, and reconstruct the retained low-frequency wavelets back into a smooth shear wave velocity trend curve, denoted as , perform depth difference on the trend curve, calculate the gradient , and the expression is: ; is the i-th measured depth point, is the value of the measured shear wave velocity trend curve at the depth , is the value of the measured shear wave velocity trend curve at the depth , perform simple linear interpolation and smoothing on the simulation shear wave velocity, and calculate the simulation gradient , and the expression is: ; represents the value of the simulation shear wave velocity at the depth , is the value of the simulation shear wave velocity at the depth ; Perform depth weighted average on GGI(z) to obtain the overall abnormal value of the foundation shear wave velocity gradient , and the expression is: ; in the formula, is the maximum depth.

[0008] ​​Preferably, the method for extracting the deviation rate of the dynamic displacement response of the isolation layer is as follows: Obtain the time history of the displacement of the isolation layer actually monitored and label it as ; the time history of the displacement of the isolation layer predicted by simulation, label it as ; perform optimal alignment of the actual displacement time history and the simulated displacement time history on the time axis, and calculate the Euclidean distance matrix between each point on the two time history curves. The expression is: ; search for a minimum cumulative cost path connecting the start to the end, representing the optimal alignment path, and output a set of aligned point pairs sequence ; represent the actual displacement corresponding to the simulated displacement ; represent the time point in the simulated displacement time history aligned with , calculate the instantaneous error for each pair of aligned points. The expression is: ; is the value of the simulated predicted displacement at time , calculate the average error of all aligned points: ; where W is the number of point pairs on the alignment path, and take the maximum predicted displacement peak value of the simulated displacement time history: ; finally, the deviation rate IDRΔ of the dynamic displacement response of the isolation layer is defined as: .

[0009] Preferably, set the weight of the abnormal value of the foundation shear wave velocity to 50%; set the weight of the deviation rate of the dynamic displacement of the isolation layer to 50%; perform normalization processing on GGI, and use the absolute value of the deviation from 1 as the metric: ∣GGI−1∣; calculate the comprehensive deviation index CPI, and the calculation formula is: .

[0010] Preferably, according to the comprehensive deviation index, divide the systematic deviation degree of the isolation performance prediction model into high deviation, medium deviation and low deviation levels, and perform corresponding processing respectively, specifically including: If CPI≥0.40, divide the systematic deviation degree of the isolation performance prediction model into a high deviation degree, indicating that the model mismatch is serious and needs to be comprehensively corrected; If 0.20≤CPI<0.40; divide the systematic deviation degree of the isolation performance prediction model into a medium deviation degree, indicating that there is a mismatch in the model and needs to be locally optimized; If CPI<0.20; divide the systematic deviation degree of the isolation performance prediction model into a low deviation degree, indicating that the model is in good agreement with the actual situation and no adjustment is required.

[0011] Preferably, for the case of medium deviation level, Gaussian process regression is used to establish a surrogate relationship between the foundation parameters and the CPI value. The surrogate model can predict the corresponding CPI after each update of the foundation parameters. The expression of the surrogate model is: ; is the total squared error between the CPI predicted by the surrogate model and the actual CPI; r is the number of samples; is the predicted CPI of the i-th sample, is the actual CPI of the i-th sample; Randomly sample several groups of foundation physical parameter combinations and actually evaluate the CPI; Predict the parameter group that will cause the CPI to decrease next step according to the surrogate model; Update the foundation physical parameters and adjust the simulation output; Recalculate the CPI and feedback it to the surrogate model; The termination condition is to find the parameter combination that makes the CPI drop to the low deviation level or reach the maximum number of iterations.

[0012] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. By deploying multi-dimensional Internet of Things sensors on the building structure and equipment of the data center, the present invention can collect vibration characteristic data in seismic events in real time, extract seismic characteristics in combination with the historical seismic wave database, and use a deep neural network to establish a non-linear mapping relationship between seismic input and structural response, realizing the intelligent prediction of seismic isolation performance. By innovatively extracting the abnormal value of the foundation shear wave velocity gradient and the deviation rate of the dynamic displacement response of the seismic isolation layer, comprehensively calculating the comprehensive deviation index, and dividing the systematic deviation level according to the CPI, and implementing hierarchical processing for different deviation degrees, it effectively makes up for the problem of inaccurate prediction of seismic isolation performance caused by the inconsistency between the simulation hypothesis and the actual boundary conditions in the prior art.

[0013] 2. The present invention can use Gaussian process regression to establish a surrogate relationship between foundation parameters and system deviation under the medium deviation level, dynamically update the foundation physical parameters through the Bayesian optimization strategy, and optimize the seismic isolation performance prediction model in combination with incremental learning, quickly correcting the model to the low deviation level, avoiding the high cost and low efficiency problems brought by full-scale reconstruction in traditional methods. Overall, the present invention greatly improves the prediction accuracy and adaptive ability of the seismic isolation system of the data center in complex seismic environments, and enhances the resilience and safety guarantee ability of key infrastructure in seismic disasters. Description of the Drawings

[0014] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other accompanying drawings can also be obtained based on these drawings.

[0015] Figure 1 This is the method mind map of the present invention. Specific embodiments

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0017] Embodiment, please refer to Figure 1 As shown, a method for detecting the seismic isolation performance of a data center based on AI intelligence in this embodiment includes: Deploy multi-dimensional Internet of Things sensors on the building structure and equipment of the data center to collect device vibration characteristic data in seismic events in real time, and extract seismic wave characteristic parameters through a seismic database containing historical seismic wave records; Based on the collected device vibration characteristic data and seismic wave characteristic data, construct and train a deep neural network model to learn the non-linear mapping relationship between seismic input and structural response, and form a seismic isolation performance prediction model; For the actually monitored device vibration characteristic data and seismic wave characteristic data, respectively extract the abnormal value of the ground shear wave velocity gradient and the deviation rate of the dynamic displacement response of the seismic isolation layer, and comprehensively calculate the comprehensive deviation index based on a preset weight; According to the comprehensive deviation index, divide the systematic deviation degree of the seismic isolation performance prediction model into high deviation, medium deviation, and low deviation levels, and perform corresponding processing respectively; For the case of medium deviation level, dynamically update the parameters of the ground physical model and retrain or optimize the seismic isolation performance prediction model.

[0018] A sensor network is arranged on the building structure of the data center and key equipment inside the computer room (such as cabinets, racks, server arrays) for the purpose of real-time monitoring of the dynamic response characteristics of the overall structure and internal equipment during seismic events, including but not limited to key indicators such as acceleration, displacement, angular velocity, and inclination.

[0019] Acceleration sensors (MEMS type, high-sensitivity type): Used to capture absolute acceleration changes under seismic excitation, with a frequency range of 0.1–200 Hz and a measurement range of ±2g to ±5g.

[0020] Displacement sensors (LVDT or laser displacement meter): Monitor displacement changes in the horizontal and vertical directions of the isolation device layer or the cabinet, with an accuracy of up to 0.1 mm.

[0021] Tilt sensors: Capture changes in the tilt angle of the equipment or local structure, with a measurement range of ±30° and a resolution of 0.001°.

[0022] Multi-axis gyroscopes: Capture three-dimensional angular velocity changes for identifying changes in the torsional and vibration modes of the equipment.

[0023] Environmental sensors: Temperature, humidity, etc., used to calibrate sensor drift caused by environmental changes.

[0024] Typical deployment locations include: Foundation, isolation device layer: Focus on monitoring the displacement of the bearings and the energy dissipation state. Frame beam-column joints: Monitor the vibration characteristics of the overall frame. Center and edges of the floor: Evaluate in-plane deformation and floor torsional effects. Bottom, middle, and top of the server cabinet: Capture the vibration amplification effect of the cabinet. Rack connection points: Monitor changes in the connection stiffness of the equipment. Foundations of UPS power supply equipment and large cooling equipment: Pay attention to the inertial response of large-mass equipment.

[0025] It is preferred to use wired (PoE Ethernet) or wireless (LoRa, Wi-Fi 6) methods to ensure communication stability during earthquakes. All sensors are uniformly synchronized at the nanosecond level using a GPS time synchronization system or IEEE 1588 PTP (Precision Time Protocol) to ensure that the data time axes are exactly the same, facilitating subsequent vibration mode analysis.

[0026] The data acquisition content of the equipment vibration characteristic data includes: Acquisition time window: Continuously record from 5 seconds before the earthquake warning to 300 seconds after the earthquake.

[0027] Types of acquired data: Triaxial acceleration, triaxial displacement, tilt angle change, spectral characteristics.

[0028] Data sampling rate: Acceleration data ≥200 Hz (meeting the high-frequency vibration sampling requirements); displacement data ≥50 Hz; tilt angle data ≥10 Hz.

[0029] Real-time data preprocessing includes: Denoising processing: Use wavelet denoising or band-pass filters to filter out high-frequency noise of the sensors.

[0030] Baseline correction: Eliminate low-frequency drift to ensure the accurate physical meaning of the displacement data.

[0031] Anomaly detection: Detect by setting thresholds (such as abnormal jumps in acceleration) or assisted by AI to eliminate error data packets.

[0032] Compression encoding: Use lossless compression (such as FLAC encoding) to compress data in real time and reduce communication pressure.

[0033] Obtain seismic data through an authoritative seismic database: The types of seismic waves cover: surface records; near-fault records; bedrock records.

[0034] All seismic waves are unified and normalized on the scales of displacement, velocity, and acceleration amplitude. Feature extraction content: Peak ground acceleration (PGA); Peak ground velocity (PGV); Peak ground displacement (PGD); Effective duration (such as the 5%-95% cumulative energy time period); Fourier spectrum characteristics (main frequency, characteristic period); Arias intensity (an index of seismic input energy). Each seismic record is converted into a set of standardized feature vectors, which are used as the input feature set of the DNN model.

[0035] Based on the collected device vibration characteristic data and seismic wave characteristic data, construct and train a deep neural network model to learn the non-linear mapping relationship between seismic input and structural response, and form a seismic isolation performance prediction model, specifically: Eliminate data samples that are incomplete, have obvious outliers, or drift and fail. Unify the unit system (such as acceleration m / s², displacement m). Extract the device vibration characteristic data (such as maximum acceleration, maximum displacement, main vibration frequency, vibration duration) into a feature set. Unify and normalize the seismic wave characteristic parameters (such as PGA, PGV, PGD, effective duration, characteristic period, etc.) to form a standard input vector. Optional feature enhancement: Add frequency domain features (such as the main frequency of the Fourier spectrum); Add time domain statistics (such as root mean square acceleration RMS). Normalize all input and output data (such as standardize to mean 0 and standard deviation 1) to improve the network convergence speed and stability. Divide the data into training set, validation set, and test set according to the ratio of 8:1:1, ensuring that the seismic waves in the test set do not overlap with the training set to examine the generalization ability of the model.

[0036] The input vector includes: a seismic wave characteristic parameter vector (such as 10-20 dimensions); a device vibration characteristic vector (such as 10 dimensions); the total input feature dimension is usually controlled within 20-50 dimensions.

[0037] Construct a multi-layer perceptron (MLP) architecture: Number of hidden layers: 3-5 layers; Number of neurons in each layer: 64-256; Activation function: ReLU (to avoid the problem of gradient disappearance); A Batch Normalization layer can be added to stabilize the training process.

[0038] The output nodes correspond to the seismic isolation performance prediction targets, such as: the maximum horizontal displacement of the seismic isolation device; the maximum acceleration response of the equipment; the energy dissipation rate of the seismic isolation system; the activation function is selected according to the output target, for example, a linear activation is used for regression problems.

[0039] The mean squared error loss function is used for regression tasks; Optimizer selection: Adam Optimizer (the initial learning rate is set to 0.001); Batch size: between 32 and 128, adjusted according to the data volume and GPU memory.

[0040] Number of training epochs: dynamically adjusted according to the convergence of the validation set error, usually 100 - 500 epochs.

[0041] Introduce the Dropout mechanism (Dropout rate between 0.2 and 0.5) Use the early stopping strategy to terminate training when the validation set error does not decrease for several consecutive epochs.

[0042] Adopt R² (coefficient of determination), MAE (mean absolute error), and RMSE (root mean square error) as the main evaluation indicators. At the same time, draw a scatter plot of the prediction results and the true observed values to test the fitting effect.

[0043] Use extreme seismic wave data that has not appeared in the test set (such as near - fault pulse - type earthquakes) to test the model stability. Calculate the performance differences of the model under earthquakes of different intensities and different types.

[0044] Freeze and package the trained and validated DNN model (such as save it in the ONNX or TensorFlow SavedModel format).

[0045] Deploy the model to the intelligent monitoring system in the data center to realize the real - time input of the characteristics of the seismic waves collected during an earthquake and the preliminary response data of the equipment; quickly predict the response of the seismic isolation system and the potential risk of performance degradation.

[0046] Regularly collect new actual earthquake response data, expand the dataset, perform incremental learning, and continuously improve the prediction accuracy of the model.

[0047] For the actually monitored equipment vibration characteristic data and seismic wave characteristic data, extract the outliers of the foundation shear wave velocity gradient and the deviation rate of the dynamic displacement response of the seismic isolation layer respectively, and comprehensively calculate the comprehensive deviation index based on the preset weights, specifically including: The method for extracting the outliers of the foundation shear wave velocity gradient is: input the discrete data point set of the measured foundation shear wave velocity with respect to depth: ; where is the depth, is the corresponding shear wave velocity. The dataset of shear wave velocity varying with depth obtained from simulation modeling: ; N is the total number of discrete data points of the measured foundation shear wave velocity varying with depth, is the total number of shear wave velocity data varying with depth obtained from simulation modeling; is the j-th depth point in the simulation model, referring to the depth position of a certain layer in the foundation profile set during the simulation modeling process, is the shear wave velocity of the simulation model at depth indicating the value of the foundation shear wave velocity set at this depth position in the simulation modeling.

[0048] Select a wavelet basis suitable for smooth trend extraction, such as Daubechies wavelet (db4) or Symlets wavelet (sym5). Perform one-dimensional discrete wavelet transform on the shear wave velocity data and decompose it to multiple scales (for example, the 3rd layer or the 4th layer). Only retain the low-frequency approximation components and discard the high-frequency detail components. Reconstruct the retained low-frequency wavelets back into a smooth shear wave velocity trend curve, denoted as . Differentiate the trend curve with respect to depth to calculate the gradient (shear wave velocity change rate) , and the expression is: ; is the i-th measured depth point, is the value of the measured shear wave velocity trend curve at depth , is the value of the measured shear wave velocity trend curve at depth . For the simulated shear wave velocity, perform simple linear interpolation and smoothing (it can be not processed by wavelet if the simulation data is ideally layered), and calculate the simulated gradient , and the expression is: ; represents the value of the simulated shear wave velocity at depth , is the value of the simulated shear wave velocity at depth ; In each depth section, calculate the ratio of the actual gradient to the simulated gradient : . If GGI(z)≈1, it means the actual and the simulated are consistent. If GGI(z)>1, it means the actual foundation gradient is steeper (hardening trend). If GGI(z)<1, it means the actual foundation is softened or stratified and weakened.

[0049] Perform depth-weighted averaging on GGI(z) to obtain the overall foundation shear wave velocity gradient anomaly value , and the expression is: ; In the formula, is the maximum depth.

[0050] The method for extracting the deviation rate of the dynamic displacement response of the seismic isolation layer is as follows: Obtain the time history of the displacement of the seismic isolation layer measured in actuality, marked as ; the time history of the displacement of the seismic isolation layer predicted by simulation, marked as ; the two time histories can have different lengths (different earthquake durations or slightly different sampling frequencies), and the sampling frequencies should be unified or simple linear interpolation processing should be performed.

[0051] Align the actual displacement time history and the simulated displacement time history optimally on the time axis. Allow local time stretching / compression to make the overall response trends as close as possible. Calculate the Euclidean distance matrix between each point on the two time history curves, and the expression is: ; search for a minimum cumulative cost path connecting the start to the end, representing the optimal alignment path, and output a set of aligned point pairs sequence ; represents the actual displacement corresponding to the simulated displacement ; represents the time point in the simulated displacement time history aligned with , and calculate the instantaneous error for each pair of aligned points, and the expression is: ; is the value of the simulated predicted displacement at time , and calculate the average error of all aligned points: ; where W is the number of point pairs on the alignment path, and take the maximum predicted displacement peak value of the simulated displacement time history: ; finally, the deviation rate IDRΔ of the dynamic displacement response of the seismic isolation layer is defined as: .

[0052] Set the weight of the ground shear wave velocity anomaly value (GGI) to 50%; set the weight of the deviation rate of the dynamic displacement of the seismic isolation layer (IDRΔ) to 50%; the weights can be flexibly adjusted according to project requirements (for example, for high-rise buildings, more emphasis is placed on displacement deviation, and for soft soil foundations, more emphasis is placed on gradient anomaly).

[0053] Normalize GGI, and use the absolute value of the deviation from 1 as the metric: ∣GGI−1∣; calculate the comprehensive deviation index CPI, and the calculation formula is: .

[0054] According to the comprehensive deviation index, divide the systematic deviation degree of the seismic isolation performance prediction model into high deviation, medium deviation, and low deviation levels, and perform corresponding processing respectively, specifically including: If CPI ≥ 0.40, the systematic deviation degree of the seismic isolation performance prediction model is classified as a high deviation degree, indicating serious model mismatch and the need for comprehensive correction; If 0.20 ≤ CPI < 0.40, the systematic deviation degree of the seismic isolation performance prediction model is classified as a medium deviation degree, indicating a certain degree of mismatch and the need for local optimization; If CPI < 0.20, the systematic deviation degree of the seismic isolation performance prediction model is classified as a low deviation degree, indicating that the model is in good agreement with the actual situation and no adjustment is required.

[0055] When the comprehensive deviation index CPI is in the medium deviation level (0.20 ≤ CPI < 0.40), although the overall trend of the seismic isolation performance prediction model is acceptable, some foundation physical parameters (such as shear wave velocity distribution, damping ratio, stiffness of the seismic isolation device) deviate from the actual situation and need to be dynamically updated in a targeted manner to avoid the accumulation of systematic errors. The goal is to dynamically correct the foundation model parameters and locally optimize the performance of the DNN prediction model at the lowest possible cost, rather than tearing down and rebuilding.

[0056] Set the range of foundation parameters to be adjusted, for example: The adjustment coefficient of the foundation shear wave velocity αVs ∈ [0.8, 1.2]; The adjustment coefficient of the foundation damping ratio αξ ∈ [0.9, 1.1]; The adjustment coefficient of the initial stiffness of the seismic isolation device αk ∈ [0.85, 1.15]; Optimization goal: Minimize the comprehensive deviation index CPI.

[0057] Use Gaussian process regression to establish a surrogate relationship between the foundation parameters and the CPI value. The surrogate model can predict the corresponding CPI after each update of the foundation parameters. The expression of the surrogate model is: ; is the total squared error between the CPI predicted by the surrogate model and the actual CPI, used to measure the fitting accuracy of the surrogate model; r is the number of samples; is the predicted CPI of the i-th sample, is the actual CPI of the i-th sample.

[0058] Randomly sample several groups of combinations of foundation physical parameters and actually evaluate the CPI.

[0059] According to the surrogate model, predict the parameter groups that may lead to a decrease in CPI in the next step (using the upper confidence bound algorithm UCB or the expected improvement EI criterion).

[0060] Update the foundation physical parameters and slightly adjust the simulation output.

[0061] Recalculate the CPI and feedback it to the surrogate model.

[0062] The termination condition is to find a parameter combination that can reduce the CPI to a low deviation level (CPI < 0.20), or reach the maximum number of iterations (e.g., 50 times).

[0063] Instead of retraining the entire deep neural network DNN, only perform small-step incremental training on the last few layers (such as the output layer or the second-to-last layer).

[0064] Use the new corrected simulation data (in small batches) for incremental training.

[0065] Retain the existing knowledge of the model, and at the same time improve the prediction accuracy for new foundation conditions.

[0066] Use independent verified seismic wave samples to evaluate the prediction accuracy of the isolation performance of the fine-tuned model under new foundation parameter conditions. After confirming that the model meets the new low deviation level standard, solidify the updated version.

[0067] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0068] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access, or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0069] It should be understood that the term "and / or" in this text is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. Additionally, the character " / " in this text generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context. Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this text can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0070] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application.

Claims

1. A data center seismic isolation performance detection method based on AI intelligence, characterized by: include: Deploy multi-dimensional IoT sensors on the data center building structure and equipment to collect equipment vibration characteristic data in real time during earthquake events, and extract seismic wave characteristic parameters through a seismic database containing historical seismic wave records; Based on the collected equipment vibration characteristic data and seismic wave characteristic data, a deep neural network model is constructed and trained to learn the nonlinear mapping relationship between seismic input and structural response, forming a seismic isolation performance prediction model; For the equipment vibration characteristic data and seismic wave characteristic data actually monitored, the foundation shear wave velocity gradient anomaly and the isolation layer dynamic displacement response deviation rate are extracted respectively, and the comprehensive deviation index is calculated based on the preset weights; According to the comprehensive deviation index, the systematic deviation degree of the seismic isolation performance prediction model is divided into high deviation, medium deviation and low deviation levels, and corresponding treatments are performed respectively; For situations with medium deviation levels, the foundation physical model parameters are dynamically updated, and the seismic isolation performance prediction model is retrained or optimized.

2. According to claim 1, a data center seismic isolation performance detection method based on AI intelligence is characterized by: The multi-dimensional Internet of Things sensor includes an acceleration sensor, a displacement sensor, a tilt sensor, a multi-axis gyroscope and an environmental sensor.

3. According to claim 1, a method for detecting seismic isolation performance of a data center based on AI intelligence is characterized in that: The method for extracting the outlier value of the foundation shear wave velocity gradient is as follows: input the discrete data point set of the measured foundation shear wave velocity with depth: ;in, is depth, is the corresponding shear wave velocity; the shear wave velocity versus depth data set obtained by simulation modeling: ; N is the total number of discrete data points of the measured foundation shear wave velocity with depth, The total number of shear wave velocity versus depth data obtained for simulation modeling; The jth depth point in the simulation model refers to the depth position of a layer in the foundation profile set during the simulation modeling process. For simulation models at depth The shear wave velocity at represents the foundation shear wave velocity value set at this depth in the simulation modeling; The shear wave velocity data is transformed into a one-dimensional discrete wavelet transform and decomposed into multiple scales, retaining only the low-frequency approximate components. The retained low-frequency wavelet is reconstructed into a smooth shear wave velocity trend curve, which is recorded as , differentiate the trend curve by depth and calculate the gradient , the expression is: ; is the i-th measured depth point, The measured shear wave velocity trend curve at depth The value at The measured shear wave velocity trend curve at depth The simulated shear wave velocity is simply linearly interpolated and smoothed to calculate the simulated gradient , the expression is: ; Indicates the simulated shear wave velocity at depth The value at To simulate the shear wave velocity at depth The value at Perform depth-weighted averaging on GGI(z) to obtain the overall foundation shear wave velocity gradient anomaly , the expression is: ; In the formula, is the maximum depth.

4. According to claim 3, a method for detecting seismic isolation performance of a data center based on AI intelligence is characterized in that: The method for extracting the deviation rate of the dynamic displacement response of the seismic isolation layer is as follows: obtain the displacement time history of the seismic isolation layer actually monitored, marked as ; The displacement time history of the isolation layer predicted by simulation is marked as ; The actual displacement time Simulation displacement time history Perform optimal alignment on the time axis and calculate the Euclidean distance matrix for each point between the two time course curves , the expression is: ; Search for a path with the minimum cumulative cost, connecting the start to the end, representing the optimal alignment path, and output a set of aligned point pair sequences ; Indicates the actual displacement Corresponding simulation displacement ; Indicates the displacement time history of the simulation Aligned time points, calculate the instantaneous error for each pair of aligned points , the expression is: ; For simulations predicting displacements in time Calculate the average error of all alignment points : ; Where W is the number of point pairs on the alignment path, and the maximum predicted displacement peak value of the simulated displacement history is taken : ; The final isolation layer dynamic displacement response deviation rate IDRΔ is defined as: .

5. According to claim 4, a method for detecting seismic isolation performance of a data center based on AI intelligence is characterized in that: The weight of the foundation shear wave velocity anomaly is set to 50%; the weight of the dynamic displacement deviation rate of the isolation layer is set to 50%; the GGI is normalized and the absolute value of the deviation from 1 is used as the metric: |GGI−1|; Calculate the comprehensive deviation index CPI, the calculation formula is: .

6. According to claim 5, a method for detecting seismic isolation performance of a data center based on AI intelligence is characterized in that: According to the comprehensive deviation index, the systematic deviation degree of the seismic isolation performance prediction model is divided into high deviation, medium deviation and low deviation levels, and corresponding treatments are performed respectively, including: If CPI ≥ 0.40, the systematic deviation degree of the seismic isolation performance prediction model is classified as high deviation degree, indicating that the model is seriously mismatched and needs to be fully corrected; If 0.20≤CPI < 0.40, the systematic deviation degree of the seismic isolation performance prediction model is classified as medium deviation degree, indicating that the model has mismatch and needs local optimization; If CPI < 0.20, the systematic deviation degree of the seismic isolation performance prediction model is classified as low deviation, indicating that the model is consistent with the actual situation and no adjustment is required.

7. The method for detecting seismic isolation performance of a data center based on AI intelligence according to claim 6 is characterized in that: For the case of medium deviation level, Gaussian process regression is used to establish the proxy relationship between foundation parameters and CPI values. The proxy model can predict the corresponding CPI after each foundation parameter update. The proxy model expression is: ; is the total square error between the CPI predicted by the proxy model and the actual CPI; r is the number of samples; is the predicted CPI of the ith sample, is the actual CPI of the i-th sample; Randomly sample several combinations of foundation physical parameters to actually evaluate CPI; Predict the parameter group that will cause CPI to fall in the next step based on the proxy model; Update foundation physical parameters and adjust simulation output; Recalculate CPI and feed it back to the proxy model; The termination condition is to find a parameter combination that reduces the CPI to a low deviation level or reach the maximum number of iterations.

Images