A seismic response prediction control system for a multi-story building structure

Abstract

The application relates to a kind of earthquake response prediction control systems for multi-storey building structure, comprising: physical layer, including sensor group and actuator group, sensor group is collected and uploaded by being laid on multi-storey building structure sensing data, and control instruction is received to control actuator group;Data layer, after receiving sensing data, carry out time alignment preprocessing and upload to digital layer;Digital layer, set for carrying out earthquake response prediction edge computing node, and for carrying out earthquake response prediction model calibration in edge computing node, whole life cycle data management and digital twin model generation cloud computing platform;Service layer, structure damage assessment, safety warning and management control instruction generation, issue to control layer;Control layer, according to the control instruction issued, to actuator group carries out closed-loop control.Compared with prior art, the application has the advantages of high real-time, high prediction accuracy, high reliability and the like.

Description

Technical Field

[0001] This invention relates to the field of earthquake response prediction, and in particular to an earthquake response prediction and control system for multi-story building structures. Background Technology

[0002] Earthquakes, as a sudden and destructive natural disaster, generate seismic waves that propagate through the soil to building structures, triggering structural vibration responses. Multi-story buildings, due to their structural characteristics, are prone to problems such as excessive inter-story drift angles and amplified floor accelerations under earthquake action. In severe cases, this can lead to damage to structural components, overall instability, or even collapse, threatening the safety of people's lives and property. Therefore, accurate prediction and effective control of the seismic response of multi-story building structures have become one of the core requirements of earthquake resistance and disaster prevention in the field of structural engineering.

[0003] Cyber-physical systems (CPS) integrate computing, communication, and control with physical processes to enable real-time monitoring and automatic control of engineering systems. Digital twin technology creates dynamic virtual representations of physical systems, utilizing bidirectional data streams to mirror, understand, and optimize physical counterparts.

[0004] In the field of structural engineering, these two technologies have great potential to improve the reliability and service life of infrastructure, but existing technologies face the following engineering application challenges: (1) Structural safety hazards caused by insufficient real-time performance: In traditional structural health monitoring systems, sensor data must undergo multi-level network transmission (sensor → data acquisition unit → local server → cloud platform) before reaching the digital twin model, resulting in a cumulative delay of 2-5 seconds. For extreme events such as earthquakes, the structural response can reach its peak within seconds, making digital twin models with delays exceeding 500 milliseconds unsuitable for real-time safety assessments.

[0005] For example, under earthquake action, the inter-story drift angle of the structure i It reaches its peak within 5-10 seconds, such as i max =0.025rad, which has reached a state of severe damage; a 5-second delay in traditional monitoring systems means that when the digital twin model shows that the structure has entered a dangerous state, the physical structure has already experienced a complete peak response, losing its early warning value; in the wind vibration response, the displacement of the structure's apex vibrates with a main period of 3-15 seconds, and a 2-second delay causes the phase difference between the digital twin model and the actual response to reach 48-240 degrees, making it unusable for active control.

[0006] (2) Model distortion caused by the lack of a data synchronization mechanism: The existing system lacks a time alignment mechanism for heterogeneous sensors (such as accelerometers at 200Hz, GPS at 10Hz, and strain gauges at 100Hz), resulting in the digital twin model receiving data that is out of sync with time, and the calculated response quantities being contradictory.

[0007] (3) Lack of closed-loop control capability leads to insufficient intelligence: The existing digital twin framework only realizes a one-way data flow of monitoring → digital model → evaluation, and lacks a feedback control channel of "digital model → physical actuator", so it cannot automatically adjust the physical structure according to the prediction results.

[0008] (4) Cloud bottlenecks caused by the lack of edge computing: Traditional systems upload all sensor data to a cloud platform for processing, with a single complete analysis (such as damage identification) taking 5-30 seconds. For large structures containing hundreds of sensors, network bandwidth becomes a bottleneck, and cloud platform failures can paralyze the entire system. For example, a large building structure may deploy 500+ sensors, generating data volume of 500 channels × 200Hz × 4 bytes per second = 400KB / s = 34.5GB / day. Continuous uploading to the cloud platform puts enormous pressure on the network and storage; modal recognition algorithms (identifying the natural frequencies of the structure) also pose a challenge. f i , vibration shape f i Damping ratio g i Processing in the cloud takes 20-60 seconds, while actual engineering requires completion within 5 seconds to support real-time decision-making.

[0009] Therefore, there is an urgent need to design a seismic response prediction and control system with high real-time performance, high prediction accuracy, and high reliability to meet the engineering requirements of real-time performance, accuracy, and intelligence for next-generation infrastructure such as smart buildings and smart bridges. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a seismic response prediction and control system for multi-story building structures that has high real-time performance, high prediction accuracy, and high reliability.

[0011] The objective of this invention can be achieved through the following technical solutions: A seismic response prediction and control system for multi-story building structures, comprising: The physical layer includes sensor groups and actuator groups. Sensor groups deployed on multi-story building structures collect sensing data and upload it to the data layer. The physical layer receives control commands from the control layer to control the actuator groups, which respond to changes in multi-story building structures during earthquakes. The data layer receives sensor data from the physical layer, performs time-aligned preprocessing on the sensor data, and then uploads the time-aligned preprocessed sensor data to the digital layer. The digital layer includes edge computing nodes for seismic response prediction, as well as a cloud computing platform for calibrating seismic response prediction models in the edge computing nodes, managing full lifecycle data, and generating digital twin models. The service layer performs structural damage assessment and safety warning based on the seismic response prediction data uploaded by the digital layer, generates management and control instructions, sends the control instructions to the control layer, and simultaneously sends the structural damage assessment results and safety warning feedback information to the cloud computing platform of the digital layer. The control layer performs closed-loop control of the actuator group on the multi-story building structure based on the control commands issued by the service layer.

[0012] Preferably, the sensor group includes an accelerometer, a GPS displacement meter, a strain gauge, and an anemometer. The accelerometer and the GPS displacement meter are respectively installed on each floor of the multi-story building structure, a set number of strain gauges are installed on the key sections of the beams and columns of the multi-story building structure, and the anemometer is installed on both sides of the roof of the multi-story building structure.

[0013] Preferably, in the data layer, after receiving sensor data from the physical layer, time-aligned preprocessing is performed on the sensor data, specifically as follows: Receive sensor data from the physical layer, where all sensors are timed via GPS or BeiDou, and the data packet format is {sensor ID, UTC timestamp, sampling rate, data value}. For sensor data with different sampling rates, a time alignment algorithm based on Kalman filtering is adopted. This algorithm fuses multi-source heterogeneous sensor data through a state-space model to perform time synchronization and alignment, resulting in synchronized data with a unified time reference. a sync (t), v sync (t), d sync (t), e sync (t)}, where a sync (t), v sync (t), d sync (t), e sync (t) represents the acceleration, velocity, displacement, and strain data at step t after time alignment.

[0014] Preferably, the digital layer includes multiple edge computing nodes, each of which executes a corresponding edge computing task, including: The first edge computing node is used for data preprocessing: filtering, baseline correction, and feature extraction of time-aligned preprocessed sensor data. The second edge computing node is used for real-time seismic response prediction: based on the most recently set-duration acceleration window data, it uses the GateBlend-LSTM model to predict the seismic response, obtaining seismic response data for the future set-duration period. This seismic response data includes acceleration response data, velocity response data, and displacement response data; and... The third edge computing node calculates the inter-story drift angle based on the displacement response data obtained from the second edge computing node, which is used by the service layer to perform structural damage assessment based on the inter-story drift angle threshold.

[0015] Preferably, in the second edge computing node, the GateBlend-LSTM model is used for seismic response prediction, specifically including: Ground acceleration window data is input, and structural displacement prediction values ​​are obtained by fusing data-driven point-by-point displacement prediction values ​​and physics-driven integral displacement prediction values ​​through the GateBlend gated fusion mechanism. In the GateBlend-LSTM model, data-driven point-by-point velocity prediction values ​​and data-driven point-by-point displacement prediction values ​​are predicted based on the LSTM backbone network and parallel velocity prediction heads and displacement prediction heads. Numerical integration is performed on the data-driven point-by-point velocity prediction values ​​to obtain the physics-driven integral displacement prediction values. The inter-story drift angle used for structural damage assessment is calculated based on the predicted structural displacement values. The inter-story drift angle is combined with the predicted values ​​of structural acceleration, structural velocity, and structural displacement to obtain the seismic structural response prediction results; wherein the predicted values ​​of structural acceleration are obtained by the LSTM backbone network and acceleration prediction head in the GateBlend-LSTM model, and the predicted values ​​of structural velocity are obtained by the LSTM backbone network and velocity prediction head in the GateBlend-LSTM model.

[0016] Preferably, the cloud computing platform performs the following tasks, including: First cloud computing task: Calibrate the GateBlend-LSTM model according to the set frequency; The second cloud computing task is long-term degradation trend analysis: extracting the natural frequency decay rate as a degradation indicator from historical data, and predicting the remaining service life. The third cloud computing task is full lifecycle data management; and The fourth cloud computing task is to update the digital twin model.

[0017] Preferably, the system employs edge-cloud collaborative monitoring, including: In normal monitoring mode, the following steps are performed: the first edge computing node uploads sensor statistics data to the cloud computing platform once per minute, and the cloud computing platform performs earthquake response prediction model calibration once per hour; When the peak acceleration exceeds the set value or any sensor value exceeds the set threshold, the earthquake response mode is triggered, and the following actions are executed: The edge computing nodes are switched to high-frequency mode, and the time-aligned preprocessed sensor data and seismic response prediction data are uploaded to the cloud computing platform in seconds. The cloud computing platform then initiates an emergency analysis process to perform structural damage assessment, safety warning, and management control at the service layer.

[0018] Preferably, the structural damage assessment, safety early warning, and management control at the service layer include: Determining the structural damage level based on the inter-story drift angle threshold; Based on the determined level of structural damage, corresponding safety warnings and management controls are implemented.

[0019] Preferably, the service layer further includes predictive maintenance based on long-term degradation trend analysis obtained from the cloud computing platform, including: When the rate of decrease of the structure's natural frequency exceeds a set value, a structural reinforcement recommendation is output. When the increase in damping ratio exceeds the set value, it is determined that there is a potential crack, and detailed inspection recommendations are output. When the peak strain exceeds the material's yield strain, the component is considered to have entered the plastic stage, and a replacement recommendation is issued.

[0020] Preferably, the control layer selects a control strategy to perform closed-loop control on the actuator group on the multi-story building structure according to the control instructions issued by the service layer. The actuator group includes a magnetorheological damper, a variable stiffness seismic isolation bearing, and an active mass tuner.

[0021] Compared with the prior art, the present invention has the following advantages: (1) A five-layer architecture of “physical layer-data layer-digital layer-service layer-control layer” is adopted. Edge computing nodes and cloud computing platform are set up in the digital layer. By using edge computing, the key task of earthquake response prediction is pushed down, which further reduces the system latency. The total system latency is reduced to 63 milliseconds. Compared with traditional cloud processing, the latency reduction is 98.7%, which meets the real-time control requirements. Moreover, the edge computing nodes process locally, so even if the cloud platform fails, the basic monitoring and early warning functions can still be maintained, ensuring the availability of the system.

[0022] (2) Using the Kalman filter algorithm to perform time alignment and fusion processing on heterogeneous sensor data can effectively solve the problem of data synchronization between sensors with different sampling rates and different time-series characteristics, significantly improve the time consistency and fusion accuracy of multi-source sensor data, not only suppress sensor noise and smooth data fluctuations, but also complete preprocessing and feature extraction at the edge, providing a high-quality and high-reliability data foundation for subsequent seismic response pattern recognition and structural damage assessment, thereby enhancing the overall stability, anti-interference ability and decision accuracy of the monitoring system.

[0023] (3) Edge-cloud collaborative monitoring is adopted. In normal monitoring mode, edge computing nodes upload statistical data at a low frequency, which ensures data continuity and effectively reduces network bandwidth usage. When the acceleration peak exceeds the limit or the sensor is abnormal, the edge node can quickly switch to high-frequency upload mode and push the preprocessed sensor data and seismic response prediction data to the cloud computing platform in a timely manner, realizing millisecond-level anomaly perception and rapid response. The cloud computing platform simultaneously starts the emergency analysis process at the service layer to complete structural damage assessment, safety early warning and management control. It not only gives full play to the real-time decision-making capability of the edge side, but also uses the powerful computing power of the cloud to realize global and refined risk assessment, which improves the anti-interference capability, emergency handling efficiency and long-term operational stability of the monitoring system.

[0024] (4) Establish a closed-loop control system for the entire process of “monitoring-prediction-decision-execution”, deeply integrate digital twin simulation prediction with real-time response of physical structure, realize active / semi-active structural control based on prediction results. The system can automatically adjust the parameters of physical actuators and form a closed-loop feedback according to the structural state prediction information output by digital twin, transforming the traditional passive response into active prediction intervention, significantly improving the timeliness, accuracy and reliability of structural control, and effectively reducing the risk of structural damage under disaster loads.

[0025] (5) The system adopts a five-layer architecture of "physical layer-data layer-digital layer-service layer-control layer". The layered architecture supports flexible expansion. In a large project, the system expanded from 50 sensors to 500 sensors + 20 actuators. Only edge computing nodes need to be added, without reconstructing the overall architecture, thus enhancing scalability. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram illustrating the response time of each stage of system data processing in the embodiment; Figure 3 This is a schematic diagram comparing the average end-to-end delay of the present invention with that of the prior art; Figure 4 This is a diagram comparing the deployment time of the present invention with that of existing technologies. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] Example like Figure 1 As shown, this embodiment provides a seismic response prediction and control system for multi-story building structures. It utilizes a digital twin-driven cyber-physical system framework and, through a five-layer architecture design, edge-cloud collaborative computing, and real-time closed-loop control, solves the latency, synchronization, control, and scalability problems of existing technologies in structural engineering applications. The system includes: The physical layer includes sensor groups and actuator groups. Sensor groups deployed on multi-story building structures collect sensing data and upload it to the data layer. The physical layer receives control commands from the control layer to control the actuator groups, which respond to changes in multi-story building structures during earthquakes. The data layer receives sensor data from the physical layer, performs time-aligned preprocessing on the sensor data, and then uploads the time-aligned preprocessed sensor data to the digital layer. The digital layer includes edge computing nodes for seismic response prediction, as well as a cloud computing platform for calibrating seismic response prediction models in the edge computing nodes, managing full lifecycle data, and generating digital twin models. The service layer performs structural damage assessment and safety warning based on the seismic response prediction data uploaded by the digital layer, generates management and control instructions, sends the control instructions to the control layer, and simultaneously sends the structural damage assessment results and safety warning feedback information to the cloud computing platform of the digital layer. The control layer performs closed-loop control of the actuator group on the multi-story building structure based on the control commands issued by the service layer.

[0029] Next, we will introduce each layer in detail.

[0030] 1. Physical layer.

[0031] The physical layer includes sensor groups and actuator groups. Sensor groups deployed on multi-story building structures collect sensing data and upload it to the data layer. The physical layer receives control commands from the control layer to control the actuator groups, which respond to changes in the multi-story building structure under earthquakes.

[0032] In this embodiment, the sensor group includes an accelerometer, a GPS displacement meter, a strain gauge, and an anemometer. The accelerometer and the GPS displacement meter are installed on each floor of the multi-story building structure, a set number of strain gauges are installed on the key sections of the beams and columns of the multi-story building structure, and the anemometers are installed on both sides of the roof of the multi-story building structure.

[0033] Specifically, for a multi-story building structure with 20 floors and a height of 70 meters, the sensor deployment scheme is as follows: Accelerometers: Three-dimensional accelerometers are arranged on each layer, with a total of 20 nodes × 3 directions = 60 channels, sampling rate of 200Hz, range of ±2g, and resolution of 0.001g; GPS displacement gauges: 20 points × 3 directions = 60 channels per floor, sampling rate 10Hz, accuracy ±5mm; Strain gauges: 100 points on key sections of beams and columns, sampling rate 100Hz, range ±3000με; Anemometer: Both sides of the rooftop, 2 points × 2 directions = 4 channels, sampling rate 1Hz; Total data volume: (60×200+60×10+100×100+4×1)×4 bytes = 88KB / s.

[0034] All sensors are timed via GPS / BeiDou with a time synchronization accuracy of ±1ms; data packet format: {sensor ID, UTC timestamp, sampling rate, data value}.

[0035] 2. Data layer.

[0036] After receiving the sensor data from the physical layer, a Kalman filter is used to perform time-aligned preprocessing on the sensor data, specifically: Receive sensor data from the physical layer, where all sensors are timed via GPS or BeiDou, and the data packet format is {sensor ID, UTC timestamp, sampling rate, data value}. For sensor data with different sampling rates, a time alignment algorithm based on Kalman filtering is adopted. This algorithm fuses multi-source heterogeneous sensor data through a state-space model to perform time synchronization and alignment, resulting in synchronized data with a unified time reference. a sync (t), v sync (t), d sync (t), e sync (t)}, where a sync (t), v sync (t), d sync (t), esync (t) represents the acceleration, velocity, displacement, and strain data at step t after time alignment.

[0037] The time interpolation process using Kalman filtering includes the following steps: Equations of state: x k =A· x k-1 +B· u k + w k ,in x k =[ d k ; v k ; a k Let A be the state vector (displacement, velocity, acceleration), B be the state transition matrix, and C be the input matrix. w k This is process noise.

[0038] Observation equation: z k =H· x k + v k ,in z k H represents the actual measured values ​​(e.g., GPS displacement 10Hz, acceleration integral displacement 200Hz); H is the observation matrix. v k For measuring noise.

[0039] The prediction and correction process is as follows: 1) Prediction x̂ k- =A· x̂ k-1 +B· u k 2) Update K k =P k- ·H T / (H·P k ·H T +R), where P k R is the state covariance, and R is the measurement noise covariance. 3) Correcting x̂ k =x̂ k- +K k ·(z k -H·x̂ k- ) Output synchronization data with a unified time base (200Hz) { a sync (t), v sync (t), d sync (t), e sync (t)}.

[0040] For sensor failures or data loss, Gaussian process regression (GPR) is used for interpolation, with the expression: f(t)~GP(μ(t),k(t,t')), where μ(t) is the mean function and k(t,t') is the kernel function; Imputation accuracy: For <5% missing data, the interpolation error is <3%.

[0041] The final output is complete, synchronized, and aligned sensor data.

[0042] In this embodiment, an application programming interface (API) is used to ensure seamless, real-time data transmission to the cloud platform for secure storage, processing, and analysis. This layer includes an IoT gateway, connectivity network, and middleware, supporting short-range technologies such as Wi-Fi and ZigBee, as well as long-range technologies such as LTE.

[0043] 3. Digital layer.

[0044] The digital layer includes a cloud computing platform and multiple edge computing nodes.

[0045] Edge computing nodes are deployed as follows: Hardware configuration: Processor is NVIDIA Jetson Xavier NX (6-core CPU + 384-core GPU), memory is 8GB LPDDR4, storage is 128GB SSD, communication is Gigabit Ethernet + 4G / 5G module; Deployment location: Structural field control box, 1-3 nodes per structure.

[0046] Multiple edge computing nodes execute corresponding edge computing tasks, including: (1) The first edge computing node is used for data preprocessing.

[0047] The time-aligned preprocessed sensor data is filtered (Butterworth filter: cutoff frequency 25Hz, to remove high-frequency noise), baseline corrected (to remove zero drift in acceleration, with a zero drift rate of <0.001g / min), and feature extracted (to calculate peak acceleration PGA and peak displacement PGD).

[0048] (2) Second edge computing node, used for real-time seismic response prediction.

[0049] Based on the most recently collected acceleration window data, the GateBlend-LSTM model is used to predict the seismic response, obtaining seismic response data for the next specified time period. This seismic response data includes acceleration response data, velocity response data, and displacement response data. For example, inputting acceleration data for the previous 10 seconds will output predicted response data for the next 50 seconds.

[0050] Specifically, the GateBlend-LSTM model is used for seismic response prediction, including: Input ground acceleration window data, and fuse data-driven point-by-point displacement predictions and physics-driven integral displacement predictions through the GateBlend gating fusion mechanism to obtain structural displacement predictions.

[0051] In the GateBlend-LSTM model, based on the LSTM backbone network and parallel velocity and displacement prediction heads, data-driven point-by-point velocity and displacement prediction values ​​are obtained. Numerical integration is performed on the data-driven point-by-point velocity prediction values ​​to obtain the physics-driven integral displacement prediction values. In addition, the structural acceleration prediction value is obtained from the LSTM backbone network and acceleration prediction head in the GateBlend-LSTM model, and the structural velocity prediction value is obtained from the LSTM backbone network and velocity prediction head in the GateBlend-LSTM model.

[0052] (3) The third edge calculation node calculates the inter-story drift angle based on the displacement response data obtained from the second edge calculation node. i i (t)=( d i (t)- d i-1 (t)) / h i (i corresponds to the story height, d is the displacement value), used by the service layer to perform structural damage assessment based on the inter-story drift angle threshold.

[0053] The inter-story drift angle is calculated based on the predicted structural displacement value to determine structural damage; the inter-story drift angle is combined with the predicted structural acceleration, predicted structural velocity, and predicted structural displacement value to obtain the seismic structural response prediction result.

[0054] Cloud computing platforms perform the following tasks, including: (1) First cloud computing task: Calibrate the GateBlend-LSTM model according to the set frequency, specifically including: based on measured data (structural natural frequencies) f i Damping ratio g iInvert structural parameters (stiffness K, mass M, damping C) and optimize using an intelligent algorithm: minΣ( f i,meas - f i,FE ) 2 The subscripts meas and FE correspond to historical data and current model data, respectively, and are generated after each earthquake and once a month.

[0055] (2) Second cloud computing task, long-term degradation trend analysis: extract the natural frequency decay rate df / dt from historical data (1 year+) as a degradation index, and based on the Paris fatigue crack propagation model (da / dN=C·(ΔK) m (where a is the crack length, N is the number of cycles, ΔK is the stress intensity factor amplitude, and C and m are material constants) to predict the remaining service life.

[0056] (3) The third cloud computing task is full lifecycle data management, including: time series database: InfluxDB, which stores all historical data of sensors; relational database: PostgreSQL, which stores structure information and detection reports; query interface: supports data backtracking and statistical analysis for any time period.

[0057] (4) The fourth cloud computing task is to update the digital twin model.

[0058] Predictive maintenance is performed based on long-term degradation trend analysis obtained from the cloud computing platform, including: When the rate of decrease of the structure's natural frequency exceeds a set value, a structural reinforcement recommendation is output. When the increase in damping ratio exceeds the set value, it is determined that there is a potential crack, and detailed inspection recommendations are output. When the peak strain exceeds the material's yield strain, the component is considered to have entered the plastic stage, and a replacement recommendation is issued.

[0059] In this embodiment, the system adopts edge-cloud collaborative monitoring, including: In normal monitoring mode, the following steps are performed: the first edge computing node uploads sensor statistics (mean, standard deviation, peak value) to the cloud computing platform once per minute, and the cloud computing platform performs seismic response prediction model calibration once per hour, with the corresponding data flow rate of 88KB / s → 5KB / min (compression rate of 94%). When the peak acceleration exceeds the set value (0.05g) or any sensor value exceeds the set threshold, the earthquake response mode is triggered, and the following actions are executed: The edge computing nodes are switched to high-frequency mode and uploaded the time-aligned preprocessed sensor data and earthquake response prediction data to the cloud computing platform in seconds (or 0.5s, depending on the actual situation). The cloud computing platform initiates the emergency analysis process, performs structural damage assessment, safety warning and management control at the service layer, and updates the digital twin model (including geometry, mechanics and damage status) in real time. The data flow rate is 88KB / s (continues until the earthquake ends).

[0060] 4. Service layer.

[0061] Based on the seismic response prediction data uploaded by the digital layer, structural damage assessment and safety warning are performed, and management and control instructions are generated. The control instructions are then sent to the control layer, while the structural damage assessment results and safety warning feedback information are sent to the cloud computing platform of the digital layer.

[0062] Determining the structural damage level based on the inter-story drift angle threshold specifically includes: D0 (Intact): i max <1 / 550 = 0.0018 rad; D1 (Mild): 1 / 550≤ i max <1 / 250 = 0.004 rad, Execution: Record as a minor response event, add to the long-term monitoring database, no immediate action required; D2 (Intermediate): 1 / 250≤ i max <1 / 120 = 0.0083 rad, D3 (Severe): 1 / 120≤ i max <1 / 50 = 0.02 rad; Execution for D2 and D3 situations: Send an early warning notification to management personnel, recommend conducting detailed post-earthquake inspections, and dispatch drones for visual inspections; D4 (collapse): i max ≥1 / 50, Execution: Trigger emergency evacuation alarm (audio-visual alarm + SMS push), digital twin model highlights dangerous floors in red, record event log for subsequent investigation; Total latency of edge computing: Data reception 5ms + preprocessing 8ms + model inference 47ms + damage identification 3ms = 63ms < 100ms requirement.

[0063] 5. Control layer.

[0064] The control layer selects a control strategy to perform closed-loop control on the actuator group on the multi-story building structure according to the control instructions issued by the service layer. The actuator group includes magnetorheological dampers, variable stiffness seismic isolation bearings, and active mass tuners.

[0065] Select the optimal control algorithm based on the type of external excitation and the structural state: Seismic excitation: LQR optimal control, objective function J=∫(x T ·Q·x+u T ·R·u)dt; LQR control state variable x=[ d 1, v 1, a 1,..., d M , v M , a M ] T ∈ℝ (3M×1) Displacement (m), velocity (m / s), and acceleration (m / s²) corresponding to M floors 2 ), control input u=[ F d1 ,..., F dN ] T ∈ℝ (N×1) The control force (kN) corresponding to N dampers; LQR performance index J=∫(x T ·Q·x+u T In the equation ·R·u)dt, the selection of weight matrices Q and R reflects the engineering objectives: the diagonal elements of matrix Q... q d , q v , q a These correspond to the penalty weights for displacement, velocity, and acceleration, respectively; if a reduction in displacement is required (to prevent damage), let... q d =1e6, q v =1e3, q a =1e0; Diagonal elements of matrix R r i The control cost (power consumption, wear) corresponding to the i-th damper; Wind-induced vibration excitation: Fuzzy control, input displacement d, velocity v → output damping force F d ; Long-term fatigue: Reinforcement learning (DQN) to maximize the remaining lifetime of the structure.

[0066] Next, the control process of each actuator will be described in detail.

[0067] (1) Magnetorheological damper Working principle: The viscosity of the magnetorheological fluid is changed by adjusting the current I (0-2A), thereby changing the damping force; Control force model: F d (t)=c0·v(t)+k0·(x(t)-x0)+α·z(t), where c0 is the viscous damping coefficient, k0 is the spring stiffness, z(t) is the hysteresis variable, and α is the hysteresis component coefficient, representing the hysteresis force intensity generated by the magnetic flux structure formed by the magnetorheological fluid.

[0068] Adjustable range: maximum damping force 0-200kN, response time <20ms; Control algorithm implementation: 1) Obtain the current state x(t) = [d(t); v(t); a(t)] from the digital twin model. 2) Obtain the state x(t+Δt) for the next 5 seconds; 4) Calculate the optimal control force u * (t)=-K·x(t), where K is the LQR gain matrix, corresponding to stiffness; 5) The inverse solution requires the current I(t) to make F d (t)=u * (t); 6) Output current command to the damper driver via D / A converter.

[0069] (2) Variable stiffness seismic isolation bearing Working principle: By adjusting the contact state between the rubber layer and the friction layer, the horizontal stiffness is changed. The stiffness adjustment range is: K∈[0.5,5.0]MN / m.

[0070] Control objective: Based on the dominant period T of the seismic motion g =2π / ω peak , where ω peak To achieve the peak frequency, optimize the support stiffness K to make the isolation layer period T... b Stay away from T g .

[0071] Control Algorithm: T b =2π√(M / K), where M is the mass of the superstructure.

[0072] If T is identified in real time g =0.5s, then set T b =3s, inverse solution K opt =0.44MN / m, K opt It is sent to the actuator as a control signal.

[0073] (3) Active mass tuner Working principle: A movable mass block (mass ratio μ=2-5%) is installed on the top layer of the structure, and a reverse inertial force is applied by a servo motor.

[0074] Control: F amd (t)= m amd · a amd (t), where m amd The mass of the block (e.g., 100 tons). a amd The acceleration of the mass block; Control algorithm: 1) Real-time measurement of acceleration at the top layer of the structure a top (t); 2) Set the acceleration of the mass block a amd (t)=-G· a top (t), where G is the control gain; 3) Calculate the required motor torque T(t) and output the command.

[0075] In this embodiment, as Figure 2 As shown, the closed-loop cycle time is: data acquisition is 5ms, edge computing is 63ms, control decision is 10ms, and actuator response is 20ms, that is, the total delay is 98ms < 100ms, which meets the requirements.

[0076] In this embodiment, numerical simulations were performed to verify both no control and closed-loop control, and the results are as follows: Peak inter-story drift angle: θmax = 0.012 rad without control → 0.007 rad after control, a decrease of 41.7%; Peak acceleration at the top layer: uncontrolled PGAtop = 0.85g → controlled 0.53g, a decrease of 37.6%; Peak base shear force: uncontrolled Vbase = 3200kN → controlled Vbase = 2300kN, a decrease of 28.1%; The final output is the optimal control command {Fd(t),K(t),aamd(t)}, which is fed back to the actuator in the physical layer.

[0077] In summary, this system adopts a five-layer architecture of "physical layer - data layer - digital layer - service layer - control layer". Each layer is connected through a real-time feedback loop (delay <50ms) to achieve a complete closed loop from sensor to actuator.

[0078] Next, the core aspects of the system of this invention will be summarized and explained: (1) Five-layer DT-driven CPS architecture and real-time feedback loop design.

[0079] A layered architecture of physical layer, data layer, digital layer, service layer and control layer is proposed. Digital twin is embedded in the digital layer as the core computing engine of CPS. Each layer is connected through real-time bidirectional data flow (latency <100ms) to form a complete closed loop of perception-modeling-decision-execution.

[0080] The physical quantities (acceleration, displacement, strain) measured by the physical layer sensors are directly input into the digital layer. The response output of the digital twin model strictly corresponds to the physical sensor measurements in terms of dimensions and magnitude. The inter-story drift angle θ=Δd / h (dimensionless) calculated by the digital layer is a damage control index clearly specified in the "Code for Seismic Design of Buildings" GB50011. The service layer makes decisions based on the threshold values ​​of θ (1 / 550, 1 / 250, 1 / 50), directly corresponding to the requirements of the engineering code. The control parameters such as damping force Fd and stiffness K output by the control layer are applied to the physical structure through actuators, changing the dynamic characteristics of the structure (natural period T, damping ratio ζ), forming a closed-loop coupling between the physical and digital layers.

[0081] Beneficial effects: 1) Reduced latency: Traditional cloud processing systems have a latency of 2-5 seconds. This invention uses edge computing to push critical tasks to the edge, reducing the total system latency to 63 milliseconds, a reduction of 98.7%, which meets the real-time control requirements (<100ms). 2) Improved reliability: Edge computing nodes process data locally, maintaining basic monitoring and early warning functions even in the event of cloud platform failure, increasing system availability from 95% to 99.5%; 3) Enhanced scalability: The layered architecture supports flexible expansion. In a large project, the system expanded from 50 sensors to 500 sensors + 20 actuators by simply adding edge computing nodes without refactoring the overall architecture.

[0082] (2) Time alignment of heterogeneous sensors based on Kalman filtering For sensors with different sampling rates, such as accelerometers (200Hz), GPS displacement gauges (10Hz), and strain gauges (100Hz), a time alignment algorithm based on Kalman filtering is proposed. By fusing multi-source data through a state-space model, time synchronization with an accuracy of ±1ms is achieved.

[0083] The Kalman filter state vector x = [ds;vs;as] corresponds to the displacement-velocity-acceleration state-space representation of structural dynamics, and the state equation x k =A·x k-1 +B·u k The transition matrix A is derived from the structure's mass M, stiffness K, and damping C matrices: A = [I, Δt·I, 0; 0, I, Δt·I; 0, -M] -1 K,-M -1 C] This is the state-space form of the Newmark-β integral method in structural dynamics; The design of the observation matrix H reflects the characteristics of different sensors measuring different physical quantities: GPS displacement measurement: H GPS =[1,0,0] (ds only observed), accelerometer: H ACC =[0,0,1] (as only), strain gauge: H ε =[0,0,K / AE] (Observed strain ε=σ / E=F / (AE)∝a, where A is the cross-sectional area and E is the elastic modulus); The measurement noise covariance R is set based on the sensor accuracy: R GPS =diag((5mm) 2 Corresponding GPS displacement accuracy, R ACC =diag((0.001g) 2 Corresponding accelerometer resolution; Beneficial effects: 1) Improved synchronization accuracy: The synchronization error of traditional timestamp alignment methods is 10-50ms. The Kalman filtering algorithm of this invention reduces the error to ±1ms, which is 10-50 times higher. 2) Data fusion accuracy: In a building structure test, the GPS measured the vertex displacement as 127.3 mm, the second integral of the acceleration was 126.8 mm, and the Kalman fusion result was 127.1 mm, with an error of only 0.1 mm compared to the true value of 127.0 mm from the laser displacement meter. 3) Missing data imputation capability: For 5% data packet loss (simulating network failure), the RMSE after Gaussian process regression interpolation is 2.3mm, and the error rate of relative peak displacement (120mm) is 1.9%, which meets the engineering requirements (<5%). (3) Edge-cloud collaborative digital twin computing engine We propose a task allocation strategy for edge computing and cloud computing, which deploys tasks with high real-time requirements (data preprocessing, response prediction, and preliminary damage identification) on edge computing nodes (latency <100ms) and deploys computationally intensive tasks (proxy model calibration and long-term trend analysis) on the cloud platform (latency <5s), achieving an optimal balance between performance and cost.

[0084] The GateBlend-LSTM model input for edge computing nodes is the ground acceleration a. g (t)∈ℝ^(N×1), the output is the structural response {a,v,d}∈ℝ^(N×M), where M is the number of monitored floors (e.g., M=3 for a 3-story structure) and N is the number of time steps (60 seconds × 200Hz = 12000 steps).

[0085] Beneficial effects: Improved computational efficiency: Edge computing node inference time is 47ms (GPU), cloud platform FE analysis time is 3.5 seconds, and the overall response time is <5 seconds, which is 6-12 times faster than the 30-60 seconds of pure cloud solutions; Bandwidth saving: In normal monitoring mode, edge computing nodes only upload statistical data (3.2KB / min), which saves 98% of bandwidth compared to uploading raw data (54KB / s), reducing cloud platform storage and communication costs; Offline robustness: During a cloud platform maintenance period (2 hours), the edge computing nodes continue to provide real-time monitoring and early warning services, ensuring that system availability is unaffected and critical security functions are not interrupted.

[0086] (4) Multi-actuator cooperative closed-loop control based on digital twin prediction A complete closed loop of "monitoring-digital twin prediction-optimal control decision-multi-actuator coordination" is established. The optimal control command is generated by using the 5-second response prediction results of the digital twin model and algorithms such as LQR / fuzzy / reinforcement learning. This coordinates multiple actuators such as magnetorheological dampers, variable stiffness seismic isolation bearings, and active mass tuners to achieve intelligent adaptive control of the structure.

[0087] Beneficial Effects: Significantly Reduced Response: Under the action of Kobe wave (PGA=0.4g), closed-loop control with 3-layer frame + 3 MR dampers: peak inter-layer displacement angle: uncontrolled 0.012rad → controlled 0.007rad, a reduction of 41.7%; peak top-layer acceleration: uncontrolled 0.85g → controlled 0.53g, a reduction of 37.6%; peak base shear force: uncontrolled 3200kN → controlled 2300kN, a reduction of 28.1%; Multi-actuator Synergistic Efficiency: On a bridge (main span 500m), deploying 10 MR dampers + 4 variable stiffness supports, the synergistic control resulted in an additional 15-20% reduction in response compared to single-type actuators; Reduced Energy Consumption: Feedforward control based on digital twin prediction, compared to pure feedback control, reduced damper current consumption by 32%, extending equipment lifespan.

[0088] (5) Standardized data interface and support for multiple application scenarios The framework features a unified API interface and data protocol (RESTful + MQTT), supporting multiple sensor types (acceleration, displacement, strain, wind speed, etc.) and multiple communication protocols (LoRa, ZigBee, 4G / 5G), enabling plug-and-play and rapid deployment of the system. It supports various application scenarios including structural health monitoring, predictive maintenance, AI-driven optimization, real-time control, and emergency response.

[0089] Beneficial effects: Improved deployment efficiency: 50 new sensors were added to a 20-story building using a plug-and-play interface. The entire process from equipment installation to system integration took only 2 days, a reduction of 85% compared to the 1-2 weeks of traditional systems.

[0090] Cross-platform compatibility: The system is compatible with accelerometers from 3 brands (PCB, B&K, Dytran) and GPS devices from 2 brands (Trimble, Leica). No core code modification is required; only sensor parameter files need to be configured.

[0091] Multi-scenario application validation: Structural health monitoring: Deployed in 15 buildings, continuous monitoring for 2+ years, with anomaly detection accuracy of 95.3%; Predictive maintenance: Predicts structural component damage 3-6 months in advance, reducing maintenance costs by 20%; Real-time control: The damper was automatically activated during two moderate earthquakes (PGA=0.15g, 0.23g), resulting in a 35% reduction in response. Emergency Response: Complete damage assessment within 60 seconds of a strong earthquake (PGA=0.38g) and guide the orderly evacuation of 2,000 people.

[0092] Figure 3 For system latency comparison (from sensor data acquisition to control command output), the bar chart in the figure compares the three architectures. The traditional cloud system has an average latency of 3.2 seconds, the edge-assisted system has an average latency of 480 milliseconds, while the DT-CPS of this invention has an average latency of 63 milliseconds, which meets the real-time control requirements.

[0093] Figure 4 For system deployment time comparison, it can be seen that the traditional system deployment time is: 3 days for hardware installation + 7 days for software integration + 3 days for debugging = 13 days, while the system deployment time of this invention is: 1.5 days for hardware installation + 0.5 days for plug-and-play software = 2 days, which is 85% shorter.

[0094] In summary, this invention addresses the technical bottlenecks of traditional structural health monitoring systems in terms of real-time performance, data synchronization, intelligent control, and scalability through a five-layer DT-driven CPS architecture, edge-cloud collaborative computing, heterogeneous sensor time alignment, and multi-actuator closed-loop control. Verification in multiple engineering projects demonstrates that: (1) Improved real-time performance: The total system latency has been reduced from the traditional 2-5 seconds to 63 milliseconds, a reduction of 98.7%, meeting the real-time control requirements (<100ms) and realizing the technological leap from "post-event analysis" to "real-time early warning".

[0095] (2) Data synchronization accuracy: The time alignment accuracy of heterogeneous sensors reaches ±1ms, and the displacement RMSE after multi-source data fusion is 1.2mm, which meets the high precision requirements of engineering (<5mm). (3) Significant control effect: seismic response is reduced by 30-50% (displacement 41.7%, acceleration 37.6%, shear force 28.1%), significantly improving structural safety and service performance.

[0096] (4) Wide range of engineering applications: deployed in 15 buildings and 3 bridges; monitored for more than 2 years; anomaly detection accuracy rate of 95.3%; maintenance costs reduced by 20%; and successfully achieved automatic control and emergency response in 2 earthquakes.

[0097] (5) The system is highly scalable: it supports expansion from a single structure (50 sensors) to a regional network (500 sensors), and the deployment cycle is shortened from 13 days to 2 days, reducing the cost by 85%.

[0098] This invention provides a complete technical solution for next-generation intelligent infrastructure, promoting a technological transformation in structural engineering from passive monitoring to active control, from single-unit structures to networked management, and from manual inspection to intelligent early warning. It has significant engineering application value and socio-economic benefits.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A seismic response prediction and control system for multi-story building structures, characterized in that, include: The physical layer includes sensor groups and actuator groups. Sensor groups deployed on multi-story building structures collect sensing data and upload it to the data layer. The physical layer receives control commands from the control layer to control the actuator groups, which respond to changes in multi-story building structures during earthquakes. The data layer receives sensor data from the physical layer, performs time-aligned preprocessing on the sensor data, and then uploads the time-aligned preprocessed sensor data to the digital layer. The digital layer includes edge computing nodes for seismic response prediction, as well as a cloud computing platform for calibrating seismic response prediction models in the edge computing nodes, managing full lifecycle data, and generating digital twin models. The service layer performs structural damage assessment and safety warning based on the seismic response prediction data uploaded by the digital layer, generates management and control instructions, sends the control instructions to the control layer, and simultaneously sends the structural damage assessment results and safety warning feedback information to the cloud computing platform of the digital layer. The control layer performs closed-loop control of the actuator group on the multi-story building structure based on the control commands issued by the service layer.

2. The seismic response prediction and control system for multi-story building structures according to claim 1, characterized in that, The sensor group includes accelerometers, GPS displacement gauges, strain gauges, and anemometers. Accelerometers and GPS displacement gauges are installed on each floor of the multi-story building structure, a set number of strain gauges are installed on key sections of the beams and columns of the multi-story building structure, and anemometers are installed on both sides of the roof of the multi-story building structure.

3. A seismic response prediction and control system for multi-story building structures according to claim 1, characterized in that, In the data layer, after receiving sensor data from the physical layer, time-aligned preprocessing is performed on the sensor data, specifically as follows: Receive sensor data from the physical layer, where all sensors are timed via GPS or BeiDou, and the data packet format is {sensor ID, UTC timestamp, sampling rate, data value}. For sensor data with different sampling rates, a time alignment algorithm based on Kalman filtering is adopted. This algorithm fuses multi-source heterogeneous sensor data through a state-space model to perform time synchronization and alignment, resulting in synchronized data with a unified time reference. a sync (t), v sync (t), d sync (t), ε sync (t)}, where a sync (t), v sync (t), d sync (t), ε sync (t) represents the acceleration, velocity, displacement, and strain data at step t after time alignment.

4. A multi-story seismic response prediction and control system for multi-story building structures according to claim 1, characterized in that, The digital layer includes multiple edge computing nodes, each of which executes a corresponding edge computing task, including: The first edge computing node is used for data preprocessing: filtering, baseline correction, and feature extraction of time-aligned preprocessed sensor data. The second edge computing node is used for real-time seismic response prediction: based on the most recently set-duration acceleration window data, it uses the GateBlend-LSTM model to predict the seismic response, obtaining seismic response data for the future set-duration period. This seismic response data includes acceleration response data, velocity response data, and displacement response data; and... The third edge computing node calculates the inter-story drift angle based on the displacement response data obtained from the second edge computing node, which is used by the service layer to perform structural damage assessment based on the inter-story drift angle threshold.

5. A multi-story seismic response prediction and control system for multi-story building structures according to claim 4, characterized in that, In the second edge computing node, the GateBlend-LSTM model is used for seismic response prediction, specifically including: Ground acceleration window data is input, and structural displacement prediction values ​​are obtained by fusing data-driven point-by-point displacement prediction values ​​and physics-driven integral displacement prediction values ​​through the GateBlend gated fusion mechanism. In the GateBlend-LSTM model, data-driven point-by-point velocity prediction values ​​and data-driven point-by-point displacement prediction values ​​are predicted based on the LSTM backbone network and parallel velocity prediction heads and displacement prediction heads. Numerical integration is performed on the data-driven point-by-point velocity prediction values ​​to obtain the physics-driven integral displacement prediction values. The inter-story drift angle used for structural damage assessment is calculated based on the predicted structural displacement values. The inter-story drift angle is combined with the predicted values ​​of structural acceleration, structural velocity, and structural displacement to obtain the seismic structural response prediction results; wherein the predicted values ​​of structural acceleration are obtained by the LSTM backbone network and acceleration prediction head in the GateBlend-LSTM model, and the predicted values ​​of structural velocity are obtained by the LSTM backbone network and velocity prediction head in the GateBlend-LSTM model.

6. A seismic response prediction and control system for multi-story building structures according to claim 4, characterized in that, The cloud computing platform performs the following tasks, including: First cloud computing task: Calibrate the GateBlend-LSTM model according to the set frequency; The second cloud computing task is long-term degradation trend analysis: extracting the natural frequency decay rate as a degradation indicator from historical data, and predicting the remaining service life. The third cloud computing task is full lifecycle data management; and The fourth cloud computing task is to update the digital twin model.

7. A seismic response prediction and control system for multi-story building structures according to claim 6, characterized in that, The system employs edge-cloud collaborative monitoring, including: In normal monitoring mode, the following steps are performed: the first edge computing node uploads sensor statistics data to the cloud computing platform once per minute, and the cloud computing platform performs earthquake response prediction model calibration once per hour; When the peak acceleration exceeds the set value or any sensor value exceeds the set threshold, the earthquake response mode is triggered, and the following actions are executed: The edge computing nodes are switched to high-frequency mode, and the time-aligned preprocessed sensor data and seismic response prediction data are uploaded to the cloud computing platform in seconds. The cloud computing platform then initiates an emergency analysis process to perform structural damage assessment, safety warning, and management control at the service layer.

8. A seismic response prediction and control system for multi-story building structures according to claim 7, characterized in that, The structural damage assessment, safety early warning, and management control at the service layer include: Determining the structural damage level based on the inter-story drift angle threshold; Based on the determined level of structural damage, corresponding safety warnings and management controls are implemented.

9. A seismic response prediction and control system for multi-story building structures according to claim 6, characterized in that, The service layer also includes predictive maintenance based on long-term degradation trend analysis obtained from the cloud computing platform, including: When the rate of decrease of the structure's natural frequency exceeds a set value, a structural reinforcement recommendation is output. When the increase in damping ratio exceeds the set value, it is determined that there is a potential crack, and detailed inspection recommendations are output. When the peak strain exceeds the material's yield strain, the component is considered to have entered the plastic stage, and a replacement recommendation is issued.

10. A seismic response prediction and control system for multi-story building structures according to claim 1, characterized in that, The control layer selects a control strategy to perform closed-loop control on the actuator group on the multi-story building structure according to the control instructions issued by the service layer. The actuator group includes a magnetorheological damper, a variable stiffness seismic isolation bearing, and an active mass tuner.

Images