Intelligent Optimization System and Method for Vibration Control of Steel Space Frame Structure

Abstract

The present invention relates to the field of steel grid structure engineering, and particularly to an intelligent optimization system and method for vibration control of steel grid structures. The present invention proposes to collect displacement, acceleration and strain data through a multi-physical quantity fusion monitoring module, perform signal filtering, feature extraction and anomaly recognition through an edge intelligent data processing module, and generate structured feature data; a dynamic response prediction module outputs structural dynamic response predictions at micro, meso and macro scales based on a deep learning model; a digital twin decision-making module conducts scenario rehearsal analysis in combination with the prediction results to generate an optimal control strategy; an adaptive active control module executes control operations to achieve vibration suppression and deformation control; the system solves the problems in the prior art of only focusing on static deformation measurement, passive monitoring method, lagging response and low processing efficiency, and improves the real-time performance and effectiveness of vibration control.

Description

Technical Field

[0001] The present invention relates to the field of steel grid structure engineering, and particularly to an intelligent optimization system and method for vibration control of steel grid structures. Background Art

[0002] With the development of modern architecture towards large-span and lightweight directions, steel grid structures are widely used in large public buildings such as stadiums, exhibition centers, and airport terminals due to their flexible shapes, large spans, and light weights. During the construction process of steel grid structures, especially in the overall lifting and installation stage, due to their large flexibility and small damping, unexpected vibrations and deformations are likely to occur, leading to potential structural safety hazards.

[0003] For example, Chinese patent application with publication number CN117128884A discloses "a deformation monitoring method for steel grid installation projects based on three-dimensional laser scanning technology". This method uses three-dimensional laser scanning equipment to perform laser measurements on the overall lifted grid at different stages, and through splicing and cleaning the measured data, key spherical joint data is obtained for structural comparison to determine whether the deformation meets expectations and whether the deformation is within the control range. However, this method has the following deficiencies:

[0004] 1) Only focuses on static deformation measurement, lacks the monitoring and analysis of the dynamic response characteristics of the structure, and cannot capture transient vibration characteristics;

[0005] 2) Adopts a passive monitoring method, and only performs manual intervention after problems are discovered, resulting in a lag in response and low processing efficiency;

[0006] 3) Lacks the ability to predict the deformation trend, can only detect the occurred deformation, and cannot give early warnings;

[0007] 4) Does not incorporate environmental factors such as temperature and wind load into the monitoring system, affecting the accuracy of monitoring results;

[0008] 5) The monitoring and control are separated, and active control of the structural vibration and deformation cannot be achieved;

[0009] 6) Data processing and analysis rely on a large amount of manual intervention, lacking intelligent and automated processing capabilities.

[0010] In view of the above technical problems, there is an urgent need for an intelligent system that can monitor, accurately predict, and actively control the vibration and deformation of steel grid structures during the lifting process to improve the safety and reliability of steel grid structure installation. Summary of the Invention

[0011] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an intelligent optimization system and method for vibration control of steel grid structures to achieve all-round intelligent management of steel grid structures during the lifting and installation process.

[0012] The present invention proposes an intelligent optimization system for vibration control of a steel grid structure, including:

[0013] A multi-physical quantity fusion monitoring module, which is used to collect displacement, acceleration and strain data of the steel grid structure, and transmit the data to the data processing module;

[0014] An edge intelligent data processing module, which is communicatively connected to the multi-physical quantity fusion monitoring module, and is used to receive the data collected by the multi-physical quantity fusion monitoring module, perform signal filtering, feature extraction and anomaly recognition, and generate structured feature data;

[0015] A dynamic response prediction module, which is communicatively connected to the edge intelligent data processing module, and is used to receive the structured feature data, and generate structural dynamic response prediction results at three scales of micro, meso and macro based on a deep learning model;

[0016] A digital twin decision-making module, which is communicatively connected to the dynamic response prediction module, and is used to receive the prediction results, perform scenario rehearsal analysis in combination with a multi-resolution digital twin model, and generate an optimal control strategy;

[0017] An adaptive active control module, which is communicatively connected to the digital twin decision-making module, and is used to receive the optimal control strategy, and perform control operations through an active vibration suppression system, a precise tension regulation system and a stiffness dynamic regulation system, so as to achieve vibration suppression and deformation control of the steel grid structure.

[0018] Preferably, the multi-physical quantity fusion monitoring module includes:

[0019] Type-I node units, which are used to collect displacement data of the steel grid structure;

[0020] Type-II node units, which are used to collect acceleration data of the steel grid structure;

[0021] Type-III node units, which are used to collect strain data of the steel grid structure;

[0022] A node layout unit, which is used to determine the optimal layout position of sensors according to the improved K-means clustering algorithm;

[0023] An adaptive sampling unit, which is used to dynamically adjust the sampling frequency and accuracy according to the lifting stage of the steel grid.

[0024] Preferably, the node layout unit determines the sensor layout density according to the following density formula:

[0025] ,

[0026] Where: is a coefficient, and its value range is 0.35 - 0.45; is the first-order frequency of the steel grid structure; is the second-order frequency of the steel grid structure; is the maximum expected displacement; is the structural characteristic dimension.

[0027] Preferably, the edge intelligent data processing module includes:

[0028] The sensor embedded processing layer is used to perform initial data screening and compression;

[0029] The regional edge node layer is used to perform wavelet noise reduction and data fusion;

[0030] The main edge server layer is used to complete feature extraction and pattern recognition;

[0031] The anomaly recognition unit is used to implement a dual-threshold anomaly detection strategy and dynamically update the threshold according to the lifting speed and acceleration factor.

[0032] Preferably, the dynamic response prediction module includes:

[0033] The micro prediction unit is used to predict the short-term response changes of key nodes based on the spatio-temporal graph convolutional network;

[0034] The meso prediction unit is used to predict the mid-term dynamic characteristic evolution of the structure based on the bidirectional gated recurrent unit network;

[0035] The macro prediction unit is used to predict the long-term trend and risks of the entire lifting process based on the attention-enhanced Transformer-XL model;

[0036] The incremental training unit is used to update the model by combining knowledge distillation and federated learning.

[0037] Preferably, the digital twin decision module includes:

[0038] The multi-resolution model unit is used to establish a digital twin system composed of a physical model, a behavior model, and a scenario model;

[0039] The state synchronization unit is used to map the monitoring data to the digital twin model in real time;

[0040] The scenario rehearsal unit is used to quickly simulate the structural responses under multiple lifting strategies;

[0041] The risk assessment unit is used to calculate the risk index and reliability of different decision-making schemes;

[0042] The scheme generation unit is used to recommend the optimal control strategy based on the multi-objective optimization algorithm.

[0043] Preferably, the risk assessment unit quantifies risks based on a state-action-value matrix, which includes a state vector, an action vector, a value matrix, and constraint conditions, and dynamically updates parameters through a hierarchical Bayesian method.

[0044] Preferably, the adaptive active control module includes:

[0045] An active vibration suppression unit for suppressing the vibration of the steel grid structure through an electromagnetic-hydraulic composite damper;

[0046] A precise tension regulation unit for balancing the tension distribution during the lifting of the steel grid structure through a distributed intelligent lifter;

[0047] A stiffness dynamic adjustment unit for adjusting the local stiffness of the structure through variable stiffness connection nodes;

[0048] A hybrid control strategy unit for switching among model predictive control, H∞ robust control, and sliding mode control strategies according to the structural state;

[0049] A multi-loop feedback unit for implementing a three-loop nested feedback control of the actuator state, local response, and overall performance.

[0050] Preferably, the hybrid control strategy unit is provided with an energy consumption adaptive optimization algorithm to control the ratio of energy consumption to vibration energy to be less than 0.35 under normal conditions and less than 0.6 under extreme conditions.

[0051] An intelligent optimization method for vibration control of a steel grid structure, applied to the system, includes the following steps:

[0052] Collect displacement, acceleration, and strain data of the steel grid structure through a multi-physical quantity fusion monitoring module;

[0053] Transmit the data to an edge intelligent data processing module for signal filtering, feature extraction, and anomaly recognition to generate structured feature data;

[0054] Input the structured feature data into a dynamic response prediction module to generate structural dynamic response prediction results at three scales of micro, meso, and macro based on a deep learning model;

[0055] Input the prediction results into a digital twin decision-making module, conduct scenario rehearsal analysis in combination with a multi-resolution digital twin model, and generate an optimal control strategy;

[0056] Receive the optimal control strategy through the adaptive active control module, and perform active vibration suppression, precise tension regulation, and stiffness dynamic adjustment to achieve vibration suppression and deformation control of the steel grid structure.

[0057] The present invention has the following beneficial effects compared with the prior art:

[0058] 1) Expand from static monitoring to dynamic response analysis, comprehensively capture the transient vibration characteristics of the steel grid structure during the lifting process, and improve the comprehensiveness and accuracy of safety monitoring;

[0059] 2) Establish a closed-loop automation system of prediction - decision - control, realize the transformation from passive monitoring to active control, can quickly respond and suppress harmful vibrations, and prevent problems before they occur;

[0060] 3) Integrate multi - physical - quantity sensing data, realize multi - scale analysis from microscopic components to macroscopic wholes, and provide strong support for comprehensively understanding the structural behavior;

[0061] 4) Introduce deep learning and digital twin technologies, the system has the ability of adaptive learning and knowledge accumulation, continuously optimize the decision - making model, and improve the prediction accuracy and control effect;

[0062] 5) The modular design and distributed architecture make the system highly scalable and adaptable, and can meet the requirements of steel grid structure projects of different scales and different forms. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is the overall framework diagram of the intelligent optimization system for vibration control of the steel grid structure of the present invention;

[0064] Figure 2 It is the structural schematic diagram of the multi - physical - quantity fusion monitoring module of the present invention;

[0065] Figure 3 It is the structural schematic diagram of the edge intelligent data processing module of the present invention;

[0066] Figure 4 It is the structural schematic diagram of the dynamic response prediction module of the present invention;

[0067] Figure 5 It is the structural schematic diagram of the digital twin decision - making module of the present invention;

[0068] Figure 6 It is the structural schematic diagram of the adaptive active control module of the present invention;

[0069] Figure 7 It is the flow chart of the intelligent optimization method for vibration control of the steel grid structure of the present invention;

[0070] Figure 8 It is the effect comparison diagram of the present invention applied to the steel grid lifting project of a stadium. DETAILED DESCRIPTION OF THE INVENTION

[0071] Please refer to the appendix Figure 1-8 , the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention, asFigure 1 As shown in the figure, the intelligent optimization system for vibration control of the steel grid structure provided by the present invention includes a multi - physical - quantity fusion monitoring module 1, an edge intelligent data processing module 2, a dynamic response prediction module 3, a digital twin decision - making module 4, and an adaptive active control module 5.

[0072] The multi - physical - quantity fusion monitoring module 1 is used to collect displacement, acceleration, and strain data of the steel grid structure, and transmit the data to the data processing module. By deploying multiple types of sensors, this module forms a distributed monitoring network to achieve a comprehensive perception of the state of the steel grid structure.

[0073] The edge intelligent data processing module 2 is communicatively connected to the multi - physical - quantity fusion monitoring module 1, and is used to receive the data collected by the multi - physical - quantity fusion monitoring module 1, perform signal filtering, feature extraction, and anomaly recognition, and generate structured feature data. This module adopts an edge - computing architecture to process data at the data source, reducing transmission delay and improving processing efficiency.

[0074] The dynamic response prediction module 3 is communicatively connected to the edge intelligent data processing module 2, and is used to receive the structured feature data and generate structural dynamic response prediction results at three scales: micro - scale, meso - scale, and macro - scale based on a deep - learning model. This module can predict multi - scale dynamic characteristics from short - term responses of key nodes to long - term overall trends.

[0075] The digital twin decision - making module 4 is communicatively connected to the dynamic response prediction module 3, and is used to receive the prediction results, conduct scenario rehearsal analysis in combination with a multi - resolution digital twin model, and generate an optimal control strategy. This module evaluates the effects of different control schemes through virtual simulation to provide theoretical support for decision - making.

[0076] The adaptive active control module 5 is communicatively connected to the digital twin decision - making module 4, and is used to receive the optimal control strategy and perform control operations through an active vibration suppression system, a precise tension regulation system, and a stiffness dynamic adjustment system to achieve vibration suppression and deformation control of the steel grid structure. This module converts decisions into actual control actions to form a closed - loop control system.

[0077] The above - mentioned modules are interconnected through a high - speed communication network to form a complete closed - loop of data acquisition → processing → prediction → decision - making → control, realizing intelligent monitoring, prediction, and control of the vibration of the steel grid structure. The overall working process of the system is as follows: The multi - physical - quantity fusion monitoring module 1 continuously collects structure state data, the edge intelligent data processing module 2 processes the original data into structured feature data, the dynamic response prediction module 3 predicts the future state of the structure based on the feature data, the digital twin decision - making module 4 generates a control strategy according to the prediction results, and the adaptive active control module 5 executes control operations to effectively suppress the vibration of the steel grid structure.

[0078] AsFigure 2 As shown in Figure 2 , the multi-physical quantity fusion monitoring module 1 includes type-I node unit 11, type-II node unit 12, type-III node unit 13, node layout unit 14 and adaptive sampling unit 15.

[0079] The type-I node unit 11 is used to collect displacement data of the steel grid structure. This unit uses a high-precision displacement sensor with an accuracy of ±0.05 mm and a sampling rate of 100 Hz. It is mainly deployed at key positions such as spherical joints to monitor the static deformation and low-frequency dynamic displacement of the structure. Preferably, the type-I node unit 11 uses a fiber Bragg grating displacement sensor, which has the advantages of anti-electromagnetic interference, corrosion resistance, and long service life.

[0080] The type-II node unit 12 is used to collect acceleration data of the steel grid structure. This unit uses a triaxial acceleration sensor with a measurement range of ±16 g and a sampling rate of 1000 Hz. It is mainly used to capture the high-frequency vibration characteristics of the structure. Preferably, the type-II node unit 12 uses a MEMS acceleration sensor, which is small in size and low in power consumption, suitable for a large number of distributed deployments.

[0081] The type-III node unit 13 is used to collect strain data of the steel grid structure. This unit uses a strain sensor with a resolution of 0.1 με and a sampling rate of 500 Hz. It is mainly deployed on load-bearing components such as rods to monitor the stress state of the structure. Preferably, the type-III node unit 13 uses a full-bridge strain gauge with temperature self-compensation to reduce the influence of ambient temperature on the measurement.

[0082] The node layout unit 14 is used to determine the optimal layout positions of sensors according to the improved K-means clustering algorithm. This unit first conducts a preliminary analysis based on the geometric and mechanical characteristics of the steel grid structure to determine the key monitoring areas, and then applies the improved K-means clustering algorithm to optimize the sensor layout scheme. Preferably, the sensor layout follows the principle of full coverage of key nodes and sampling detection in general areas to control costs while ensuring the monitoring effect.

[0083] The node layout unit 14 determines the sensor layout density according to the following density formula:

[0084] ,

[0085] where, is the sensor layout density, with the unit of number / m², is the sensitivity coefficient, dimensionless, with a value range of 0.35 - 0.45, is the first natural frequency of the structure, with the unit of Hz, is the second natural frequency of the structure, with the unit of Hz, is the maximum expected displacement, with the unit of m, is the structural characteristic dimension, with the unit of m, is the scale conversion factor, with the unit of s², and its value is s² / m, which is used for unit conversion. The coefficient C represents the structural response sensitivity coefficient, reflecting the sensitivity of different types of steel grid structures to vibration. A smaller C value (0.35) is applicable to structures with greater stiffness and lower vibration sensitivity; a larger C value (0.45) is applicable to structures with greater flexibility and higher vibration sensitivity. This formula optimizes the number of sensors while ensuring the monitoring effect by combining the structural dynamic characteristics ( ), and the possible deformation amount ( ). The formula comprehensively considers the structural dynamic characteristics and the expected deformation, making the sensor layout both economical and effective.

[0086] The adaptive sampling unit 15 is used to dynamically adjust the sampling frequency and accuracy according to the lifting stage of the steel grid structure. When the structure is stationary or changing slowly, a low-frequency sampling mode is adopted; when lifting operations are performed or an anomaly is detected, it automatically switches to a high-frequency sampling mode. Preferably, the adaptive sampling unit 15 also has an energy management function, which extends the working time of the sensor network by intelligently scheduling the sampling strategy.

[0087] The multi-physical quantity fusion monitoring module 1 also includes a network topology self-repair function. Based on the dynamic routing mechanism of the Ad-hoc network protocol, when a certain sensing node fails, the system can automatically adjust the network topology structure to ensure the stability and reliability of data transmission.

[0088] As Figure 3 shown, the edge intelligent data processing module 2 includes a sensor embedded processing layer 21, a regional edge node layer 22, a main edge server layer 23, and an anomaly recognition unit 24.

[0089] The sensor embedded processing layer 21 is used to perform initial data screening and compression. This layer directly performs preliminary processing at the sensor end, eliminates obvious abnormal data, and uses a lightweight compression algorithm to reduce the amount of data transmitted. Preferably, the sensor embedded processing layer 21 deploys a Savitzky-Golay filter to effectively suppress high-frequency noise, reduce the signal-to-noise ratio, and improve the accuracy of subsequent processing.

[0090] The regional edge node layer 22 is used to perform wavelet denoising and data fusion. This layer collects data from multiple sensors within the region and uses multi-resolution wavelet analysis for signal denoising and feature extraction. Preferably, the regional edge node layer 22 uses a five-level discrete wavelet transform, which can effectively separate the signal characteristics of different frequency bands for subsequent analysis.

[0091] The main edge server layer 23 is used to complete feature extraction and pattern recognition. This layer aggregates the processing results of all regional nodes, performs advanced feature extraction and pattern recognition, and generates structured feature data. Preferably, the main edge server layer 23 adopts a feature enhancement matrix construction method, combines time-domain and frequency-domain features, and forms a feature vector set that comprehensively describes the structural state.

[0092] The anomaly recognition unit 24 is used to implement a dual-threshold anomaly detection strategy and dynamically update the threshold according to the lifting speed and acceleration factor.

[0093] The dual-threshold strategy of the anomaly recognition unit 24 needs to clarify the anomaly recognition index and processing flow:

[0094] Anomaly recognition index:

[0095] 1. Displacement anomaly index: ;

[0096] 2. Acceleration anomaly index: ;

[0097] 3. Strain anomaly index: ;

[0098] 4. Comprehensive anomaly index: Where is the weight coefficient, which is dynamically adjusted according to different structural characteristics.

[0099] Among them, is the displacement anomaly index, dimensionless, is the measured displacement value, unit m, is the predicted displacement value, unit m, is the allowable displacement limit of the steel grid structure design, unit m, is the acceleration anomaly index, dimensionless, is the measured acceleration value, unit m / s2, is the reference acceleration value under normal conditions of the steel grid, unit m / s2, is the structural safety critical acceleration value, unit m / s2, is the strain anomaly index, dimensionless, is the measured strain value, unit με (microstrain).

[0100] Threshold determination method:

[0101] 1. Soft threshold Determined by statistical analysis of historical data, generally taking the value of Where is the mean value of the comprehensive index under normal conditions, is the standard deviation;

[0102] 2. Hard Threshold Determined based on the structural safety limit, generally taking a value of or directly taking 80% of the allowable value of the structural design;

[0103] Threshold dynamic update formula:

[0104] ,

[0105] where, is the basic threshold, determined according to the structural safety standard; is the lifting speed factor, , is the lifting acceleration factor, , α and β are weight coefficients, usually with a value range of 0.1 - 0.3. This dynamic threshold mechanism can adapt to the monitoring requirements under different working conditions and improve the accuracy of anomaly detection.

[0106] Anomaly handling process:

[0107] Normal state, conventional monitoring mode;

[0108] 2. When : Enhance the monitoring state, increase the sampling frequency, and activate the early warning;

[0109] : Emergency state, trigger the emergency control strategy, and pause the lifting if necessary;

[0110] The computing resource allocation ratio of the edge intelligent data processing module 2 is 10% / 30% / 60% (sensor / regional node / main server), the processing delay threshold is <50ms (emergency event) / <200ms (conventional monitoring), and the data compression rate is >85% (lossless compression) / >95% (lossy compression, allowing 0.5% error). This hierarchical edge computing architecture not only reduces the data transmission volume and delay but also ensures the accuracy and reliability of the processing results.

[0111] As Figure 4 shown, the dynamic response prediction module 3 includes a micro prediction unit 31, a meso prediction unit 32, a macro prediction unit 33, and an incremental training unit 34.

[0112] The micro-prediction unit 31 is used to predict the short-term response changes of key nodes based on the spatio-temporal graph convolutional network. This unit constructs a graph structure with the spherical nodes of the steel grid structure as vertices and the members as edges, and uses the spatio-temporal graph convolutional network (ST-GCN) to capture the spatial correlation and temporal changes between local nodes, and predict the response changes of key nodes within 5 to 30 seconds. Preferably, the micro-prediction unit 31 uses a convolution kernel combining 3×3 spatial convolution and 5×1 temporal convolution, and uses LeakyReLU as the activation function with an α value of 0.2 to effectively extract local spatio-temporal features.

[0113] The meso-prediction unit 32 is used to predict the evolution of the mid-term dynamic characteristics of the structure based on the bidirectional gated recurrent unit network. This unit uses the bidirectional gated recurrent unit network (Bi-GRU) to analyze the evolution trend of the overall dynamic characteristics of the structure within 1 to 10 minutes. Preferably, the meso-prediction unit 32 sets the hidden layer dimension to 256, the time step is dynamically adjusted between 5 and 15 seconds, and the sequence length is dynamically adjusted to 64 - 256 according to the current lifting stage, which can effectively capture the change law of the mid-term structural response.

[0114] The macro-prediction unit 33 is used to predict the long-term trend and risks throughout the lifting process based on the attention-enhanced Transformer-XL model. This unit focuses on the long-term trend prediction of 10 to 60 minutes and evaluates the potential risks of the entire lifting process. Preferably, the macro-prediction unit 33 uses an 8-head attention mechanism, 6-layer encoder, and 512-dimensional embedding vector, which can effectively process long time series data and capture long-distance dependence relationships.

[0115] The incremental training unit 34 is used to update the model by combining knowledge distillation and federated learning. This unit continuously learns from historical data and real-time monitoring and continuously optimizes the prediction model. Preferably, the incremental training unit 34 uses the following loss function:

[0116] ,

[0117] where: MSE is the mean square error, which is used to evaluate the overall deviation between the predicted value and the actual value; MAE is the mean absolute error, which reduces the influence of outliers; KL is the KL divergence, which measures the difference between the predicted distribution and the actual distribution; 、 、 are weight coefficients, which are dynamically adjusted according to the prediction task.

[0118] The prediction accuracy evaluation metrics of the dynamic response prediction module 3 are as follows: short-term prediction (5 seconds) accuracy > 95%, (30 seconds) accuracy > 92%; medium-term prediction (1 minute) accuracy > 90%, (10 minutes) accuracy > 85%; long-term prediction (10 minutes) accuracy > 80%, (60 minutes) accuracy > 75%. This multi-scale prediction architecture can meet the prediction requirements of different time spans and provide comprehensive data support for control decisions.

[0119] As Figure 5 shown, the digital twin decision module 4 includes a multi-resolution model unit 41, a state synchronization unit 42, a scenario rehearsal unit 43, a risk assessment unit 44, and a solution generation unit 45.

[0120] The multi-resolution model unit 41 is used to establish a digital twin system composed of a physical model, a behavior model, and a scenario model. The physical model constructs a fine structure model based on the finite element method; the behavior model describes the system dynamics characteristics based on neural differential equations; the scenario model considers environmental factors such as temperature and wind load to construct a multi-physical field coupling model. Preferably, the multi-resolution model unit 41 adopts an adaptive grid refinement technique, using finer grids in key areas to improve the calculation efficiency while ensuring the simulation accuracy.

[0121] The state synchronization unit 42 is used to map the monitoring data to the digital twin model in real time. This unit fuses the real-time monitoring data with the digital twin model through data assimilation technology to achieve state synchronization between the physical space and the digital space. Preferably, the state synchronization unit 42 adopts an ensemble Kalman filter algorithm to effectively handle non-linear and non-Gaussian characteristics and improve the accuracy of state estimation.

[0122] The scenario rehearsal unit 43 is used to quickly simulate the structural responses under various lifting strategies. This unit presets multiple operation scenarios, such as adjusting the lifting speed and changing the lifting order, and quickly evaluates the effects of different strategies through the digital twin model. Preferably, the scenario rehearsal unit 43 adopts a fast calculation strategy to complete the simulation analysis of a single scenario in milliseconds to support real-time decision-making requirements.

[0123] The risk assessment unit 44 is used to calculate the risk index and reliability of different decision-making solutions. This unit quantifies the risk based on the state-action-value (SAV) matrix, comprehensively considering factors such as structural safety and control costs. The SAV matrix includes a state vector, an action vector, a value matrix, and constraint conditions, and realizes dynamic parameter update through a hierarchical Bayesian method. Preferably, the risk assessment is based on the following formula:

[0124] ,

[0125] where: is the risk index of solution i; is the occurrence probability of the event; is the event under the plan the degree of influence; is the plan against the event the resistance ability; is the total number of risk events considered.

[0126] The plan generation unit 45 is used to recommend the optimal control strategy based on the multi-objective optimization algorithm. This unit comprehensively considers multiple objectives such as safety, economy, and reliability, and generates an optimal solution that balances the requirements of all aspects. Preferably, the plan generation unit 45 adopts the Pareto optimization method, gives multiple non-inferior solutions for decision-making reference, and automatically selects the most suitable control strategy according to the current working conditions.

[0127] The digital twin decision-making module 4 provides three-level decision-making support: full-automatic mode, the system directly executes the optimal decision; semi-automatic mode, the system provides recommendations and the operator confirms; expert-assisted mode, the system makes decisions in collaboration with experts, complementing and enhancing each other. This flexible decision-making mechanism not only ensures the autonomy of the system but also retains the possibility of manual intervention, adapting to decision-making scenarios of different complexities.

[0128] As Figure 6 shown, the adaptive active control module 5 includes an active vibration suppression unit 51, a precise tension regulation unit 52, a stiffness dynamic adjustment unit 53, a hybrid control strategy unit 54, and a multi-loop feedback unit 55.

[0129] The active vibration suppression unit 51 is used to suppress the vibration of the steel grid structure through an electromagnetic-hydraulic composite damper. This unit is deployed at key positions of the structure and effectively suppresses the structure vibration by applying a damping force opposite to the vibration velocity. Preferably, the active vibration suppression unit 51 adopts an electromagnetic-hydraulic composite damper, with a maximum damping coefficient of 25 kN·s / m, a response time less than 20 ms, a working temperature range of -30°C to 70°C, a control accuracy of ±2%, and a power consumption of <15 W (standby) / <120 W (maximum load).

[0130] The precise tension regulation unit 52 is used to balance the tension distribution during the lifting of the steel grid through a distributed intelligent lifter. This unit coordinates the working states of the multi-point lifting equipment according to the real-time monitored structure state to ensure uniform stress on the structure during the lifting process. Preferably, the lifting control parameters of the precise tension regulation unit 52 include: a speed range of 0.01 - 0.5 m / min, an acceleration limit less than 0.01 m / s², a synchronization accuracy less than 1 mm, a load sensing accuracy of 0.1% F.S, and an emergency braking time less than 0.5 s.

[0131] The stiffness dynamic adjustment unit 53 is used to adjust the local stiffness of the structure through variable stiffness connection nodes. This unit optimizes the dynamic characteristics of the structure and improves its vibration resistance performance by changing the node connection stiffness. Preferably, the stiffness dynamic adjustment unit 53 adopts magnetorheological material technology, with a stiffness adjustment range of 200% - 500% and a response time of less than 50 ms, capable of quickly changing the node stiffness within milliseconds to meet the sudden vibration requirements.

[0132] The hybrid control strategy unit 54 is used to switch among model predictive control, H∞ robust control, and sliding mode control strategies according to the structural state. Under normal operating conditions, model predictive control is adopted, with a prediction time domain of 15 - 30 seconds; when facing disturbances, it switches to H∞ robust control to suppress the influence of uncertainties; in critical states, sliding mode control based on Lyapunov stability analysis is applied to ensure system stability. In addition, the hybrid control strategy unit 54 is equipped with an energy consumption adaptive optimization algorithm, controlling the ratio of energy consumption to vibration energy to be less than 0.35 under normal conditions and less than 0.6 under extreme conditions, achieving a balance between control effect and energy consumption.

[0133] The multi-loop feedback unit 55 is used to implement triple-loop nested feedback control of the actuator state, local response, and overall performance.

[0134] Fast local sensors: High-speed sensors with a sampling rate of 200 Hz (5 ms) are equipped at key nodes;

[0135] Conventional global sensors: Most nodes maintain a sampling rate of 100 Hz (10 ms);

[0136] State inference: The state at unsampled time points is obtained through model inference;

[0137] Multi-loop feedback coordination mechanism:

[0138] The inner loop (2 ms) mainly performs real-time control based on local state estimation and fast local sensor data. The middle loop (20 ms) and the outer loop (200 ms) rely on more comprehensive but lower-frequency monitoring data for overall structural optimization control, forming a hierarchical and complementary control strategy.

[0139] This multi-loop nested feedback mechanism ensures the stability and reliability of control, can quickly respond to changes at the actuator level, and at the same time takes into account the optimization of the overall structural performance.

[0140] High-frequency state estimation method: The system uses a state estimator to generate high-frequency state estimates between sensor sampling points to achieve high-frequency feedback:

[0141] ,

[0142] where, is the state estimation vector, For the control input, For the measurement output, is the system state space matrix, is the Kalman filter gain matrix, is the state update time step (2 ms),

[0143] The adaptive active control module 5, through the dynamic energy consumption distribution algorithm, adaptively distributes control energy according to the structural importance, response amplitude, and control effect, achieving the optimal balance between control effect and energy consumption. The entire control system forms a closed-loop control, continuously monitors the structural state through sensors, evaluates the control effect, and dynamically adjusts the control strategy to ensure the effectiveness and adaptability of the control.

[0144] As Figure 7 shown, the present invention also provides an intelligent optimization method for the vibration control of a steel grid structure, which is applied to the above system and includes the following steps:

[0145] Step 1, collect the displacement, acceleration, and strain data of the steel grid structure through the multi-physical quantity fusion monitoring module. Specifically, it includes: collecting displacement data through type I node units with an accuracy of ±0.05 mm and a sampling rate of 100 Hz; collecting acceleration data through type II node units with a range of ±16 g and a sampling rate of 1000 Hz; collecting strain data through type III node units with a resolution of 0.1 με and a sampling rate of 500 Hz. The sensor layout follows the density formula to ensure the monitoring coverage of key parts. At different lifting stages, the sampling strategy is adaptively adjusted to optimize resource utilization.

[0146] Step 2, transmit the data to the edge intelligent data processing module for signal filtering, feature extraction, and anomaly recognition to generate structured feature data. Specifically, it includes: performing preliminary screening and compression of the data in the in-sensor embedded processing layer to reduce the transmission burden; performing wavelet denoising and data fusion in the regional edge node layer to improve the signal quality; completing feature extraction and pattern recognition in the main edge server layer to generate structured feature data; implementing a dual-threshold detection strategy through the anomaly recognition unit to detect anomalies in a timely manner.

[0147] Step 3, input the structured feature data into the dynamic response prediction module to generate the structural dynamic response prediction results at three scales: micro, meso, and macro based on the deep learning model. Specifically, it includes: the micro prediction unit predicts the short-term response changes of key nodes within 5 to 30 seconds based on the spatio-temporal graph convolutional network; the meso prediction unit predicts the evolution of the structural dynamic characteristics within 1 to 10 minutes based on the bidirectional gated recurrent unit network; the macro prediction unit predicts the long-term trend and risks within 10 to 60 minutes based on the attention-enhanced Transformer-XL model; continuously optimize the prediction model through the incremental training unit to improve the prediction accuracy.

[0148] Step 4: Input the prediction results into the digital twin decision-making module, and conduct scenario rehearsal analysis in combination with the multi-resolution digital twin model to generate the optimal control strategy. Specifically, it includes: establishing a digital twin system composed of a physical model, a behavior model, and a scenario model through the multi-resolution model unit; mapping the monitoring data to the digital twin model in real time through the state synchronization unit; quickly simulating the structural responses under various lifting strategies through the scenario rehearsal unit; calculating the risk index and reliability of different decision-making schemes through the risk assessment unit; and recommending the optimal control strategy based on the multi-objective optimization algorithm through the scheme generation unit.

[0149] Step 5: Receive the optimal control strategy through the adaptive active control module, and perform active vibration suppression, precise tension regulation, and dynamic stiffness adjustment to achieve vibration suppression and deformation control of the steel grid structure. Specifically, it includes: the active vibration suppression unit suppressing the vibration of the steel grid structure through an electromagnetic-hydraulic composite damper; the precise tension regulation unit balancing the tension distribution during the lifting process of the steel grid through a distributed intelligent lifter; the dynamic stiffness adjustment unit adjusting the local stiffness of the structure through variable stiffness connection nodes; the hybrid control strategy unit switching different control strategies according to the structural state; and the multi-loop feedback unit implementing triple-loop nested feedback control of the actuator state, local response, and overall performance.

[0150] The system of the present invention is applied to the steel grid lifting project of a stadium. The steel grid structure has a span of 120 m, a weight of about 2,500 tons, and a lifting height of 35 m. During the lifting process, 120 multi-physical quantity monitoring nodes are deployed in the system, a digital twin model including the structural dynamic characteristics, environmental factors, and the working conditions of the lifting equipment is established, and the structural vibration is predicted and controlled in real time.

[0151] During the lifting process, the system successfully predicted and suppressed the risks of two structural resonances caused by wind loads. The vibration amplitude was reduced by 78% through the active vibration suppression unit, and the structure was kept stable; the precise tension regulation unit achieved millimeter-level synchronous accuracy for multi-point lifting, significantly superior to the centimeter-level accuracy of traditional manual control; the dynamic stiffness adjustment unit optimized the structural stiffness distribution at key nodes and improved the overall anti-vibration performance.

[0152] Compared with the traditional method, the system of the present invention shortens the lifting construction period by 25%, reduces the number of manual interventions by 85%, reduces the safety risk by more than 90%, saves 30% of energy consumption, and has remarkable comprehensive benefits.

[0153] The intelligent optimization system and method for vibration control of steel grid structure provided by the present invention realize the comprehensive monitoring, accurate prediction and active control of the vibration of the steel grid structure through the collaborative work of five major modules: multi-physical quantity fusion monitoring, edge intelligent data processing, dynamic response prediction, digital twin decision-making and adaptive active control, significantly improving the safety, reliability and efficiency of the lifting and installation process of the steel grid structure. This system expands from static monitoring to dynamic response analysis, from passive detection to active control, from a single space to multi-scale analysis, from deterministic judgment to uncertainty quantification, and from passive response to intelligent learning, representing the technical development direction of the steel grid structure installation project and having broad application prospects.

[0154] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An intelligent optimization system for vibration control of steel space frame structures, characterized in that, Including: A multi-physical quantity fusion monitoring module, which is used to collect displacement, acceleration and strain data of the steel grid structure and transmit the data to the data processing module; An edge intelligent data processing module, which is communicatively connected to the multi-physical quantity fusion monitoring module, and is used to receive the data collected by the multi-physical quantity fusion monitoring module, perform signal filtering, feature extraction and anomaly recognition, and generate structured feature data; A dynamic response prediction module, which is communicatively connected to the edge intelligent data processing module, and is used to receive the structured feature data and generate structural dynamic response prediction results at three scales of micro, meso and macro based on a deep learning model; A digital twin decision-making module, which is communicatively connected to the dynamic response prediction module, and is used to receive the prediction results, conduct scenario rehearsal analysis in combination with a multi-resolution digital twin model, and generate an optimal control strategy; An adaptive active control module, which is communicatively connected to the digital twin decision-making module, and is used to receive the optimal control strategy and perform control operations through an active vibration suppression system, a precise tension regulation system and a stiffness dynamic regulation system to achieve vibration suppression and deformation control of the steel grid structure; The dynamic response prediction module includes: A micro prediction unit, which is used to predict short-term response changes of key nodes based on a spatio-temporal graph convolutional network; A meso prediction unit, which is used to predict the evolution of the mid-term dynamic characteristics of the structure based on a bidirectional gated recurrent unit network; A macro prediction unit, which is used to predict the improvement of the overall long-term trend and risks based on an attention-enhanced Transformer-XL model; An incremental training unit, which is used to update the model by combining knowledge distillation and federated learning; The digital twin decision-making module includes: A multi-resolution model unit, which is used to establish a digital twin system composed of a physical model, a behavior model and a scenario model; A state synchronization unit, which is used to map the monitoring data to the digital twin model in real time; A scenario rehearsal unit, which is used to quickly simulate the structural responses under various lifting strategies; A risk assessment unit, which is used to calculate the risk index and reliability of different decision-making schemes; A scheme generation unit, which is used to recommend an optimal control strategy based on a multi-objective optimization algorithm.

2. The system according to claim 1, wherein The multi-physical quantity fusion monitoring module includes: Type I node unit, which is used to collect displacement data of the steel grid structure; Type II node unit, which is used to collect acceleration data of the steel grid structure; Type III node unit, which is used to collect strain data of the steel grid structure; A node layout unit, which is used to determine the optimal layout position of sensors according to the improved K-means clustering algorithm; An adaptive sampling unit, which is used to dynamically adjust the sampling frequency and accuracy according to the lifting stage of the steel grid.

3. The system according to claim 1, wherein The edge intelligent data processing module includes: A sensor embedded processing layer, which is used to perform initial data screening and compression; A regional edge node layer, which is used to perform wavelet denoising and data fusion; A main edge server layer, which is used to complete feature extraction and pattern recognition; An anomaly recognition unit, which is used to implement a double-threshold anomaly detection strategy and dynamically update the threshold according to the lifting speed and acceleration factor.

4. The system according to claim 1, wherein The risk assessment unit performs risk quantification based on a state-action-value matrix, wherein the matrix includes a state vector, an action vector, a value matrix and constraint conditions, and realizes dynamic parameter updating through a hierarchical Bayesian method.

5. The system according to claim 1, characterized in that The adaptive active control module includes: Active vibration suppression unit, used to suppress the vibration of the steel grid structure through electromagnetic-hydraulic composite damper; Precise tension control unit, used to balance the tension distribution during steel grid lifting through distributed intelligent lifters; Stiffness dynamic adjustment unit, used to adjust the local stiffness of the structure through variable stiffness connection nodes; A hybrid control strategy unit, used to switch between model predictive control, H∞ robust control and sliding mode control strategies according to the structural state; The multi-loop feedback unit is used to implement three-loop nested feedback control of actuator state, local response and overall performance.

6. The system according to claim 5, wherein The hybrid control strategy unit is provided with an energy consumption adaptive optimization algorithm, and the ratio of control energy consumption to vibration energy is less than 0.35 under normal conditions and less than 0.6 under extreme conditions.

7. An intelligent optimization method for vibration control of a steel space frame structure, applied to the system according to any one of claims 1-6, characterized in that, The following steps are involved: The displacement, acceleration and strain data of the steel grid structure are collected through the multi-physical quantity fusion monitoring module; Transmitting the data to an edge intelligent data processing module for signal filtering, feature extraction, and anomaly identification to generate structured feature data; Inputting the structural feature data into a dynamic response prediction module, and generating structural dynamic response prediction results at three scales, micro, meso and macro, based on a deep learning model; The prediction results are input into the digital twin decision module, and the scenario preview analysis is performed in combination with the multi-resolution digital twin model to generate the optimal control strategy; The optimal control strategy is received through the adaptive active control module to perform active vibration suppression, precise tension control and dynamic stiffness adjustment to achieve vibration suppression and deformation control of the steel grid structure.

Images