Method and system for evaluating performance of fabricated steel beam-column joint

Abstract

This invention discloses a method and system for evaluating the performance of beam-column joints in prefabricated steel structures, relating to the field of prefabricated building construction technology. The method involves determining the evaluation dimensions of beam-column joints in prefabricated steel structures and generating an evaluation index system. It collects experimental data, acceptance measurement data, and operation and maintenance monitoring data for each evaluation index. The experimental data, acceptance measurement data, and operation and maintenance monitoring data are preprocessed to obtain a standardized index dataset. The individual performance scores of each evaluation index are output through a joint performance evaluation model. The comprehensive weight of each evaluation index is determined using the analytic hierarchy process combined with the entropy weight method. Finally, the comprehensive performance evaluation value of the joint is obtained by combining the evaluation results of each evaluation index with the comprehensive weight. This invention overcomes the technical shortcomings of existing evaluation methods, such as single-dimensionality, insufficient accuracy, lack of dynamic evaluation capability, and cumbersome processes, achieving a comprehensive performance evaluation of beam-column joints in prefabricated steel structures throughout their entire lifecycle, in multiple dimensions, and with precision and dynamism.

Description

Technical Field

[0001] This invention relates to the field of prefabricated building construction technology, and more specifically to a method and system for evaluating the performance of beam-column joints in prefabricated steel structures. Background Technology

[0002] Currently, prefabricated steel structures, with their advantages of high construction efficiency, environmental friendliness and energy conservation, material recyclability, and high degree of assembly, have been widely used in high-rise, super high-rise, and high-seismic-intensity buildings, becoming one of the core components of prefabricated building systems. Beam-column joints, as the key force-bearing and force-transmitting core of prefabricated steel structures, directly bear the axial force, bending moment, shear force, and other internal forces transmitted by beams and columns. Their performance directly determines the stability, safety, and durability of the entire steel structure system. They are also a core link in controlling the structure's seismic performance, assembly quality, and operational reliability, playing a decisive role in ensuring the safety of the building throughout its entire life cycle.

[0003] With the continuous development of prefabricated steel structure technology, the structural forms of beam-column joints have become increasingly diverse, forming various types such as reinforced, weakened, and recoverable joints. Joints with different structural forms exhibit significant differences in mechanical performance, seismic resistance, assembly difficulty, and operation and maintenance requirements. Existing evaluation methods suffer from several technical shortcomings: First, they are limited in scope, neglecting the performance requirements at different stages throughout the joint's lifecycle, such as assembly quality during construction and durability and operation and maintenance performance during use. This leads to biased evaluation results that are difficult to support engineering design optimization, construction quality control, and operation and maintenance decisions. Second, their evaluation accuracy is insufficient. Traditional evaluations often rely on manual experience threshold judgments or single experimental data, lacking multi-source data fusion analysis and a scientific weighting mechanism. This makes it impossible to quantify the impact of different indicators on the overall performance of the joint, resulting in strong subjectivity. Third, they lack dynamic evaluation capabilities. Existing methods are mostly static evaluations, conducting only a one-time evaluation of a specific stage of the joint. They cannot capture the performance degradation process of the joint under long-term use, environmental erosion, and load conditions in real time, making it difficult to achieve dynamic tracking and prediction of joint performance.

[0004] Therefore, in view of the shortcomings of existing technologies, how to provide a performance evaluation method and system for prefabricated steel structure beam-column joints, and realize a comprehensive performance evaluation of prefabricated steel structure beam-column joints throughout their entire life cycle, in multiple dimensions, with precision and dynamism, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for evaluating the performance of beam-column joints in prefabricated steel structures, which solves the technical defects of existing evaluation methods such as single dimension, insufficient accuracy, lack of dynamic evaluation capability, and cumbersome process, and realizes a comprehensive performance evaluation of beam-column joints in prefabricated steel structures throughout their entire life cycle, in multiple dimensions, with precision and dynamism.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for evaluating the performance of prefabricated steel structure beam-column joints, comprising: Determine the evaluation dimensions for beam-column joints in prefabricated steel structures, and generate an evaluation index system based on the evaluation dimensions; Collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator; The test data, the acceptance test data, and the operation and maintenance monitoring data are preprocessed to obtain a standardized index dataset; Based on the standardized index dataset, and combined with the improved machine learning algorithm, a node performance evaluation model for prefabricated steel structure beams and columns is constructed, and the individual performance score of each evaluation index is output through the node performance evaluation model. The combined weights of each evaluation index are determined by using the analytic hierarchy process (AHP) combined with the entropy weight method. By combining the evaluation results of each evaluation indicator with their comprehensive weights, a comprehensive evaluation value for node performance is obtained.

[0007] Preferably, the evaluation dimensions include mechanical performance, seismic performance, assembly quality, durability, and operation and maintenance performance. Each evaluation dimension has several primary evaluation indicators, and each primary evaluation indicator has several secondary evaluation indicators, forming a hierarchical and multi-dimensional evaluation indicator system.

[0008] Preferably, the improved machine learning algorithm integrates an adaptive dynamic attention residual convolutional network and an adaptive particle swarm optimization algorithm, with the goal of minimizing node performance evaluation error, to perform adaptive iteration of model parameters and accurate performance prediction.

[0009] Preferably, the adaptive dynamic attention residual convolutional network takes a standardized index dataset as input to construct a node performance monitoring graph structure; The node performance monitoring graph structure uses sensor measurement points and component connection points as graph nodes. The node features correspond to the data of each evaluation index, and the node connection relationship is determined by the dual constraints of spatial distance and dynamic performance difference. The local and extended neighborhood performance features are extracted by dual-scale graph convolution, and feature weights are dynamically allocated by graph attention mechanism. Temporal encoding is embedded to fuse temporal evolution information, and a dynamic residual fusion gating mechanism is used to suppress redundant features, outputting node performance feature vectors. The node performance feature vector is mapped through a fully connected layer to obtain the performance prediction value.

[0010] Preferably, the adaptive particle swarm optimization algorithm uses the mean square error between the performance prediction value output by the adaptive dynamic attention residual convolutional network and the true label as the objective function to initialize the parameter combination of the particle swarm representation network; it adaptively adjusts the inertia weight coefficient based on the node performance evolution rate, dynamically optimizes the particle search speed and position, avoids premature convergence of the algorithm, outputs the optimal network parameter combination, and improves the fitting accuracy and generalization ability of the evaluation model.

[0011] Preferably, the weights of the indicators are determined using the analytic hierarchy process (AHP) combined with the entropy weight method, including: A judgment matrix is ​​constructed using the analytic hierarchy process (AHP) to obtain the subjective weights of each dimension and indicator. The objective weights of each indicator are obtained by calculating the dispersion of the data of each indicator using the entropy weight method. The subjective weights and objective weights are linearly fused to obtain a comprehensive weight.

[0012] Preferably, a performance evaluation system for prefabricated steel structure beam-column joints includes: The evaluation index system generation module is used to determine the evaluation dimensions of the beam-column joints of the prefabricated steel structure and generate an evaluation index system based on the evaluation dimensions. The data acquisition module is used to collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator. The standardization module is used to preprocess the test data, the acceptance test data, and the operation and maintenance monitoring data to obtain a standardized index dataset. The model building module is used to construct a node performance evaluation model for prefabricated steel structure beams and columns based on the standardized index dataset and combined with an improved machine learning algorithm, and output the individual performance score value of each evaluation index through the node performance evaluation model. The weight calculation module is used to determine the comprehensive weight of each evaluation indicator by combining the analytic hierarchy process (AHP) with the entropy weight method. The comprehensive evaluation module is used to combine the evaluation results and comprehensive weights of each evaluation indicator to obtain the comprehensive evaluation value of node performance.

[0013] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for evaluating the performance of prefabricated steel structure beam-column joints, which breaks through the technical bottlenecks in the traditional evaluation of prefabricated steel structure beam-column joint performance, such as insufficient feature extraction, parameter dependence on manual work, low evaluation accuracy, and weak generalization ability. It realizes accurate, efficient, and comprehensive evaluation of joint performance. The present invention has the following significant beneficial effects: (1) It constructs a hierarchical evaluation index system covering five dimensions: mechanical performance, seismic performance, assembly quality, durability, and operation and maintenance performance. It comprehensively covers the key performance parameters of the entire life cycle of joint design, construction, and operation and maintenance, and solves the problem that the existing evaluation index is single and one-sided and cannot comprehensively characterize the comprehensive performance of the joint; (2) It adopts a distributed acquisition and edge computing collaborative mode to integrate three types of multi-source heterogeneous data: test data, acceptance measured data, and operation and maintenance monitoring data. Through a multi-step data preprocessing process, it ensures the reliability, consistency, and standardization of the data, providing a basis for subsequent evaluation models. The accuracy of the model provides solid data support, overcoming the defects of poor compatibility of multi-source data, large noise interference, and low data utilization in the existing technology; (3) The adaptive dynamic attention residual convolutional network and the adaptive particle swarm optimization algorithm are integrated to form an intelligent evaluation architecture. The adaptive dynamic attention residual convolutional network can accurately capture the spatial correlation and temporal evolution law of node performance through dynamic graph structure, dual-scale feature extraction, attention mechanism and temporal embedding, and output a comprehensive node performance feature vector. The reliable performance prediction value is obtained by mapping through the fully connected layer. The adaptive particle swarm optimization algorithm takes the mean square error between the prediction value and the real label as the objective function, and realizes the global optimal search of network parameters through adaptive inertial weight adjustment. It avoids the subjectivity of manual parameter tuning and the problem of premature convergence of the algorithm, significantly improves the fitting accuracy and generalization ability of the evaluation model, and solves the technical problem of low evaluation accuracy and poor robustness of traditional machine learning algorithms in small sample, high-dimensional, multi-source heterogeneous data scenarios. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a method for evaluating the performance of prefabricated steel structure beam-column joints provided by the present invention.

[0016] Figure 2 This invention provides a structural schematic diagram of a prefabricated steel structure beam-column joint performance evaluation system. Detailed Implementation

[0017] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention discloses a method for evaluating the performance of beam-column joints in prefabricated steel structures, such as... Figure 1 As shown, it includes: Determine the evaluation dimensions for beam-column joints in prefabricated steel structures, and generate an evaluation index system based on the evaluation dimensions; Collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator; The test data, the acceptance test data, and the operation and maintenance monitoring data are preprocessed to obtain a standardized index dataset; Based on the standardized index dataset, and combined with the improved machine learning algorithm, a node performance evaluation model for prefabricated steel structure beams and columns is constructed, and the individual performance score of each evaluation index is output through the node performance evaluation model. The combined weights of each evaluation index are determined by using the analytic hierarchy process (AHP) combined with the entropy weight method. By combining the evaluation results of each evaluation indicator with their comprehensive weights, a comprehensive evaluation value for node performance is obtained.

[0019] Specifically, the test data includes mechanical and seismic index data obtained from laboratory quasi-static tests, dynamic loading tests, tensile tests, fatigue tests, and corrosion tests; The acceptance test data includes quality acceptance data obtained from on-site torque testing, dimensional testing, non-destructive testing, weld testing, and assembly deviation testing. The operation and maintenance monitoring data includes long-term service dynamic data collected by structural health monitoring systems, stress sensors, displacement sensors, corrosion sensors, and vibration sensors. The data acquisition process employs a collaborative approach between distributed acquisition terminals and edge computing nodes to achieve real-time data uploading, categorized storage, and traceability management. It constructs a multi-source heterogeneous node performance database containing text, numerical data, and images, supporting incremental data updates and dynamic data retrieval.

[0020] Specifically, the data collection process adopts a collaborative working mode of distributed data collection terminals and edge computing nodes to achieve full-domain collection, real-time transmission, intelligent classification, and end-to-end traceability of experimental data, acceptance test data, and operation and maintenance monitoring data. It constructs a multi-source heterogeneous node performance database containing text, numerical data, and images. The specific implementation process is as follows: Multiple types of sensors, including those for stress, displacement, pore water pressure, vibration, corrosion, temperature and humidity, and tilt angle, are deployed at the beam-column joints of the prefabricated steel structure. Each sensor corresponds to a distributed acquisition terminal, which has independent acquisition, local caching, time synchronization, and edge preprocessing functions. The acquisition terminals acquire data in real time according to a preset sampling frequency: during the testing phase, high-frequency mode is used to acquire mechanical and seismic response data for quasi-static, dynamic loading, and fatigue tests; during the acceptance phase, measured data such as torque, dimensional deviation, weld images, and assembly gaps are acquired; during the operation and maintenance phase, low-frequency continuous acquisition of dynamic data on structural service is used. All terminals adopt a unified timestamp synchronization mechanism to ensure strict alignment of multi-source data in the time dimension.

[0021] Each data acquisition terminal uploads raw data to the nearest edge computing node deployed at the construction site via wired Ethernet or wireless LoRa / 5G, enabling preliminary processing without relying on the cloud. The edge computing nodes perform three main tasks: first, real-time data verification, instantly marking and alarming for out-of-range, disconnected, or abrupt data changes; second, data classification and structuring, automatically categorizing test data, acceptance data, and maintenance data, uniformly encoding unstructured data such as weld images and inspection reports, and formatting and storing numerical data such as load-bearing capacity, displacement, and corrosion rate; and third, data traceability and binding, assigning a unique identifier ID to each data point and associating it with sensor number, node number, spatial coordinates, acquisition time, acquisition conditions, and operator information to form an immutable traceability chain.

[0022] Edge computing nodes upload cleaned, categorized, and traced data to the central database via a secure, encrypted channel, constructing a multi-source heterogeneous node performance database. The database employs a hybrid storage architecture: a relational database stores standardized values ​​such as load-bearing capacity, displacement, and torque; a document database stores acceptance reports, maintenance logs, and expert judgment texts; and an image database stores non-destructive testing photos of welds, node installation images, and crack monitoring images. The database supports incremental updates, with edge nodes only uploading changed and newly added data, reducing transmission pressure. It also provides standardized calling interfaces, supporting evaluation models, monitoring platforms, and maintenance systems to dynamically access data as needed.

[0023] To ensure data reliability, the system is equipped with a three-level storage mechanism: local caching at the acquisition terminal, temporary storage at edge nodes, and permanent storage in the central database. Failure in any of these links will not affect data integrity, and the system will automatically retransmit the data at the point of interruption after recovery.

[0024] The test data, the acceptance test data, and the operation and maintenance monitoring data are preprocessed to obtain a standardized indicator dataset, including: Data preprocessing includes: S31. Data cleaning: Remove duplicate data, missing data, and logically abnormal data. Use the Laida criterion to identify and remove outliers. Use linear interpolation, mean imputation, and K-nearest neighbor imputation to complete missing data. S32. Data noise reduction: Wavelet transform, moving average filtering, and Kalman filtering are used to remove monitoring noise and interference signals while preserving the true performance characteristics. S33. Data Standardization: Extreme value normalization, Z-score standardization, and interval mapping are used to uniformly map indicator data of different dimensions and magnitudes to the [0,1] interval. The formula is as follows: Positive indicators: ; Negative indicators: ; In the formula: For standardized data, The original data, , Let be the maximum and minimum values ​​of the j-th indicator; S34. Data Consistency Verification: Cronbach's Alpha model is used for credibility verification to ensure data reliability and consistency. The verification formula is as follows: ; In the formula: m is the number of indicators. For the variance of a single indicator, This represents the total variance.

[0025] Specifically, the evaluation dimensions include mechanical performance, seismic performance, assembly quality, durability, and operation and maintenance performance. Each evaluation dimension has several primary evaluation indicators, and each primary evaluation indicator has several secondary evaluation indicators, forming a hierarchical and multi-dimensional evaluation indicator system.

[0026] Among them, the mechanical performance dimension includes: primary indicators such as bearing capacity, stiffness characteristics, connection strength, and deformation compatibility; secondary indicators such as ultimate bearing capacity, yield load, initial stiffness, residual deformation, bolt slippage, weld tensile strength, and end face friction coefficient. Seismic performance dimensions: Primary indicators include energy dissipation capacity, ductility performance, low-cycle fatigue performance, and damage evolution characteristics; Secondary indicators include equivalent viscous damping coefficient, ductility coefficient, fatigue life, damage factor, stiffness degradation rate, and bearing capacity degradation rate. Assembly quality dimensions: Primary indicators include installation accuracy, connection reliability, construction compliance, and component matching; Secondary indicators include axis deviation, perpendicularity deviation, clearance deviation, final bolt tightening torque, welding quality pass rate, component dimensional error, and assembly efficiency. Durability dimension: Primary indicators include corrosion resistance, fatigue resistance, aging resistance, and environmental adaptability; secondary indicators include corrosion rate, fatigue crack propagation rate, coating integrity, low-temperature toughness, high-temperature stability, and dry-wet cycle durability. Operation and maintenance performance dimensions: primary indicators include monitorability, maintainability, replaceability, and security early warning; secondary indicators include sensor adaptability, maintenance convenience, component replacement cycle, anomaly response speed, and early warning accuracy.

[0027] Specifically, the improved machine learning algorithm integrates an adaptive dynamic attention residual convolutional network and an adaptive particle swarm optimization algorithm, with the goal of minimizing node performance evaluation error, to perform adaptive iteration of model parameters and accurate performance prediction.

[0028] The improved machine learning algorithm used in this invention is not a single neural network or a traditional optimization algorithm. Instead, it uses an adaptive dynamic attention residual convolutional network as the core structure for feature extraction and performance prediction, and an adaptive particle swarm optimization algorithm as the global optimization unit for model hyperparameters and network weights, forming a closed-loop intelligent evaluation architecture of feature learning, parameter optimization, error feedback, and iterative updates. The algorithm constructs a loss function based on the mean square error between the predicted and actual calibration values ​​of prefabricated steel structure beam-column nodes, and uses minimizing the evaluation error as the global optimization objective. During training, it adaptively iteratively updates key parameters such as network structure parameters, attention weight coefficients, residual gating thresholds, and convolutional kernel scale. This eliminates the need for manual parameter tuning, allowing it to adapt to performance evaluation scenarios with different node types, service environments, and data distributions. Ultimately, it achieves comprehensive performance prediction with high accuracy, high robustness, and high generalization ability across five dimensions: node mechanics, seismic resistance, assembly, durability, and operation and maintenance.

[0029] Compared with traditional machine learning methods such as graph convolutional networks, backpropagation neural networks, support vector machines, and random forests, the improved algorithm in this embodiment adapts to changes in node monitoring points through a dynamic graph structure, focuses on high-contribution performance features through an attention mechanism, alleviates gradient vanishing in deep networks through residual gating, and avoids local optima through adaptive particle swarm optimization. It maintains stable evaluation accuracy in beam-column node performance data with small samples, high dimensions, multi-source heterogeneity, and time-series evolution, effectively solving the technical defects of traditional algorithms in steel structure node performance evaluation, such as insufficient feature extraction, parameter dependence on manual methods, easy overfitting, and poor generalization.

[0030] Specifically, the adaptive dynamic attention residual convolutional network takes a standardized metric dataset as input to construct a node performance monitoring graph structure; The node performance monitoring graph structure uses sensor measurement points and component connection points as graph nodes. The node features correspond to the data of each evaluation index, and the node connection relationship is determined by the dual constraints of spatial distance and dynamic performance difference. The local and extended neighborhood performance features are extracted by dual-scale graph convolution, and feature weights are dynamically allocated by graph attention mechanism. Temporal encoding is embedded to fuse temporal evolution information, and a dynamic residual fusion gating mechanism is used to suppress redundant features, outputting node performance feature vectors. The node performance feature vector is mapped through a fully connected layer to obtain the performance prediction value; Specifically, the adaptive dynamic attention residual convolutional network uses a preprocessed standardized index dataset as its original input. This standardized index dataset contains 28 secondary evaluation indicators across five dimensions: mechanical performance, seismic performance, assembly quality, durability, and operation and maintenance performance. All index data have been normalized to the [0,1] interval, possessing a unified dimension and numerical range, and can be directly used for training and inference of the network model.

[0031] To fully characterize the spatial topological relationships of beam-column joints in prefabricated steel structures, the force transmission paths of components, the correlation of monitoring points, and the spatial dependence of performance evolution, this embodiment of the invention constructs a node performance monitoring graph structure, denoted as: G=(V, E, A,X), where the parameters are defined as follows: V is the set of graph nodes, V={v1,v2,…,v...} n}, where n is the sum of the total number of sensor measuring points and the number of key connection points of the component. Each node corresponds to a physical monitoring position or a key stress part, specifically including bolt ball joints, end plate connection positions, flange butt joint positions, web splicing positions, stress sensor placement points, displacement monitoring points, corrosion sensor points, etc. E is the set of graph edges, used to characterize the performance correlation and spatial influence relationship between nodes. It is not a fixed connection relationship, but is updated in real time with the dynamic changes in the performance status of the nodes. A is an adaptive weighted adjacency matrix, A∈R n×n The elements of this matrix are calculated using a combination of spatial distance and dynamic performance differences, and are used to quantify the strength of associations between nodes. X is the feature matrix of the graph nodes, X∈R n×d Where d is the number of evaluation indicators contained in a single node, and the feature vector of each node is composed of standardized indicator data such as bearing capacity, stiffness, ductility, damping, installation deviation, corrosion rate, and monitoring response at the corresponding location, which can comprehensively reflect the performance status of the corresponding point.

[0032] Specifically, the node connection relationship is determined by the dual constraints of spatial distance and dynamic performance difference, including: only when two nodes meet the proximity condition in space and their performance indicators show synchronous anomalies or significant correlations are they assigned higher edge weights; for regions with stable performance, the connection weights between nodes are appropriately reduced or even disconnected to avoid invalid information interfering with the model evaluation accuracy and to ensure that the graph structure can accurately depict the real evolution law of node performance.

[0033] Spatial distance calculation uses a three-dimensional spatial distance formula, as follows: ; In the formula, ( ), ( , , ) are the three-dimensional spatial coordinates of nodes i and j, respectively. These coordinates are determined by on-site measurements to ensure the accuracy of the spatial position.

[0034] The dynamic performance difference index is calculated using the L2 norm, and the specific formula is as follows: In the formula, , Let be the feature vectors of node i and node j, respectively. The L2 norm is used to quantify the degree of performance difference between two nodes. This is an indicator of the dynamic performance difference between node i and node j.

[0035] The rules for calculating the elements of the adaptive weighted adjacency matrix A are as follows: ; in, Let be the element in the i-th row and j-th column of the adjacency matrix, representing the connection weight between nodes i and j. This is the spatial distance attenuation coefficient. Both are performance difference attenuation coefficients, set according to the actual size of the node and the monitoring range, and are used to adjust the degree of influence of spatial distance and performance difference on connection weight. This is the spatial distance threshold; As an exponential function, the connection weights decay exponentially and rapidly with increasing spatial distance and performance differences, which conforms to the actual distribution law of the influence range of node performance.

[0036] This dual constraint mechanism enables the graph structure to dynamically focus on areas of abnormal performance and key force transmission paths, closely matching the actual stress characteristics and damage evolution patterns of steel structure beam-column joints, thus ensuring the accuracy of subsequent feature extraction.

[0037] Specifically, the adaptive dynamic attention residual convolutional network adopts a dual-scale parallel extraction structure of local neighborhood convolution and extended neighborhood convolution, which respectively captures the subtle local performance changes of nodes and the overall structural collaborative performance trend. This effectively solves the technical problem that single-scale feature extraction cannot take into account both local anomalies and global trends, and improves the comprehensiveness and accuracy of feature extraction.

[0038] The local neighborhood feature extraction process includes: taking the target node as the center, performing convolution aggregation operation on its first-order directly adjacent nodes to extract the performance features of the target node itself and its directly connected parts. This can effectively reflect subtle performance changes such as bolt tightness, weld quality, and local deformation, providing support for the evaluation of local node performance.

[0039] The formula for local neighborhood feature extraction is as follows: ; in, is the local neighborhood feature matrix of the l-th layer; Ã is the normalized adjacency matrix, which is obtained by normalizing the adaptive weighted adjacency matrix A and is used to eliminate the influence of node degree differences on feature aggregation; This is the output feature matrix of the (l-1)th layer network, used to characterize the node features extracted from the previous layer; σ is the trainable weight matrix of the local convolutional layer, used to perform weighted transformation on the features of neighboring nodes; σ(·) is the activation function, using the ReLU activation function to introduce non-linear features and improve the feature representation ability of the network.

[0040] The extended neighborhood feature extraction process includes: based on the local neighborhood feature extraction, the neighborhood range is extended to second-order adjacent nodes, that is, nodes that have a two-hop connection with the target node. By squaring the normalized adjacency matrix, an indirect information propagation path is constructed, and the overall performance characteristics of the region where the target node is located are extracted. This can effectively reflect the macroscopic performance trends such as stiffness degradation, damage propagation, and cooperative deformation in the beam-column region where the node is located.

[0041] The extended neighborhood feature extraction formula is as follows: ; In the formula, This is the extended neighborhood feature matrix of the l-th layer; The square of the normalized adjacency matrix, To expand the trainable weight matrix of the convolutional layer.

[0042] Specifically, in order to enable the network to automatically focus on features and neighboring nodes that contribute more to the evaluation of node performance and avoid subjective bias caused by manually setting weights, this embodiment of the invention introduces a dynamic graph attention mechanism into the network to adaptively weight and fuse local and extended neighborhood features extracted by dual-scale graph convolution, thereby improving the effectiveness of features.

[0043] First, an attention mapping function is constructed using a multilayer perceptron to calculate the attention coefficients between adjacent nodes, which are used to quantify the importance of feature associations between nodes. The specific formula is as follows: ; In the formula, Let be the attention coefficients between nodes i and j, and || be the feature vector concatenation operation. (·) is the attention mapping function, composed of two layers of perception mechanisms, used to map the concatenated feature vector to attention coefficients. , , These are the feature vectors of nodes i and j in the (l-1)th layer, respectively.

[0044] Secondly, the attention coefficients are normalized using the softmax function to obtain normalized attention weights, ensuring that the sum of the weights is 1. The specific formula is as follows: ; In the formula, These are the normalized attention weights; It is an exponential function; This is the exponential sum of the attention coefficients of all neighboring nodes, used to normalize the attention weights.

[0045] Finally, based on the normalized attention weights, the dual-scale features are weighted and fused to obtain the feature fusion output of the l-th layer, as shown in the following formula: ; In the formula, This is the feature fusion output matrix for the l-th layer; , These are the attention weights for local and extended neighborhood features, respectively, satisfying... It can update in real time according to the node performance status. When a node is damaged, deformed, or corroded, it can automatically increase the weight of the corresponding feature and improve the network’s sensitivity to identifying performance anomalies.

[0046] Specifically, since the performance of prefabricated steel structure beam-column joints undergoes temporal evolution such as degradation, fatigue accumulation, and damage development over service time, this invention introduces a time coding module into the network to enable the network to accurately model the temporal changes in node performance. This module integrates temporal information into graph convolutional features, enhancing the network's ability to model the temporal evolution of node performance. Specifically, this includes: Map the collection timestamp of each node to the corresponding time series vector T. i The dimension of the time series vector is consistent with the dimension of the node feature vector, and the time series vectors of all nodes are summed to form the time series matrix T∈R. n×d The temporal matrix T is fused with the multi-scale graph convolution of the current layer to output the result. By adding the corresponding nodes row by row, the updated graph convolutional feature matrix is ​​obtained, as shown in the following formula: ; in, The feature matrix after embedding time-series information; This is the output matrix of the multi-scale feature fusion at layer l; It is a time series matrix, which is formed by summing the time series vectors of all nodes, and its dimension is the same as that of the node feature vectors.

[0047] This operation enables the network to simultaneously extract spatial features and model temporal evolution, allowing it to not only characterize the current performance status of nodes but also predict future performance trends, thus providing support for early warning of node performance.

[0048] Specifically, to address issues such as gradient vanishing, redundant feature accumulation, and model convergence difficulties that easily occur during deep network training, this invention constructs a dynamic residual fusion gating mechanism. By adaptively controlling the fusion ratio of current layer features and historical layer features through gating coefficients, redundant features are suppressed, thereby improving the training stability and feature representation capability of the network.

[0049] The formula for calculating the gating coefficient is as follows: ; in, The l-th layer gating coefficient has a value range of [0,1]. For the trainable weight matrix of the gated layer, These are the gate layer bias terms, and both are adaptively updated through network training. This is a matrix flattening operation used to flatten the output feature matrix of the (l-1)th layer. Flattened into a one-dimensional vector, it facilitates the calculation of the gating coefficient; This is the Sigmoid activation function, used to adjust the fusion ratio between current layer features and historical layer features.

[0050] Based on the gating coefficients, a residual fusion operation is performed to obtain the convolutional feature matrix of the residual map of the l-th layer. The specific formula is as follows: ; In the formula, The feature matrix of the residual map at layer l is the convolution feature matrix. This is an element-wise multiplication operation. The gating coefficient is set when the node performance is stable. The gating coefficient approaches 0, tending to preserve historical features and ensure feature stability; when node performance undergoes a sudden change, the gating coefficient... Approaching 1, it tends to adopt newly extracted features from the current layer, effectively improving the network's ability to capture performance mutations, thereby improving the network's stability and evaluation accuracy.

[0051] After operations such as multi-layer dual-scale graph convolution, dynamic graph attention weight allocation, temporal information embedding, and dynamic residual fusion gating, the adaptive dynamic attention residual convolutional network outputs the final feature matrix. ∈R n×m Where m is the final feature dimension, adaptively determined by the network structure. Each row in this feature matrix corresponds to a node performance feature vector of a graph node. This vector integrates the node's spatial topology information, multi-dimensional performance index information, temporal evolution information, and attention weight information, and can comprehensively and accurately characterize the overall performance status of the corresponding monitoring point or connection part, providing core data support for the subsequent individual performance scoring of each evaluation index and the comprehensive evaluation of node performance.

[0052] Specifically, the adaptive particle swarm optimization algorithm uses the mean square error between the performance prediction value output by the adaptive dynamic attention residual convolutional network and the true label as the objective function to initialize the parameter combination of the particle swarm representation network; it adaptively adjusts the inertia weight coefficient based on the node performance evolution rate, dynamically optimizes the particle search speed and position, avoids premature convergence of the algorithm, outputs the optimal network parameter combination, and improves the fitting accuracy and generalization ability of the evaluation model.

[0053] Specifically, the goal of the adaptive particle swarm optimization algorithm is to minimize the node performance evaluation error of the adaptive dynamic attention residual convolutional network, so that the performance prediction value output by the network is as close as possible to the actual performance state of the node, thus ensuring the accuracy and reliability of the evaluation model.

[0054] In this embodiment of the invention, the mean squared error (MSE) is used as the objective function of the adaptive particle swarm optimization algorithm, and the specific formula is as follows: ; in, The objective function is to evaluate the error. This is the set of parameters to be optimized for an adaptive dynamic attention residual convolutional network, specifically including all network parameters that need to be determined through training, such as convolutional kernel weights, attention coefficients, residual gating thresholds, learning rate, and batch size. The number of samples is the total number of samples in the standardized indicator dataset. The samples are all from preprocessed experimental data, acceptance test data, and operation and maintenance monitoring data. The true label for the i-th sample is determined jointly by laboratory testing, on-site acceptance testing, and industry experts to ensure the accuracy and authority of the true label. The performance prediction value of the i-th sample output by the adaptive dynamic attention residual convolutional network is compared with the true label y. i One-to-one correspondence.

[0055] The adaptive particle swarm optimization algorithm minimizes the objective function. With the goal of global optimization, the optimal combination of network parameters Θ is obtained through iterative search. This minimizes the error in node performance evaluation and ensures the prediction accuracy of the evaluation model.

[0056] All parameters to be optimized in the adaptive dynamic attention residual convolutional network are encoded to form a particle. Each particle corresponds to a complete set of network parameter configurations. Multiple particles form a particle swarm to achieve global optimization of network parameters.

[0057] The initialization expression for the particle swarm is as follows: P = {Θ1,Θ2,…,Θ} s In the formula, s is the number of particles, which is set according to the network complexity and optimization efficiency. It is usually taken as 20-50 and can be flexibly adjusted according to the actual application scenario.

[0058] Each particle contains the following four core parameters: Current position Θ s : Corresponds to a complete set of network parameter configurations, with each parameter within a reasonable range of values; Current speed V s Used to determine the position update direction and step size of the particle in the next iteration, which directly affects the optimization efficiency and optimization accuracy; Individual optimal position P best The combination of parameters that minimizes the objective function f(Θ) during all iterations; Global optimal position G best The combination of parameters that minimizes the objective function f(Θ) during all iterations of the entire particle swarm.

[0059] During initialization, the current position of the particle is randomly generated within a reasonable range of values ​​for each parameter. The current velocity of the particle is initialized to 0 or a small range of random values. The individual optimal position is initialized to the initial position of the particle, and the global optimal position is initialized to the position that minimizes the objective function value among all the initial positions of the particles. This ensures the rationality and randomness of the particle swarm initialization and lays the foundation for subsequent optimization.

[0060] Specifically, the process of adaptive particle velocity and position update is as follows: Traditional particle swarm optimization (PSO) algorithms employ fixed inertia weights, which are prone to premature convergence, getting stuck in local optima, and sluggish searching in later stages, failing to adapt to the dynamic characteristics of node performance evolution. This invention adopts an adaptive inertia weight adjustment strategy based on the node performance evolution rate, enhancing the algorithm's global search capability when node performance undergoes abrupt changes and strengthening its local fine-tuning capability when node performance stabilizes, achieving an adaptive balance between global search and local optimization.

[0061] The particle velocity update formula is as follows: ; in, , These are the velocity vectors of the s-th particle in the t-th and t+1-th iterations, respectively. For the t-th iteration, the adaptive inertia weight is dynamically adjusted according to the node performance evolution rate, and is used to balance the global search and local optimization capabilities of particles. , The learning factor is used to adjust the degree of influence of the individual optimal position and the global optimal position on the particle velocity update. It is usually set to c1=c2=2, and can be fine-tuned according to the actual optimization effect. , The [0,1] random weighting coefficients are used to increase the randomness of particle search and avoid the algorithm getting trapped in local optima. The optimal position for each individual particle; This represents the globally optimal position for the particle swarm. This represents the position of the s-th particle in the t-th iteration (network parameter combination).

[0062] The particle position update formula is as follows: ; In the formula, This is the position vector of the s-th particle in the (t+1)-th iteration, i.e., the combination of network parameters; Specifically, the adaptive inertia weight calculation process is as follows: The adaptive inertia weight The objective function is dynamically adjusted based on its rate of decline (i.e., the rate of node performance evolution). The rate of decline of the objective function directly reflects the drastic nature of the node performance evolution. The specific calculation formula is as follows: ; in, Let be the inertia weight for the t-th iteration; , These represent the maximum and minimum values ​​of the inertia weight, respectively. The global optimal objective function value in generation t is the minimum evaluation error achieved by the entire particle swarm in generation t. Let K be the global optimal objective function value for the tKth generation, where K is a sliding window used to calculate the rate of descent of the objective function, typically taken as 5-10. It is a very small positive number, and its value is usually 10. -8 This is used to avoid cases where the denominator is 0, ensuring the validity of the formula calculation.

[0063] When node performance evolves drastically (i.e., the objective function decreases rapidly), As the value increases, it approaches... This enhances the global search capability of particles, facilitating the rapid finding of the globally optimal parameter combination; when node performance tends to stabilize (i.e., the objective function decreases at a slower rate), The value decreases, approaching... This enhances the particle's ability to perform local fine-tuning, fine-tunes the optimal parameter combination, further reduces evaluation errors, and achieves an adaptive balance between global and local optimization.

[0064] The adaptive particle swarm optimization algorithm stops iterating and outputs the globally optimal combination of network parameters when any of the following termination conditions are met during the iteration process: The maximum number of iterations is set based on network complexity and optimization efficiency, typically ranging from 100 to 300, and can be flexibly adjusted according to the actual application scenario. The objective function changes less than a set error threshold for several consecutive generations (usually 5–10 generations), with the error threshold typically set to 10. -6 This indicates that the algorithm has converged, and further iterations cannot significantly reduce the evaluation error; The node performance evaluation accuracy meets the preset engineering requirements, that is, the evaluation error is less than the allowable error range of the engineering, which can directly meet the actual engineering application needs.

[0065] After the iteration terminates, output the globally optimal parameter combination Θ. The optimal parameter combination is then substituted into an adaptive dynamic attention residual convolutional network to obtain the optimal node performance evaluation model. This model has the best fitting accuracy and generalization ability and can be used for accurate performance evaluation of beam-column nodes in prefabricated steel structures.

[0066] After adaptive particle swarm optimization, the parameter configuration of the adaptive dynamic attention residual convolutional network reaches the global optimum, which can significantly reduce the node performance evaluation error and effectively improve the fitting accuracy and generalization ability of the evaluation model under different node types, service environments, and data distributions. Simultaneously, it avoids overfitting and underfitting problems in the evaluation model, enabling the model to adapt to the performance evaluation needs of prefabricated steel structure beam-column nodes in different scenarios. This ensures the reliability, stability, and engineering practicality of the evaluation results, providing accurate data support and technical assurance for the design optimization, construction acceptance, and operation and maintenance management of prefabricated steel structure beam-column nodes.

[0067] In this embodiment of the invention, an adaptive dynamic attention residual convolutional network is used to extract multi-dimensional spatiotemporal features, outputting node performance feature vectors and mapping them to performance prediction values. An objective function is constructed using the error between the predicted values ​​and the true labels, and an adaptive particle swarm optimization algorithm is used to globally optimize the key parameters of the network. The optimal parameters obtained by optimization are substituted back into the network to complete model calibration, improving prediction accuracy and generalization ability. The above process is iteratively executed until the error converges, and finally, a stable and reliable node performance evaluation result is output.

[0068] Specifically, the weights of the indicators are determined using the analytic hierarchy process (AHP) combined with the entropy weight method, including: A judgment matrix is ​​constructed using the analytic hierarchy process (AHP) to obtain the subjective weights of each dimension and indicator. The objective weights of each indicator are obtained by calculating the dispersion of the data of each indicator using the entropy weight method. The subjective weights and objective weights are linearly fused to obtain a comprehensive weight, which takes into account both expert experience and objective data patterns, thereby improving the rationality of weight allocation.

[0069] The embodiments of the present invention use a combination of the analytic hierarchy process (AHP) and the entropy weight method to determine the comprehensive weight, taking into account both the subjectivity of expert experience and the objectivity of data patterns, avoiding the one-sidedness of a single weight allocation method, and improving the rationality and scientificity of weight allocation. Specifically, the comprehensive performance evaluation value of the node is calculated using a weighted summation method, and the specific formula is as follows: ; In the formula, S is the comprehensive performance evaluation value of the node. The comprehensive weight of the i-th evaluation indicator is... Let be the individual performance score of the i-th evaluation indicator, and n be the total number of evaluation indicators. The individual performance score is output by the node performance evaluation model based on standardized indicator data, and the score range is [0, 100]. The higher the score, the better the performance of the corresponding evaluation indicator.

[0070] Specifically, based on the comprehensive performance evaluation value of nodes, node performance levels are divided into four levels: excellent (S≥90), good (80≤S<90), qualified (70≤S<80), and unqualified (S<70). Differentiated operation and maintenance management strategies are formulated for different performance levels. Excellent level nodes only require routine inspections, good level nodes require regular monitoring of key indicators, qualified level nodes require targeted rectification of weak indicators, and unqualified level nodes are immediately taken out of service and repaired or replaced, so as to achieve precise management and control of the entire life cycle of nodes.

[0071] This invention classifies performance into four levels based on comprehensive evaluation values ​​and formulates differentiated operation and maintenance management strategies to achieve precise management and control of nodes throughout their entire lifecycle. It provides accurate data support and technical assurance for the design optimization, construction acceptance, operation and maintenance of prefabricated steel structure beam-column nodes, filling the gap in the existing technology for a closed-loop management and control scheme for the entire lifecycle of node performance.

[0072] The evaluation method and system of this invention have standardized operation procedures, high evaluation accuracy, strong adaptability, and outstanding practicality. They can complete the automated and intelligent evaluation of node performance without extensive manual intervention, effectively solving many defects of existing prefabricated steel structure beam-column node performance evaluation methods. They can be widely applied to various prefabricated steel structure beam-column node performance evaluation scenarios, promoting the high-quality development of the prefabricated steel structure industry.

[0073] In one specific embodiment of the present invention, a performance evaluation system for prefabricated steel structure beam-column joints is provided, such as... Figure 2 As shown, it includes: The evaluation index system generation module is used to determine the evaluation dimensions of the beam-column joints of the prefabricated steel structure and generate an evaluation index system based on the evaluation dimensions. The data acquisition module is used to collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator. The standardization module is used to preprocess the test data, the acceptance test data, and the operation and maintenance monitoring data to obtain a standardized index dataset. The model building module is used to construct a node performance evaluation model for prefabricated steel structure beams and columns based on the standardized index dataset and combined with an improved machine learning algorithm, and output the individual performance score value of each evaluation index through the node performance evaluation model. The weight calculation module is used to determine the comprehensive weight of each evaluation indicator by combining the analytic hierarchy process (AHP) with the entropy weight method. The comprehensive evaluation module is used to combine the evaluation results and comprehensive weights of each evaluation indicator to obtain the comprehensive evaluation value of node performance.

[0074] In a specific embodiment of the present invention, this embodiment takes the performance evaluation of beam-column joints in a high-rise prefabricated steel structure residential project as a specific application scenario. The project adopts end-plate type prefabricated steel structure beam-column joints, and a total of 30 typical joints are selected as evaluation objects. The specific implementation process is as follows: The first step is to construct an evaluation index system: determine five major evaluation dimensions, namely mechanics, seismic resistance, assembly quality, durability, and operation and maintenance performance. Each dimension has primary and secondary indicators, totaling 28 secondary indicators. Clarify the testing standards and value ranges of each indicator to form a hierarchical evaluation index system.

[0075] The second step is multi-source data acquisition: a distributed acquisition terminal and edge computing node collaborative mode is adopted to collect three types of data from 30 nodes: test data comes from laboratory quasi-static, fatigue, and corrosion tests to obtain mechanical seismic indicators such as ultimate bearing capacity; acceptance measurement data comes from on-site torque detection, weld non-destructive testing, etc. to obtain assembly quality indicators such as bolt final tightening torque; operation and maintenance monitoring data is collected over a long period of time through stress, displacement, and corrosion sensors to obtain dynamic data during service and build a multi-source heterogeneous database.

[0076] The third step is data preprocessing: removing duplicate and outlier data, using the Laida criterion to remove outliers, and filling missing data with K-nearest neighbors; using wavelet transform for noise reduction, extreme value normalization to map all data to the [0,1] interval, and Cronbach's Alpha test to confirm data consistency, thus obtaining a standardized index dataset.

[0077] The fourth step is model construction and parameter optimization: Based on a standardized dataset, an evaluation model is constructed by integrating an adaptive dynamic attention residual convolutional network and an adaptive particle swarm optimization algorithm. A node performance monitoring graph structure is built, with sensor measurement points and connection points as nodes, and the connection relationship is determined by dual constraints of spatial distance and performance difference. Dual-scale graph convolution extracts local and global features, and combines attention mechanism, temporal coding and residual gating to output feature vectors. The adaptive particle swarm optimization initializes 30 particles with mean squared error as the objective function, outputs the optimal network parameters, and completes model calibration.

[0078] The fifth step is weight determination and comprehensive evaluation: the analytic hierarchy process is used to construct a judgment matrix to obtain subjective weights, the entropy weight method is used to calculate the data dispersion to obtain objective weights, and linear fusion is used to obtain comprehensive weights; the individual scores of each indicator are output by the model, and the weighted sum is used to obtain the comprehensive evaluation value of the node, and the performance level is divided according to the four-level standard.

[0079] This embodiment verifies the feasibility and accuracy of the method of the present invention, with the evaluation error controlled within 3%, solving many defects of traditional methods, realizing precise management and control of the entire life cycle of nodes, and adapting to the needs of actual engineering applications.

[0080] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0081] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating the performance of a fabricated steel beam-column joint, characterized in that, include: Determine the evaluation dimensions for beam-column joints in prefabricated steel structures, and generate an evaluation index system based on the evaluation dimensions; Collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator; The test data, the acceptance test data, and the operation and maintenance monitoring data are preprocessed to obtain a standardized index dataset; Based on the standardized index dataset, and combined with the improved machine learning algorithm, a node performance evaluation model for prefabricated steel structure beams and columns is constructed, and the individual performance score of each evaluation index is output through the node performance evaluation model. The combined weights of each evaluation index are determined by using the analytic hierarchy process (AHP) combined with the entropy weight method. By combining the evaluation results of each evaluation indicator with their comprehensive weights, a comprehensive evaluation value for node performance is obtained.

2. The method for evaluating the performance of prefabricated steel structure beam-column joints according to claim 1, characterized in that, The evaluation dimensions include mechanical performance, seismic performance, assembly quality, durability, and operation and maintenance performance. Each evaluation dimension has several primary evaluation indicators, and each primary evaluation indicator has several secondary evaluation indicators, forming an evaluation indicator system.

3. The method for evaluating the performance of prefabricated steel structure beam-column joints according to claim 1, characterized in that, The improved machine learning algorithm integrates an adaptive dynamic attention residual convolutional network and an adaptive particle swarm optimization algorithm, with the goal of minimizing node performance evaluation error, to perform adaptive iteration of model parameters and performance prediction.

4. The method for evaluating the performance of prefabricated steel structure beam-column joints according to claim 3, characterized in that, The adaptive dynamic attention residual convolutional network takes a standardized index dataset as input to construct a node performance monitoring graph structure; The node performance monitoring graph structure uses sensor measurement points and component connection points as graph nodes. The node features correspond to the data of each evaluation index, and the node connection relationship is determined by the dual constraints of spatial distance and dynamic performance difference. The local and extended neighborhood performance features are extracted by dual-scale graph convolution, and feature weights are dynamically allocated by graph attention mechanism. Temporal encoding is embedded to fuse temporal evolution information, and a dynamic residual fusion gating mechanism is used to suppress redundant features, outputting node performance feature vectors. The node performance feature vector is mapped through a fully connected layer to obtain the performance prediction value.

5. The method for evaluating the performance of prefabricated steel structure beam-column joints according to claim 4, characterized in that, The adaptive particle swarm optimization algorithm uses the mean square error between the performance prediction value output by the adaptive dynamic attention residual convolutional network and the true label as the objective function to initialize the parameter combination of the particle swarm representation network; it adaptively adjusts the inertia weight coefficient based on the node performance evolution rate to dynamically optimize the particle search speed and position, and outputs the optimal network parameter combination.

6. The method for evaluating the performance of prefabricated steel structure beam-column joints according to claim 3, characterized in that, The weights of the indicators are determined using the analytic hierarchy process (AHP) combined with the entropy weight method, including: A judgment matrix is ​​constructed using the analytic hierarchy process (AHP) to obtain the subjective weights of each dimension and indicator. The objective weights of each indicator are obtained by calculating the dispersion of the data of each indicator using the entropy weight method. The subjective weights and objective weights are linearly fused to obtain a comprehensive weight.

7. A performance evaluation system for prefabricated steel structure beam-column joints, employing the performance evaluation method for prefabricated steel structure beam-column joints as described in any one of claims 1-6, characterized in that, include: The evaluation index system generation module is used to determine the evaluation dimensions of the beam-column joints of the prefabricated steel structure and generate an evaluation index system based on the evaluation dimensions. The data acquisition module is used to collect test data, acceptance test data, and operation and maintenance monitoring data for each evaluation indicator. The standardization module is used to preprocess the test data, the acceptance test data, and the operation and maintenance monitoring data to obtain a standardized index dataset. The model building module is used to construct a node performance evaluation model for prefabricated steel structure beams and columns based on the standardized index dataset and combined with an improved machine learning algorithm, and output the individual performance score value of each evaluation index through the node performance evaluation model. The weight calculation module is used to determine the comprehensive weight of each evaluation indicator by combining the analytic hierarchy process (AHP) with the entropy weight method. The comprehensive evaluation module is used to combine the evaluation results and comprehensive weights of each evaluation indicator to obtain the comprehensive evaluation value of node performance.

Images