Building structure wind load monitoring method and system based on BIM model

The wind load data set is optimized through Bootstrap statistical method and entropy weight method, and the non-stationary wind speed spectrum is generated using recursive convolutional neural network and CBAM attention mechanism, which solves the accuracy problem of the BIM model when simulating wind loads in high-rise buildings and improves the The accuracy and prediction accuracy of wind load monitoring and prediction.

CN119089386AInactive Publication Date: 2024-12-06GUANGZHOU VOCATIONAL COLLEGE OF SCI & TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411218195.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing BIM model is difficult to accurately simulate the non-stationary wind speed spectrum when simulating wind loads in high-rise buildings, resulting in major challenges in wind load monitoring and prediction.

Method used

The Bootstrap statistical method and entropy weight method are used to optimize the wind load data set, combined with the recursive convolutional neural network and the CBAM attention mechanism, to generate more accurate non-stationary wind speed spectrum and predict the wind load response.

Benefits of technology

It improves the accuracy and prediction accuracy of wind load monitoring, generates high-quality non-stationary wind speed spectrum, enhances the prediction ability of wind load response, and provides reliable support for the safety assessment and design of building structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089386A_ABST
    Figure CN119089386A_ABST
Patent Text Reader

Abstract

The invention discloses a building structure wind load monitoring method and system based on a BIM model. The invention relates to the technical field of constructional engineering. Performing entropy weight method fusion on the wind speed data in the data set D, and considering data of different sensors S and different time points t; the information entropy e of each independent wind load S is calculated, the non-stationary wind speed spectrum D'at a certain moment serves as the input of a recursive convolutional neural network (with a CBAM attention mechanism module), and wind load responses at several moments after the moment are output; according to the method, the key wind load point location of the building is accurately extracted, partial response identification is carried out by using the Bootstrap statistical method, the number of independent wind loads is effectively reduced, the variation degree and the information amount of the data are fully considered by using the entropy weight method, the fused wind speed data is more accurate and comprehensive, the convolutional neural network in a recursive form is utilized, and the method is more accurate and comprehensive. And a CBAM attention mechanism is embedded, so that spatial structure characteristics and time dynamic change characteristics in a wind speed spectrum can be captured at the same time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of building engineering, and in particular to a detection technology for building wind loads, and in particular to a method and system for monitoring building structure wind loads based on a BIM model. Background Art

[0002] Wind load on high-rise buildings refers to the dynamic pressure caused by air flow on high-rise buildings. Wind load is actually a random time-varying live load, which is different from general static loads, such as floor and roof live loads, crane loads, snow loads, etc. The impact of wind loads on buildings is mainly reflected in horizontal displacement, especially for super high-rise buildings, whose structural design often needs to consider wind loads as the main control load. When airflow passes through tall structures, it will produce obvious three-dimensional wind load effects, including downwind, crosswind and torsional wind loads, which will cause the structure to vibrate in three directions.

[0003] BIM (Building Information Modeling) technology plays an important role in wind load simulation of high-rise buildings. BIM models can be presented to designers, owners and related stakeholders through 3D visualization technology, enabling them to better understand the design concept and building form. More importantly, BIM software has the ability to simulate wind loads, helping designers optimize structural design by simulating the performance of building structures under different wind speeds, wind directions and other conditions.

[0004] When simulating wind loads on high-rise buildings, the wind speed spectrum is a crucial input parameter. The wind speed spectrum describes the statistical characteristics of wind speed changes over time and is essential for accurately simulating wind loads. However, a major problem faced by current BIM models when simulating wind loads is that the existing wind speed spectrum often cannot simulate different stable or non-steady states.

[0005] The steady wind speed spectrum assumes that the statistical characteristics of wind speed (such as mean and variance) do not change over time. This simplified treatment may not accurately reflect the complexity of the actual wind field in some cases. The non-steady wind speed spectrum takes into account the changing characteristics of wind speed over time and is closer to the actual situation of the actual wind field. However, due to the complexity and diversity of the non-steady wind speed spectrum, the current BIM model still faces great challenges in simulating non-steady wind loads.

[0006] To this end, the present invention proposes a building structure wind load monitoring method and system based on a BIM model. Summary of the invention

[0007] In view of this, the present invention hopes to provide a method and system for monitoring wind loads on building structures based on a BIM model to solve or alleviate the technical problems existing in the prior art, namely, how to optimize the generation of non-stationary wind speed spectra to achieve wind load monitoring of building structures, and at least provide a beneficial option for this; the technical solution of the present invention is implemented as follows:

[0008] First, the wind load monitoring method of building structure based on BIM model:

[0009] 1. Overview:

[0010] The present invention aims to optimize the monitoring and prediction of wind loads on high-rise buildings through advanced statistical analysis and data fusion technology. First, the Bootstrap statistical method is used to resample and evaluate the stability of wind load points in the BIM model or the actual building model, aiming to reduce the number of independent wind loads, improve the efficiency of partial response identification, and form a high-quality wind load data set. Secondly, the entropy weight method is used to fuse the wind speed data from different sensors and different time points, calculate the information entropy of each independent wind load, and assign corresponding wind speed drift weights according to the size of the information entropy, so as to generate a more accurate non-stationary wind speed spectrum. Finally, with the help of a recursive convolutional neural network and the CBAM attention mechanism embedded, the non-stationary wind speed spectrum is used as input to predict the wind load response at several moments in the future, providing strong support for the structural design and safety assessment of high-rise buildings.

[0011] (II) Technical solution:

[0012] To achieve the above technical objectives, based on the BIM model or the actual building model, after identifying and extracting the key wind load points s on the building, the following steps S1 to S3 are performed.

[0013] 2.1 Step S1, wind load point extraction and partial response identification:

[0014] The data set D of partial response identification wind load is formed through statistical analysis;

[0015] Among them, Bootstrap resampling is performed on the extracted wind load points to evaluate the stability and correlation of the wind loads at each point. Then, through statistical analysis, independent wind loads are selected and deleted to form a data set D for partial response identification of wind loads. The data set D includes different independent wind loads S:

[0016] Assume that the original wind load point data set is P = {p1, p2, ..., p n}, where p i Represents the i-th wind load point. After Bootstrap resampling, a series of resampled data sets are obtained. Each element of this contains m (repeatable) random samples drawn from P, and the variance of the wind load is calculated and covariance ρ j , filter the original wind load points:

[0017]

[0018] Where D is the data set for partial response identification of wind loads; S i represents the ith independent wind load; P is the original wind load point data set; Indicates that it contains S i The resampled dataset P j * The variance of j (S i ) indicates that it includes independent wind load S i The resampled dataset P j * covariance of; threshold and threshold T ρ are pre-set stability and correlation thresholds.

[0019] 2.2 Step S2, data fusion and non-stationary wind speed spectrum generation:

[0020] The wind speed data in the data set D are fused by entropy weight method, considering the data from different sensors S and different time points t; the information entropy e of each independent wind load S is calculated. The smaller the information entropy, the greater the degree of variation and the more information. According to the result of information entropy, a corresponding wind speed drift weight w is assigned to each independent wind load S. All weighted independent wind loads S' constitute a non-stationary wind speed spectrum D' in the form of a weight matrix.

[0021] 2.2.1 Step S200, obtaining information entropy:

[0022] Suppose the wind speed data in the data set D is {vS,t}, where S represents different sensors and t represents different time points. Calculate the information entropy e of each independent wind load S S :

[0023] e S =-∑ t p S,t logp S,t ;

[0024] Among them, p S,t Represents the wind speed data v of sensor S at time point t S,t Probability of occurrence.

[0025] 2.2.2 Step S201, weight assignment:

[0026] According to the information entropy result, each independent wind load S is assigned a corresponding wind speed drift weight w. S .

[0027] Weight w S and information entropy e S Inversely proportional, that is, the smaller the information entropy, the greater the weight:

[0028]

[0029] in, is a normalization factor to ensure that the sum of all weights is 1.

[0030] 2.2.3 Step S203, forming a non-stationary wind speed spectrum D′:

[0031] All weighted independent wind loads S′ are used to form a non-stationary wind speed spectrum D′ in the form of a weight matrix, namely:

[0032] D′={w S ·S′|S′∈D};

[0033] Among them, w S S′ represents the independent wind load S′ multiplied by its corresponding weight w S .

[0034] 2.3 Step S3, wind load response prediction:

[0035] The non-stationary wind speed spectrum D' at a certain moment is used as the input of the recursive convolutional neural network (with the CBAM attention mechanism module), and the wind load response at the next few moments after that moment is output. The structure of the recursive convolutional neural network is connected in sequence as follows:

[0036] 2.3.1 Input layer:

[0037] Input non-stationary wind speed spectrum D′∈R n×m , where n is the number of sensors and m is the time step.

[0038] 2.3.2 First CBAM convolutional layer:

[0039] Use convolution kernels to extract features from input data. The convolution operation can capture local features in the input data and cover the entire input through a sliding window. The CBAM attention mechanism module further enhances the focus on important features by adaptively adjusting the weights of feature maps, allowing the network to focus more on information that is useful for prediction tasks. Including:

[0040] 1) Convolution kernel K1∈R k1×k1 , where k1 is the convolution kernel size;

[0041] 2) ReLU activation function;

[0042] 3) CBAM outputter F1: F1 = CBAM (ReLU (D′*K1)).

[0043] 2.3.3 Recursive layer:

[0044] Process time series data and capture the time dependency in the input data. The recursive layer receives the input of the current time step and the recursive output of the previous time step, uses the weight matrix and bias for linear transformation, and then applies the activation function for nonlinear transformation, thereby achieving modeling and prediction of time series data. Including:

[0045] Recursive function: R(x t )=σ(W r x t +U r R(x t-1 )+b r );

[0046] Among them, x t is the input of the current time step, R(x t-1 ) is the recursive output of the previous time step, σ is the ReLU activation function, W r and U r is the weight matrix, b r is bias.

[0047] 2.3.4 The second CBAM convolutional layer:

[0048] It uses convolution kernels for feature extraction and uses the CBAM attention mechanism module to enhance the focus on important features. However, since it receives the output of the recursive layer, the features it extracts contain the temporal dependencies of the time series data. Including:

[0049] 1) Convolution kernel: K2∈R k2×k2 , where k2 is the convolution kernel size;

[0050] 2) ReLU activation function;

[0051] 3) CBAM output and recursive input F2 = CBAM (ReLU (F1 * K2)), x t =F2.

[0052] 2.3.5 Output layer:

[0053] The output of the recursive layer is linearly transformed through a fully connected layer to obtain the predicted wind load response. The fully connected layer uses a weight matrix and bias to linearly combine the input data to generate the final prediction result.

[0054] The output layer is a fully connected layer O=W oR(x t )+b o ;

[0055] Among them, W o is the weight matrix, b o is the offset and O is the predicted wind load response.

[0056] 2.3.6 Training method of recursive convolutional neural network:

[0057] P1. Data preprocessing: Remove noise, outliers and irrelevant information in the non-stationary wind speed spectrum D′, and adjust the numerical range of the input data to a relatively small range through Z-score normalization.

[0058] P2. Model initialization: Use normal distribution to randomly initialize the weights and biases in the model.

[0059] P3, forward propagation: The preprocessed non-stationary wind speed spectrum D′ is input into the network. The features of the input data are extracted layer by layer through the convolutional layer and the CBAM attention mechanism module. The recursive layer processes the extracted features to capture the temporal dependencies in the time series data. The fully connected layer converts the output of the recursive layer into the final wind load response prediction value.

[0060] P4. Loss calculation: Use cross-entropy to measure the difference between the model prediction value and the true label. And calculate the loss value of the current batch of data based on the loss function.

[0061] P5, Back Propagation: Use the chain rule to calculate the gradient of the loss function with respect to the model parameters layer by layer. Based on the calculated gradient and the pre-set learning rate, use the Adam optimizer to update the model's weights and biases.

[0062] P6. Iterative training: Repeat forward propagation, loss calculation and back propagation, and continuously iterate the training process until the stopping condition is met (the maximum number of iterations is reached or the loss function value is less than the threshold).

[0063] (III) Mechanism for solving technical problems:

[0064] 3.1 Wind load point extraction and partial response identification:

[0065] The Bootstrap resampling technique is applied to perform multiple random sampling of the extracted wind load points to evaluate the stability and correlation of the wind loads at each point. Through statistical analysis, the independent wind loads that contribute more to the overall wind load response are selected, and the redundant information is deleted to form a data set D for partial response identification of wind loads. This process effectively reduces the number of wind loads that need to be processed independently and improves the efficiency and accuracy of subsequent analysis.

[0066] 3.2 Data fusion and non-stationary wind speed spectrum generation:

[0067] For the wind speed data in the data set D, the entropy weight method is used for fusion processing. This method is based on the information entropy theory and can objectively reflect the degree of variation and the amount of information of the data. By calculating the information entropy of each independent wind load, a greater weight is given to the wind load with a large degree of variation and a large amount of information, and vice versa. This process fully considers the differences and importance of data from different sensors and at different time points, and improves the overall quality of wind speed data. All weighted independent wind loads are combined according to certain rules to form a non-stationary wind speed spectrum D' in the form of a weight matrix. This wind speed spectrum not only includes the non-stationary characteristics of wind speed changes over time, but also reflects the spatial correlation and mutual influence between different wind loads.

[0068] 3.3 Wind load response prediction:

[0069] The recursive convolutional neural network model captures the spatial structural characteristics and temporal dynamic change characteristics in the wind speed spectrum, improving the accuracy of wind load response prediction. The non-stationary wind speed spectrum D' at a certain moment is used as the input of the model, and the wind load response at the next few moments after that moment is output through the recursive calculation and convolution operation of the model. This process makes full use of the rich information in the non-stationary wind speed spectrum and realizes the accurate prediction of the future wind load response of the building structure.

[0070] Second, the building structure wind load monitoring system based on the BIM model:

[0071] like Figure 3 As shown, the system includes a processor and a memory connected to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the wind load monitoring method as described above; wherein the processor is connected to:

[0072] (1) Apply the Bootstrap statistical method to perform multiple random sampling of the extracted wind load points. Wind load point extraction and partial response identification module: Based on BIM (Building Information Model) or actual building model, accurately identify and extract the key points on the building surface that are most affected by wind loads. Through statistical analysis, select independent wind loads that contribute more to the overall wind load response, delete redundant information, and form a data set D for partial response wind load identification.

[0073] Furthermore, the pertinence and efficiency of wind load monitoring can be improved, the amount of data processing can be reduced, and high-quality basic data can be provided for subsequent data fusion and wind speed spectrum generation.

[0074] (2) For the wind speed data in the data set D, the entropy weight method is used to perform data fusion and non-stationary wind speed spectrum generation module: the information entropy of each independent wind load is calculated, and the corresponding wind speed drift weight is assigned according to the size of the information entropy. All weighted independent wind loads are combined according to certain rules to form a non-stationary wind speed spectrum D' in the form of a weight matrix.

[0075] Thus, the overall quality and reliability of wind speed data can be improved. A non-stationary wind speed spectrum containing rich information is generated, providing strong support for the subsequent wind load response prediction.

[0076] (3) Wind load response prediction module for implementing recursive convolutional neural network model: embedded with CBAM (Convolutional Block Attention Module) attention mechanism. The non-stationary wind speed spectrum D' at a certain moment is used as the input of the model, and the wind load response at several moments after that moment is output through the recursive calculation and convolution operation of the model.

[0077] Furthermore, the accuracy and timeliness of wind load response prediction can be improved, providing accurate and reliable prediction results for safety assessment and design optimization of building structures.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. Improve the accuracy of wind load monitoring: This invention effectively reduces the number of independent wind loads and data redundancy by accurately extracting key wind load points of buildings and using the Bootstrap statistical method for partial response identification, thereby improving the accuracy of wind load monitoring. The entropy weight method is used to fuse wind speed data from different sensors and at different time points, fully considering the degree of data variation and the amount of information, making the fused wind speed data more accurate and comprehensive, providing a reliable basis for subsequent wind load response prediction.

[0080] 2. Generate high-quality non-stationary wind speed spectrum: The present invention generates a non-stationary wind speed spectrum in the form of a weight matrix by assigning corresponding wind speed drift weights to different independent wind loads. The wind speed spectrum not only includes the non-stationary characteristics of the wind speed, but also reflects the spatial correlation and mutual influence between wind loads, providing rich information for wind load response prediction.

[0081] 3. Improve the prediction ability of wind load response: The present invention utilizes a recursive convolutional neural network and embeds the CBAM attention mechanism, which can simultaneously capture the spatial structure characteristics and temporal dynamic change characteristics in the wind speed spectrum, thereby improving the prediction accuracy and timeliness of wind load response. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0083] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0084] Figure 2 It is a schematic diagram of the model training method of the present invention;

[0085] Figure 3 It is a schematic diagram of the system composition of the present invention;

[0086] Figure 4 Schematic diagram of the execution procedure of step S1 and step S2 of the present invention;

[0087] Figure 5 It is a schematic diagram of the model architecture program of the present invention;

[0088] Figure 6 It is a schematic diagram of the model training method of the present invention;

[0089] Figure 7 It is a visualization diagram of the application example; the horizontal axis is the wind load degree (%), and the vertical axis is the predicted wind load response value;

[0090] Figure 8 It is an operational diagram for Revit to simulate the wind load of a specified engineering structure based on the input structural response coefficient. DETAILED DESCRIPTION

[0091] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0092] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0093] Explanation of relevant terms:

[0094] (1) Wind load: The pressure exerted by air flow on the engineering structure. Its magnitude is proportional to the square of wind speed;

[0095] (2) BIM model: For example, the existing Revit model can simulate the wind load of a specified engineering structure based on the input structural response coefficient, such as Figure 7 As shown;

[0096] (3) Bootstrap resampling: It is a statistical resampling technique that generates a series of new resampled data sets by randomly extracting samples (repeatable) from the original data set. These resampled data sets are used to evaluate the stability and relevance of the original data set.

[0097] Embodiment 1: Figure 1 As shown, this embodiment will provide a method for monitoring wind loads on building structures based on a BIM model:

[0098] In this embodiment, regarding step S1, wind load point extraction and partial response identification:

[0099] Based on the BIM model or the actual building model, the key points on the building that may be affected by wind loads are determined through professional structural analysis software or manual identification. Bootstrap resampling is performed on the extracted wind load points to evaluate the stability and correlation of the wind loads at each point. Then, through statistical analysis, independent wind loads are selected and deleted to form a data set D for partial response identification of wind loads. Data set D includes different independent wind loads S:

[0100] Assume that the original wind load point data set is P = {p1, p2, ..., p n}, where p i Represents the i-th wind load point. After Bootstrap resampling, a series of resampled data sets are obtained. Each element of this contains m (repeatable) random samples drawn from P, and the variance of the wind load is calculated and covariance ρ j , filter the original wind load points:

[0101]

[0102] Where D is the data set for partial response identification of wind loads; S i represents the ith independent wind load; P is the original wind load point data set; Indicates that it contains S i The resampled dataset P j * The variance of j (S i) indicates that it includes independent wind load S i The resampled dataset P j * covariance of; threshold and threshold T ρ are pre-set stability and correlation thresholds.

[0103] For each resampled data set, the variance and covariance of its wind load are calculated. Variance is used to measure the stability of wind load points, while covariance is used to measure the correlation between different points. By calculating variance and covariance, we can quantify the stability and correlation of wind load points, providing a clear basis for subsequent screening. This helps us remove those points with poor stability or weak correlation, thereby improving the efficiency and accuracy of the prediction model.

[0104] It is understandable that Bootstrap resampling can help us evaluate the stability and relevance of wind load points, thereby screening out points that are more valuable for wind load response prediction. This method can improve the reliability of data and the accuracy of prediction.

[0105] It is understandable that the original wind load points are screened according to the pre-set stability and correlation thresholds. Only those points with high stability and strong correlation will be retained and enter the subsequent analysis and prediction as independent wind loads. Screening independent wind loads can remove those points that have little impact on the prediction results, thereby reducing data redundancy and noise. This helps to improve the generalization ability and accuracy of the prediction model, making the prediction results more reliable and useful.

[0106] In this embodiment, regarding step S2, data fusion and non-stationary wind speed spectrum generation: the wind speed data in the data set D are fused by entropy weight method, considering the data of different sensors S and different time points t; the information entropy e of each independent wind load S is calculated, and the smaller the information entropy, the greater the degree of variation and the more information. According to the result of the information entropy, a corresponding wind speed drift weight w is assigned to each independent wind load S. All weighted independent wind loads S' constitute a non-stationary wind speed spectrum D' in the form of a weight matrix.

[0107] Specifically, step S200, obtain information entropy: suppose the wind speed data in the data set D is {vS, t}, where S represents different sensors and t represents different time points. Calculate the information entropy e of each independent wind load S S :e S =-∑ t p S,t logp S,t ;

[0108] Among them, p S,t Represents the wind speed data v of sensor S at time point tS,t Probability of occurrence.

[0109] It can be understood that information entropy is an important indicator to measure the degree of data variation and the amount of information. S When we consider the wind speed data from different sensors S and different time points t. The smaller the information entropy, the greater the degree of variation of the wind speed data of the wind load, that is, the more information it contains, and the more valuable the prediction of the wind load response. By calculating the information entropy, we can quantify the degree of variation and amount of information of the wind speed data of each independent wind load, providing a basis for the subsequent wind speed drift weight assignment. This helps us more accurately identify wind speed data that is valuable for predicting wind load responses.

[0110] Specifically, step S201, weight assignment: assign a corresponding wind speed drift weight w to each independent wind load S according to the result of information entropy S .

[0111] Preferably, the weight w S and information entropy e S Inversely proportional, that is, the smaller the information entropy, the greater the weight:

[0112]

[0113] in, is a normalization factor to ensure that the sum of all weights is 1.

[0114] It can be understood that according to the result of information entropy, we assign a corresponding wind speed drift weight w to each independent wind load S S . Weight wS and information entropy e S Inversely proportional, that is, the smaller the information entropy, the greater the weight. This means that wind loads with large variability and large amounts of information will occupy a more important position in the prediction. By assigning weights, we can weightedly fuse the wind speed data of different independent wind loads according to the amount of information, thereby obtaining a more accurate and comprehensive wind speed spectrum. This helps to improve the accuracy and reliability of wind load response prediction.

[0115] Specifically, in step S203, a non-stationary wind speed spectrum D′ is formed: all weighted independent wind loads S′ are used to form a non-stationary wind speed spectrum D′ in the form of a weight matrix, that is:

[0116] D′={w s ·S′|S′∈D};

[0117] Among them, w S S′ represents the independent wind load S′ multiplied by its corresponding weight w S .

[0118] It can be understood that the non-stationary wind speed spectrum D′ takes into account the temporal and spatial variation characteristics of wind speed and can more accurately describe the wind speed distribution of the actual wind field. Forming the non-stationary wind speed spectrum D′ is an important step in the prediction of wind load response. It takes into account the temporal and spatial variation characteristics of wind speed and can provide more accurate and comprehensive wind speed information, providing a reliable data basis for subsequent wind load response prediction. This helps to improve the accuracy and practicality of the prediction.

[0119] In this embodiment, the Python execution program of the above steps S1 to S2 is as follows: Figure 4 As shown in the figure, the principle is as follows: the original wind load point data set is extracted from the BIM model or the actual building model. The Bootstrap method is used for resampling to generate multiple resampled data sets. The variance and covariance of each resampled data set are calculated to evaluate the stability and correlation of the wind load point. Independent wind load points are screened out according to the preset stability and correlation thresholds.

[0120] For the selected independent wind load points, simulate and generate wind speed data for each sensor at different time points. Calculate the information entropy of each wind load. The smaller the information entropy, the greater the degree of variation of the wind speed data and the more information. According to the result of the information entropy, assign a corresponding wind speed drift weight to each wind load. The smaller the information entropy, the greater the weight. Use the outer product operation to combine the weight with the wind load point to form a non-stationary wind speed spectrum in the form of a weight matrix.

[0121] In this embodiment, regarding step S3, wind load response prediction: the non-stationary wind speed spectrum D' at a certain moment is used as the input of a recursive convolutional neural network (with a CBAM attention mechanism module), and the wind load response at a few moments after the moment is output. The structure of the recursive convolutional neural network is connected in sequence as follows:

[0122] (1) Input layer:

[0123] The input layer receives the non-stationary wind speed spectrum D′ as the input of the recursive convolutional neural network. D′ is a two-dimensional matrix, where n represents the number of sensors and m represents the time step. Such an input design allows the network to simultaneously consider the wind speed data of multiple sensors at different time points.

[0124] By receiving multi-dimensional input data, the network can capture the spatiotemporal characteristics of the wind field more comprehensively and provide richer information for subsequent wind load response prediction.

[0125] (2) First CBAM convolutional layer:

[0126] The first CBAM convolution layer uses convolution kernels to extract features from the input data. The convolution operation can capture local features in the input data and cover the entire input through a sliding window. The CBAM attention mechanism module further enhances the focus on important features by adaptively adjusting the weights of feature maps, allowing the network to focus more on information that is useful for prediction tasks. Including:

[0127] 1) Convolution kernel K1∈R k1×k1 , where k1 is the convolution kernel size;

[0128] 2) ReLU activation function;

[0129] 3) CBAM outputter F1: F1 = CBAM (ReLU (D′*K1)).

[0130] It can be understood that through the convolution operation, the network is able to extract local features in the input data and enhance the attention to important features through the CBAM attention mechanism module, thereby improving the accuracy and robustness of the prediction.

[0131] (3) Recursive layer:

[0132] The recursive layer processes time series data and captures the temporal dependencies in the input data. It receives the input of the current time step and the recursive output of the previous time step, performs linear transformation using the weight matrix and bias, and then applies the activation function for nonlinear transformation, thereby achieving modeling and prediction of time series data. It includes:

[0133] Recursive function: R(x t )=σ(W r x t +U r R(x t-1 )+b r );

[0134] Among them, x t is the input of the current time step, R(x t-1 ) is the recursive output of the previous time step, σ is the ReLU activation function, W r and U r is the weight matrix, b r is bias.

[0135] It can be understood that the design of the recurrent layer enables the network to capture and exploit the temporal dependencies in the time series data, which is crucial for the prediction of wind load response. By considering the influence of historical data, the network is able to predict future wind load responses more accurately.

[0136] (4) Second CBAM convolutional layer:

[0137] The second CBAM convolutional layer is similar to the first CBAM convolutional layer. It also uses convolution kernels for feature extraction and enhances the focus on important features through the CBAM attention mechanism module. However, since it receives the output of the recursive layer, the features it extracts contain the temporal dependencies of the time series data. Including:

[0138] 1) Convolution kernel: K2∈R k2×k2 , where k2 is the convolution kernel size;

[0139] 2) ReLU activation function;

[0140] 3) CBAM output and recursive input F2 = CBAM (ReLU (F1 * K2)), x t =F2.

[0141] It can be understood that by applying the convolution operation and the CBAM attention mechanism module again, the network is able to further extract and enhance features containing temporal dependencies, thereby improving the accuracy and reliability of predictions.

[0142] (5) Output layer:

[0143] The output of the recursive layer is linearly transformed through a fully connected layer to obtain the predicted wind load response. The fully connected layer uses a weight matrix and bias to linearly combine the input data to generate the final prediction result.

[0144] The output layer is a fully connected layer O=W o R(x t )+b o ;

[0145] Among them, W o is the weight matrix, b o is the offset and O is the predicted wind load response.

[0146] Through the linear transformation of the fully connected layer, the network can convert the extracted features into specific prediction values, thereby achieving accurate prediction of wind load response.

[0147] In this embodiment, the python architecture program of the above recursive convolutional neural network is as follows Figure 5 As shown in the figure, the principle is to first import the necessary TensorFlow and Keras modules. Then a RecursiveConvNetWithCBAM class is defined, which inherits from tf.keras.Model. All layers of the network are constructed in this class, and the forward propagation process is defined in the call method. Finally, a network instance is created and the summary method is called to print the structure of the network.

[0148] Embodiment 2: Figure 2 , 6 As shown, this embodiment further provides a training method for a recursive convolutional neural network based on the first embodiment:

[0149] P1. Data preprocessing: Remove noise, outliers and irrelevant information in the non-stationary wind speed spectrum D′, and adjust the numerical range of the input data to a relatively small range through Z-score normalization.

[0150] P2. Model initialization: Use normal distribution to randomly initialize the weights and biases in the model.

[0151] P3, forward propagation: The preprocessed non-stationary wind speed spectrum D′ is input into the network. The features of the input data are extracted layer by layer through the convolutional layer and the CBAM attention mechanism module. The recursive layer processes the extracted features to capture the temporal dependencies in the time series data. The fully connected layer converts the output of the recursive layer into the final wind load response prediction value.

[0152] P4. Loss calculation: Use cross-entropy to measure the difference between the model prediction value and the true label. And calculate the loss value of the current batch of data based on the loss function.

[0153] P5, Back Propagation: Use the chain rule to calculate the gradient of the loss function with respect to the model parameters layer by layer. Based on the calculated gradient and the pre-set learning rate, use the Adam optimizer to update the model's weights and biases.

[0154] P6. Iterative training: Repeat forward propagation, loss calculation and back propagation, and continuously iterate the training process until the stopping condition is met (the maximum number of iterations is reached or the loss function value is less than the threshold).

[0155] Application examples:

[0156] 1. Overview:

[0157] This example aims to predict the wind load response generated by different steady or non-steady wind speed spectra through the BIM building model combined with a recursive convolutional neural network (equipped with a CBAM attention mechanism module), and to visualize the fit between the prediction results and the actual simulated wind load degree curve to verify the effectiveness of the model.

[0158] (II) Experimental data:

[0159] A BIM building model of a 76-benchmark with one-dimensional structural response and a 16-story asymmetric structure with two-dimensional structural response was selected.

[0160] A set of data containing steady and non-steady wind speed spectra is used to simulate wind load excitation. The data numerical examples are as follows:

[0161] (1) Steady wind speed spectrum: [0.05, 0.06, ..., 5] m / s (100 data points in total);

[0162] (2) Non-stationary wind speed spectrum: [0.03, 0.04, 0.8, ..., 3] m / s (a total of 100 data points, including mutations and fluctuations);

[0163] (3) Actual wind load degree curve: obtained through BIM software simulation and used as a comparison benchmark for the prediction results.

[0164] (III) Experimental methods:

[0165] The trained model is applied to the new wind speed spectrum data to generate 100 sets of scattered point predicted wind load responses O.

[0166] (IV) Experimental results and visualization:

[0167] like Figure 7 As shown in Figure 1, the wind load response O predicted by 100 sets of scattered points is attached to the visualization diagram of the wind load degree curve simulated in reality. The visualization diagram shows that most of the scattered points fit the wind load degree curve closely, indicating that the model prediction results are highly consistent with the actual situation.

[0168] (V) Conclusion

[0169] This example verifies the effectiveness of the wind load response prediction method based on the BIM building model and the recursive convolutional neural network (equipped with the CBAM attention mechanism module). The wind load response predicted by 100 groups of scattered points is basically consistent with the wind load degree curve of the real simulation, proving the prediction accuracy of the model under different steady or non-steady wind speed spectrum excitations. This provides strong technical support for the wind-resistant design of building structures.

[0170] All the above embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

[0171] For those skilled in the art, it can be further appreciated that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0172] At the same time, those skilled in the art can understand that all or part of the processes in all the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Claims

1. A method for monitoring wind loads on building structures based on a BIM model, comprising identifying and extracting key wind load points s on a building based on a BIM model or an actual building model, characterized in that: Perform the following operation process: S1, forming a data set D of partial response identification wind loads through statistical analysis; S2, the wind speed data in the data set D are fused by entropy weight method, considering the data from different sensors S and different time points t; the information entropy e of each independent wind load S is calculated and the corresponding wind speed drift weight w is assigned to each independent wind load S, and all the weighted independent wind loads S' constitute a non-stationary wind speed spectrum D'; S3, the non-stationary wind speed spectrum D' at a certain moment is used as a recursive convolutional neural network to output the wind load response at several moments after that moment.

2. The wind load monitoring method according to claim 1, characterized in that: The statistical analysis method of S1 is the Bootstrap statistical method: Assume that the original wind load point data set is P = {p1, p2, ..., p n }, where p i Represents the i-th wind load point; after Bootstrap resampling, a series of resampled data sets are obtained Each element contains m random samples drawn from P, and the variance of their wind loads is calculated and covariance ρ j , filter the original wind load points: Where D is the data set for partial response identification of wind loads; S i represents the ith independent wind load; P is the original wind load point data set; Indicates that it contains S i The resampled dataset P j * The variance of ρ j (S i ) indicates that it includes independent wind load S i The resampled dataset P j * covariance of; threshold and threshold T ρ are pre-set stability and correlation thresholds.

3. The wind load monitoring method according to claim 1, characterized in that: In S2, the execution process of the entropy weight method is as follows: S200, assuming that the wind speed data in the data set D is {vS, t}, where S represents different sensors and t represents different time points; calculate the information entropy e of each independent wind load S S : e S =-∑ t p S,t logp s,t ; Among them, p S,t Represents the wind speed data v of sensor S at time point t S,t Probability of occurrence; S201, assign a corresponding wind speed drift weight w to each independent wind load S according to the information entropy result. S ; Weight w S and information entropy e S Inversely proportional: in, is the normalization factor.

4. The wind load monitoring method according to claim 3, characterized in that: The method for forming the non-stationary wind speed spectrum D′ is as follows: all weighted independent wind loads S′ are combined to form a non-stationary wind speed spectrum D′ in the form of a weight matrix: D′={w S ·S′|S′∈D}; Among them, w S S′ represents the independent wind load S′ multiplied by its corresponding weight w S .

5. The wind load monitoring method according to any one of claims 1 to 3, characterized in that: In S3, the recursive convolutional neural network includes the following sequentially connected: Input layer of non-stationary wind speed spectrum D′; The first CBAM convolutional layer uses convolution kernels to extract features from the input data; Recursive layers for processing time series data; The second CBAM convolution layer uses convolution kernels for feature extraction; The output of the recursive layer is linearly transformed through a fully connected layer to obtain the output layer of the predicted wind load response.

6. The wind load monitoring method according to claim 5, characterized in that: The first CBAM convolutional layer includes: 1) Convolution kernel K1∈R k1×k1 , where k1 is the convolution kernel size; 2) ReLU activation function; 3) CBAM outputter F1: F1 = CBAM (ReLU (D′*K1)).

7. The wind load monitoring method according to claim 5, characterized in that: The recursive layer includes the recursive function: R(x t )=σ(W r x t +U r R(x t-1 )+b r ); Among them, x t is the input of the current time step, R(x t-1 ) is the recursive output of the previous time step, σ is the ReLU activation function, W r and U r is the weight matrix, b r is bias.

8. The wind load monitoring method according to claim 5, characterized in that: The second CBAM convolutional layer includes: 1) Convolution kernel: K2∈R k2×k2 , where k2 is the convolution kernel size; 2) ReLU activation function; 3) CBAM output and recursive input F2 = CBAM (ReLU (F1 * K2)), x t =F2.

9. The wind load monitoring method according to claim 5, characterized in that: The output layer is a fully connected layer O=W o R(x t )+b o ; Among them, W o is the weight matrix, b o is the offset and O is the predicted wind load response.

10. Building structure wind load monitoring system based on BIM model, characterized by: The system comprises a processor and a memory connected to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the wind load monitoring method according to any one of claims 1 to 9; wherein the processor is connected to, The Bootstrap statistical method is used to perform multiple random sampling of the extracted wind load points to extract the wind load points and identify the partial response module; For the wind speed data in data set D, the entropy weight method is used to perform data fusion and non-stationary wind speed spectrum generation module; A wind load response prediction module for implementing a recursive convolutional neural network model.

Citation Information

Cited By

  • Servo cable bending stress adaptive prediction method based on multi-sensor fusion

    CN120372520A