Method for determining installation opportunity of connecting truss of super high-rise building based on GBRT-LSTM hybrid model

Abstract

The invention discloses a method for determining the installation time of a connecting truss of a super high-rise building based on a GBRT-LSTM hybrid model, and the method comprises the steps: operating a finite element model under a part of construction conditions through employing a sparse simulation strategy, and building a sparse structure response data set; a GBRT-LSTM hybrid model is constructed to effectively capture spatial and temporal features of a data set, and the network structure of the model comprises an input layer, a feature extraction layer, a weighted fusion layer, an optimization layer and an output layer; inputting the sparse structure response data set into the model for training and verification, taking the final GBRT-LSTM hybrid model as an agent model to predict the structure response of the unsimulated construction condition, and complementing the structure response data set; performing multi-index analysis on the data set by adopting a TOPSIS algorithm to obtain a comprehensive score of each construction condition, and determining an optimal installation time node; and performing clustering analysis on the comprehensive score by adopting a K-means + + algorithm to determine an optimal installation time window. According to the method disclosed by the invention, the installation time of the connecting truss of the super high-rise building can be quickly determined, and support is provided for construction of the truss of the super high-rise building.

Description

Technical Field

[0002] This invention relates to a method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model. By combining GBRT and LSTM models and TOPSIS and K-means++ algorithms, the installation timing of connecting trusses in super high-rise buildings can be quickly determined, which belongs to the field of construction technology. Background Technology

[0003] Supertall buildings are key to achieving high-quality urban development. They often employ a twin-tower structure with connecting trusses between the towers to meet the requirements of different functional zones. These connecting trusses integrate the individual towers of a multi-tower building, providing sufficient lateral stiffness. The timing of the connecting truss installation significantly impacts the overall performance of the building. Therefore, developing an effective method for rapidly determining the installation timing of connecting trusses in supertall buildings has significant engineering application value.

[0004] Current methods for determining truss installation timing primarily involve establishing finite element models to simulate structural deformation and internal forces under all installation conditions, thereby identifying the optimal installation time. However, these methods require extensive computation, making the determination of the optimal solution extremely time-consuming. Surrogate models, by replacing finite element simulation with data-driven predictions, can significantly improve computational efficiency. However, these methods typically focus only on local characteristics of the data and cannot effectively capture the spatiotemporal correlations of the data.

[0005] Furthermore, current methods for determining the timing of truss installation typically focus on a single indicator (such as deformation or stress) to determine the optimal installation time for connecting trusses. They do not consider a comprehensive evaluation of multiple structural indicators (such as maximum floor displacement, inter-story drift ratio, inter-story drift angle, and truss internal forces), and fail to adequately quantify the evaluation results. Therefore, it is necessary to develop a rapid method for determining the installation timing of connecting trusses in super high-rise buildings that can effectively capture the spatiotemporal correlation between data and comprehensively consider multiple indicators. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model, thereby improving the efficiency and reliability of optimizing the installation timing of connecting trusses in super high-rise buildings. Its specific contents include:

[0007] A method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model includes the following steps:

[0008] A. Use a sparse simulation strategy to run finite element models under some construction conditions and establish a sparse structure response dataset;

[0009] B. Constructing a gradient boosting regression tree-long short-term memory (GBRT-LSTM) hybrid model effectively captures the spatiotemporal features of the dataset. Its network structure consists of: input layer, feature extraction layer, weighted fusion layer, optimization layer, and output layer.

[0010] C. Input the sparse structural response dataset into the model for training. The trained GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response of unsimulated construction conditions and complete the structural response dataset.

[0011] D. The TOPSIS algorithm (Technique for Order Preference by Similarity to an Ideal Solution) is used to perform multi-index analysis on the dataset to obtain the comprehensive score of each construction condition and determine the optimal installation time node.

[0012] E. The K-means++ algorithm is used to perform cluster analysis on the comprehensive score to determine the optimal installation time window.

[0013] Furthermore, step A specifically includes:

[0014] A1. Based on the construction drawings, establish a detailed three-dimensional finite element model of the super high-rise building in finite element software;

[0015] A2. Starting from the floor where the connecting truss is located, the construction of the main tower core tube is defined as a construction case for each floor upwards until the top is reached;

[0016] A3. Employing a sparse simulation strategy every... Perform a finite element analysis once for each working condition to obtain a structural response dataset under sparse finite element simulation;

[0017] A4. Divide the sparse simulation data to construct training and validation datasets. The training set accounts for 70% for training the GBRT-LSTM hybrid model, and the test set accounts for 30% for evaluating the model's accuracy.

[0018] Furthermore, step B specifically includes:

[0019] B1. The GBRT model calculates the negative gradient of the loss function for each sample as the residual. Based on these residuals, a new regression tree is constructed, dividing the input space into non-overlapping regions. And determine the optimal output value in each region. To reduce loss, the final spatial feature prediction result is obtained by adding the results of the previous iteration to the newly constructed decision tree. The sum of all regression tree outputs is calculated using the following formula:

[0020] (1)

[0021] In the formula, This represents the initial input data; The number of iterations; Number of partitions for the input space;

[0022] B2. The LSTM model selectively controls the forgetting of historical information and the input of new information through three gating units: the forget gate, the input gate, and the output gate. This updates the cell state to capture long-term dependencies in the time series. The output gate generates the hidden state at the current time step based on the updated cell state and iteratively propagates it along the time steps, ultimately obtaining the predicted value of the time feature. , can be represented as:

[0023] (2)

[0024] (3)

[0025] In the formula, This is a new cell state; It is in the old cell state; , and The outputs of the forget gate, input gate, and output gate; This is the weight matrix; This is the hidden state from the previous stage; This is the current input; For bias terms; Currently hidden;

[0026] B3. The weighted fusion layer will convert GBRT spatial predictions. Compared with LSTM time prediction value Dynamic fusion through learnable weights:

[0027] (4)

[0028] In the formula, This is the final prediction result; and For weighting; For bias terms;

[0029] B4. The optimization layer calculates the gradient of the loss function with respect to the fusion parameters through backpropagation, and uses the optimizer to iteratively update the parameters until the loss converges to its minimum:

[0030] (5)

[0031] In the formula, and These are the fusion parameters before and after the update, respectively; For learning rate, This is the residual.

[0032] Furthermore, step C specifically includes:

[0033] C1. The training set is input into the input layer of the GBRT-LSTM hybrid model for training. The input layer standardizes the original data and divides it into two types of sub-inputs: GBRT sub-input (single-time-step spatial feature vectors) and LSTM sub-input (multi-time-step sequence data).

[0034] C2. The two types of sub-inputs enter the two types of feature extraction layers respectively, and are then fused and optimized by the weighted fusion layer and the optimization layer, finally obtaining the complete structural response dataset in the output layer;

[0035] C3. Input the validation set into the trained GBRT-LSTM hybrid model for validation to obtain the final GBRT-LSTM hybrid model;

[0036] C4. The final GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response data of the unsimulated construction conditions. The input is the structural response of the construction conditions of the connecting truss part of the super high-rise building, and the output is the structural response of all construction conditions.

[0037] Furthermore, step D specifically includes:

[0038] D1. Entropy Weight Method (EWM) is based on the entropy value of data. Determine the weight of each spatial node , can be represented as:

[0039] (6)

[0040] (7)

[0041] In the formula, Indicates input data; The total number of data;

[0042] D2. TOPSIS utilizes We perform weighted aggregation of spatial dimension data on the structural response dataset, transforming the three-dimensional structural response dataset into a two-dimensional decision matrix, thereby achieving spatial dimension reduction.

[0043] D3. The Analytic Hierarchy Process (AHP) combines expert judgment to assess the relative importance of each indicator and constructs a judgment matrix for the structural response indicators. The system is then quantized using the scaling method, and the principal eigenvectors are calculated and normalized to obtain the weight vectors. Ensure random consistency ratio ;

[0044] D4. TOPSIS utilizes The weighted Euclidean distances to the positive and negative ideal solutions are calculated to achieve weighted aggregation of index dimension data, resulting in a comprehensive evaluation score for each construction condition. The construction condition with the highest comprehensive score is determined as the optimal installation time for the connecting truss of the super high-rise building.

[0045] Furthermore, step E specifically includes:

[0046] E1. Determine the optimal number of clusters using the elbow rule combined with the profile coefficient method. ;

[0047] E2. Use the K-means++ algorithm to randomly select a sample in the dataset as the initial center and iteratively distribute the sample, updating the cluster mean until the center moves less than 0.001 or the iteration exceeds 100 times;

[0048] E3. The category with the highest overall score at the cluster center is determined as the optimal installation time window for the connecting trusses of super high-rise buildings.

[0049] The working principle of this invention is as follows: It solves efficiency problems by combining sparse simulation with surrogate model prediction, and ensures reliability through multi-index decision-making. Sparse simulation is used to obtain structural response data for some operating conditions, reducing initial computation. A GBRT-LSTM hybrid model is then constructed, using the GBRT algorithm to capture spatial correlations and the LSTM algorithm's three-gating mechanism to capture time dependencies. Weighted fusion and parameter optimization form a high-precision surrogate model, replacing the finite element simulation prediction of unsimulated operating conditions and completing the sparse structural response dataset. Subsequently, EWM, AHP, and TOPSIS algorithms are combined to calculate the distance between each operating condition and the positive and negative ideal solutions, obtaining a comprehensive score based on multiple indicators to lock in the optimal node. Finally, the K-means++ algorithm is used to group operating conditions with similar scores into one category, and the category with the highest cluster center score is the optimal time window, ultimately achieving rapid and accurate determination of the installation timing.

[0050] The innovation of this invention is:

[0051] (1) Construct a GBRT-LSTM hybrid model to allow GBRT and LSTM to capture the spatial correlation and temporal dependence of data in parallel, and effectively capture the spatiotemporal correlation of data through weighted fusion and parameter optimization.

[0052] (2) A strategy of “sparse simulation + model prediction” is proposed, which only performs finite element simulation on some construction conditions and uses the trained GBRT-LSTM model to predict the response of unsimulated conditions, thereby reducing the amount of computation while ensuring data accuracy.

[0053] (3) The TOPSIS algorithm combines EWM and AHP to determine the optimal time node through multi-index quantification, and the K-means++ algorithm clusters the scores to obtain the time window, breaking through the limitation of relying on a single index.

[0054] The advantages of this invention are:

[0055] (1) The prediction error index of the GBRT-LSTM hybrid model is better than that of the LSTM model. It can more accurately predict the structural response of the unsimulated working conditions, and the prediction data is closer to the real finite element simulation results.

[0056] (2) The GBRT-LSTM hybrid model is used as a proxy model to replace finite element simulation with data-driven prediction, which significantly improves computational efficiency compared with the finite element simulation method.

[0057] (3) The TOPSIS and K-means++ algorithms quantify and evaluate different construction conditions and cluster the evaluation results to determine the optimal installation time node and window for the connecting truss, making the evaluation more comprehensive and reliable. Attached Figure Description

[0058] Figure 1 This is a flowchart of the method of the present invention;

[0059] Figure 2 This is a graph comparing the computational efficiency of different methods;

[0060] Figures 3a-3c This is a comparison chart of data prediction metrics from different methods. Figure 3a This is a comparison chart of the MAE indicator. Figure 3b This is a comparison chart of the RMSE indicator. Figure 3c This is a comparison chart of MAPE indicators;

[0061] Figure 4a This is a comparison chart of scores from different evaluation methods. Figure 4b This is a comparison chart of the combined scoring errors of different methods;

[0062] Figure 5 This is a schematic diagram showing the classification results of truss installation timing. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings and an embodiment.

[0064] like Figure 1This example relates to a method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model, specifically including the following steps:

[0065] A. A sparse simulation strategy is employed to run finite element models under certain construction conditions, establishing a sparse structural response dataset, specifically including:

[0066] A1. Based on the design and construction drawings of the South Taihu CBD Building, a three-dimensional refined finite element model of the super high-rise building was established in finite element software;

[0067] A2. The connecting truss is located on the 20th floor of the building. Starting from the 20th floor, the construction of the main tower core tube is defined as a construction condition for each floor above the building, until the main tower core tube is capped at the 68th floor. A total of 49 construction conditions for the installation of the connecting truss of the super high-rise building are set.

[0068] A3. Using a sparse simulation strategy, finite element analysis was performed every 4 working conditions, for a total of 13 finite element models under construction conditions, to obtain the structural response dataset under sparse finite element simulation.

[0069] A4. Divide the sparse simulation data to construct training and validation datasets. The training set accounts for 70% for training the GBRT-LSTM hybrid model, and the test set accounts for 30% for evaluating the model's accuracy.

[0070] B. Constructing a GBRT-LSTM hybrid model to effectively capture the spatiotemporal features of the dataset. Its network structure consists of: an input layer, a feature extraction layer, a weighted fusion layer, an optimization layer, and an output layer, specifically including:

[0071] B1. The GBRT model calculates the negative gradient of the loss function for each sample as the residual. Based on these residuals, a new regression tree is constructed, dividing the input space into non-overlapping regions. And determine the optimal output value in each region. To reduce loss, the final spatial feature prediction result is obtained by adding the results of the previous iteration to the newly constructed decision tree. This is the sum of the outputs of all regression trees;

[0072] B2. The LSTM model selectively controls the forgetting of historical information and the input of new information through three gating units: the forget gate, the input gate, and the output gate. This updates the cell state to capture long-term dependencies in the time series. The output gate generates the hidden state at the current time step based on the updated cell state and iteratively propagates it along the time steps, ultimately obtaining the predicted value of the time feature. ;

[0073] B3. The weighted fusion layer will convert GBRT spatial predictions. Compared with LSTM time prediction value Dynamic fusion through learnable weights;

[0074] B4. The optimization layer calculates the gradient of the loss function with respect to the fusion parameters through backpropagation, and uses the optimizer to iteratively update the parameters until the loss converges to the minimum value.

[0075] C. The sparse structural response dataset is input into the model for training and validation. The final GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response of unsimulated construction conditions, thus completing the structural response dataset. Compared with the finite element simulation method, this significantly improves computational efficiency. Figure 2 As shown, it specifically includes:

[0076] C1. The training set is input into the input layer of the GBRT-LSTM hybrid model for training. The input layer standardizes the original data and divides it into two types of sub-inputs: GBRT sub-input (single-time-step spatial feature vectors) and LSTM sub-input (multi-time-step sequence data).

[0077] C2. The two types of sub-inputs enter the two types of feature extraction layers respectively, and are then fused and optimized by the weighted fusion layer and the optimization layer, finally obtaining the complete structural response dataset in the output layer;

[0078] C3. Input the validation set into the trained GBRT-LSTM hybrid model for validation to obtain the final GBRT-LSTM hybrid model;

[0079] C4. The final GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response data of unsimulated construction conditions. The input is the structural response of the connecting truss section of the super high-rise building during construction, and the output is the structural response of all construction conditions. Compared with the LSTM model, it has higher prediction accuracy, such as... Figures 3a-3c As shown.

[0080] D. The TOPSIS algorithm is used to perform multi-index analysis on the dataset to obtain the comprehensive score for each construction condition and determine the optimal installation time node, specifically including:

[0081] D1. EWM is based on data entropy. Determine the weight of each spatial node ;

[0082] D2. TOPSIS algorithm utilizes We perform weighted aggregation of spatial dimension data on the structural response dataset, transforming the three-dimensional structural response dataset into a two-dimensional decision matrix, thereby achieving spatial dimension reduction.

[0083] D3. AHP combines expert judgment to assess the relative importance of each indicator and constructs a judgment matrix for structural response indicators. The system is then quantized using the scaling method, and the principal eigenvectors are calculated and normalized to obtain the weight vectors. Ensure random consistency ratio ;

[0084] D4. TOPSIS algorithm utilizes The weighted Euclidean distances to the positive and negative ideal solutions are calculated to achieve weighted aggregation of index dimension data, yielding a comprehensive evaluation score for each construction condition. The construction condition with the highest comprehensive score is determined as the optimal installation time for the connecting truss of the super high-rise building, such as... Figures 4a-4b As shown.

[0085] E. The K-means++ algorithm is used to perform cluster analysis on the comprehensive score to determine the optimal installation time window, specifically including:

[0086] E1. Determine the optimal number of clusters using the elbow rule combined with the profile coefficient method. ;

[0087] E2. Use the K-means++ algorithm to randomly select a sample in the dataset as the initial center and iteratively distribute the sample, updating the cluster mean until the center moves less than 0.001 or the iteration exceeds 100 times;

[0088] E3. The category with the highest overall score at the cluster center is determined as the optimal installation time window for the connecting trusses of super high-rise buildings, such as... Figure 5 As shown.

[0089] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model, comprising the following steps: A. Use a sparse simulation strategy to run finite element models under some construction conditions and establish a sparse structure response dataset; B. Construct a GBRT-LSTM hybrid model to capture the spatiotemporal features of the dataset. Its network structure includes a single-connected input layer, a feature extraction layer, a weighted fusion layer, an optimization layer, and an output layer. in, The input layer standardizes the raw data and divides it into two sub-inputs: GBRT sub-input and LSTM sub-input; the feature extraction layer includes GBRT and LSTM sub-layers, which work in parallel to extract the spatial correlation and temporal dependency of the data respectively; the weighted fusion layer weights and fuses the spatiotemporal feature prediction values ​​to generate the final prediction value; the optimization layer quantifies the prediction error and optimizes the parameters. C. Input the sparse structural response dataset into the model for training and validation. The final GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response of the unsimulated construction conditions and complete the structural response dataset. D. The TOPSIS algorithm, which approximates the ideal solution ranking algorithm, is used to perform multi-index analysis on the dataset to obtain the comprehensive score of each construction condition and determine the optimal installation time node; E. Adopt K The -means++ algorithm performs cluster analysis on the comprehensive score to determine the optimal installation time window.

2. The method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model as described in claim 1, characterized in that, Step A specifically includes: A1. Based on the construction drawings, establish a detailed three-dimensional finite element model of the super high-rise building in finite element software; A2. Starting from the floor where the connecting truss is located, the construction of the main tower core tube is defined as a construction case for each floor upwards until the top is reached; A3. Employing a sparse simulation strategy every... Perform a finite element analysis once for each working condition to obtain a structural response dataset under sparse finite element simulation; A4. Divide the sparse simulation data to construct training and validation datasets. The training set accounts for 70% and is used for training the GBRT-LSTM hybrid model, while the test set accounts for 30% and is used for evaluating the accuracy of the hybrid model.

3. The method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model as described in claim 1, characterized in that, Step B specifically includes: B1. The GBRT model calculates the negative gradient of the loss function for each sample as the residual. Based on these residuals, a new regression tree is constructed, dividing the input space into non-overlapping regions. And determine the optimal output value in each region. To reduce loss; by adding the results of the previous iteration to the newly constructed decision tree, the final spatial feature prediction results are improved. This is the sum of the outputs of all regression trees; B2. The LSTM model selectively controls the forgetting of historical information and the input of new information through three gating units: the forget gate, the input gate, and the output gate. This updates the cell state to capture long-term dependencies in the time series. The output gate generates the hidden state at the current time step based on the updated cell state and iteratively propagates it along the time steps, ultimately obtaining the predicted value of the time feature. ; B3. The weighted fusion layer will convert the GBRT spatial prediction values. Compared with LSTM time prediction value Dynamic fusion through learnable weights; B4. The optimization layer calculates the gradient of the loss function with respect to the fusion parameters through backpropagation, and uses the optimizer to iteratively update the parameters until the loss converges to the minimum value.

4. The method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model as described in claim 1, characterized in that, Step C specifically includes: C1. The training set is input into the input layer of the GBRT-LSTM hybrid model for training. The input layer standardizes the original data and divides it into GBRT sub-input and LSTM sub-input. The GBRT sub-input is a single-time-step spatial feature vector, and the LSTM sub-input is multi-time-step sequence data. C2. The two types of sub-inputs enter the two types of feature extraction layers respectively, and are then fused and optimized by the weighted fusion layer and the optimization layer, finally obtaining the complete structural response dataset in the output layer; C3. Input the validation set into the trained GBRT-LSTM hybrid model for validation to obtain the final GBRT-LSTM hybrid model; C4. The final GBRT-LSTM hybrid model is used as a surrogate model to predict the structural response data of the unsimulated construction conditions. The input is the structural response of the construction conditions of the connecting truss part of the super high-rise building, and the output is the structural response of all construction conditions.

5. The method for determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model as described in claim 1, characterized in that, Step D specifically includes: D1. EWM is based on data entropy. Determine the weight of each spatial node ; D2. TOPSIS algorithm utilizes We perform weighted aggregation of spatial dimension data on the structural response dataset, transforming the three-dimensional structural response dataset into a two-dimensional decision matrix, thereby achieving spatial dimension reduction. D3. AHP combines expert judgment to assess the relative importance of each indicator and constructs a judgment matrix for structural response indicators. The system is then quantized using the scaling method, and the principal eigenvectors are calculated and normalized to obtain the weight vectors. Ensure random consistency ratio ; D4. TOPSIS algorithm utilizes The weighted Euclidean distances to the positive and negative ideal solutions are calculated to achieve weighted aggregation of index dimension data, resulting in a comprehensive evaluation score for each construction condition. The construction condition with the highest comprehensive score is determined as the optimal installation time for the connecting truss of the super high-rise building.

6. The method for rapidly determining the installation timing of connecting trusses in super high-rise buildings based on a GBRT-LSTM hybrid model as described in claim 1, characterized in that, Step E specifically includes: E1. Determine the optimal number of clusters using the elbow rule combined with the profile coefficient method. ; E2. Adopt K The -means++ algorithm randomly selects a sample in the dataset as the initial center and iteratively distributes samples, updating the cluster mean until the center moves a distance of less than 0.001 or the iteration exceeds 100 times; E3. The category with the highest overall score at the cluster center is determined as the optimal installation time window for the connecting trusses of super high-rise buildings.

Images