Steel structure intelligent type selection method based on parameterization and neural network fusion

Abstract

The invention provides a steel structure intelligent type selection method based on parameterization and neural network fusion. A large amount of steel structure scheme data is rapidly generated through parametric modeling, and key indexes such as safety and steel consumption are obtained. The method is not only suitable for a linear structure, but also can perform model selection for an L-shaped structure and a U-shaped structure, and the parameterization process comprises shape parameterization. An improved BP neural network model is utilized, structural safety prediction is used as a classification task, steel consumption prediction is used as a regression task for training, and rapid and high-precision prediction of a new scheme is achieved. On the premise that finite element analysis does not need to be frequently called, the model can conduct multi-round iterative screening and evaluation on the scheme, and therefore the optimal structure scheme considering safety and economical efficiency is efficiently obtained. According to the method, the efficiency and accuracy of steel structure model selection are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of building structure engineering and intelligent design technology, and more specifically, to an intelligent selection method for assembled steel structures based on parametric modeling and intelligent neural networks and a platform thereof, which is suitable for various structural layout forms such as straight-line, L-type and U-type. Background Art

[0002] For steel structure design, structural selection is a key step to ensure structural safety, economy and construction efficiency. However, traditional structural selection methods are highly dependent on finite element analysis, which requires repeated construction and modification of models. The process is time-consuming and labor-intensive, and requires high professional experience and analytical ability of designers. In addition, finite element analysis is often difficult to respond quickly to multiple rounds of scheme iterations, which restricts the design efficiency and optimization speed of engineering practice. With the rapid development of machine learning and artificial intelligence technology, introducing intelligent algorithms into structural selection has become an efficient and cutting-edge solution. At present, the existing steel structure selection models still have the following limitations when dealing with high-dimensional parameters and multi-objective optimization problems:

[0003] 1. Low efficiency of data generation and processing: It is difficult to quickly generate multi-scheme data in batches that meet security requirements.

[0004] 2. Insufficient model accuracy and generalization ability: When multiple factors such as safety indicators (such as inter-story displacement angle, component stress ratio) and economic indicators (steel consumption) are considered simultaneously, it is difficult to achieve high-precision prediction and rapid decision-making.

[0005] 3. Over-reliance on finite element analysis: Traditional methods repeatedly call for finite element analysis, resulting in high time costs and low efficiency.

[0006] Therefore, how to integrate parametric modeling with intelligent neural networks to achieve rapid generation of multiple steel structure schemes, high-precision prediction of key indicators and efficient selection, so as to reduce dependence on finite element analysis and improve design efficiency has become a technical problem that needs to be solved urgently. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide an intelligent steel structure selection method based on the fusion of parameterization and neural network. The method generates a large number of steel structure scheme data through automated parametric modeling, and uses a neural network model to classify and regressively predict structural safety and steel consumption, thereby reducing reliance on traditional finite element repeated calculations, improving the efficiency and accuracy of selection decisions, and quickly providing engineers with feasible structural selection solutions.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0009] An intelligent steel structure selection method based on parameterization and neural network fusion includes the following steps:

[0010] (1) Based on the given building layout (including straight, L-shaped, and U-shaped), component section combination parameters, and building structure parameter range, a number of steel structure schemes are randomly generated using structural analysis software; finite element analysis and design verification are performed on the generated steel structure schemes, and key characteristic data of the steel structure are extracted in batches. The key characteristic data include inter-story displacement angle, component stress ratio, deflection, and steel consumption data;

[0011] (2) Data preprocessing is performed on the extracted key feature data of the steel structure, including data cleaning, denoising, normalization, and feature importance ranking, to improve the quality of the input data of the neural network model;

[0012] (3) Using the preprocessed data set, a neural network model is constructed and trained: structural safety prediction is used as a classification task to train the model, and steel consumption prediction is used as a regression task to train the model;

[0013] (4) Based on the trained neural network model, the safety judgment and steel consumption prediction values ​​are quickly output when the input parameters are given, realizing the rapid selection and evaluation of new solutions.

[0014] The beneficial effects of the present invention are:

[0015] 1. Reduce calculation dependence: significantly reduce the dependence on a large number of repeated finite element calculations and improve the calculation efficiency of structure selection.

[0016] 2. High-precision prediction: Use the neural network model to make high-precision predictions on safety and steel usage indicators, and quickly screen and generate economical selection plans that meet safety requirements.

[0017] 3. Multi-form applicability: The method is applicable to various steel structure layout forms such as I-shaped, L-shaped and U-shaped, and has wide practicality and scalability.

[0018] 4. Data-driven platform: Through data automated modeling and neural network training, it provides an intelligent, data-driven steel structure selection platform.

[0019] Through the BP neural network model optimized by batch stochastic gradient descent, the accuracy of safety classification can reach 85.5%, and the steel consumption prediction error is within 10%, which significantly improves the efficiency and accuracy of steel structure selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 The overall process of the present invention is demonstrated, including parameterized modeling, data feature extraction, neural network model training and selection decision process.

[0022] Figure 2 It is a plan view of a typical steel structure layout in the present invention, showing the relationship between input parameters (such as column spacing, beam span, floor height and cross-sectional dimensions) and the overall structural layout: where: a is a straight-line layout; b is an L-shaped layout; c is a U-shaped layout.

[0023] Figure 3 The figure is a schematic diagram showing the performance comparison between the improved BP neural network in the present invention and the traditional BP neural network in terms of security classification accuracy; wherein: a is the security classification accuracy of the traditional BP neural network; b is the security classification accuracy of the BP neural network based on batch stochastic gradient descent (msgd); c is the security classification accuracy of the BP neural network based on the momentum method.

[0024] Figure 4 This is an error analysis diagram for predicting the amount of steel used using the neural network model in the present invention: wherein: a is a loss function trend diagram for the training set and the validation set; b is a model iteration number-accuracy trend diagram.

[0025] Figure 5 This is the comparison result between SAP2000 analysis and neural network model prediction, where: a is the horizontal deformation diagram; b is the rod stress diagram; c is the vertical deformation diagram. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Figure 1The overall process of the present invention is presented, which includes steps such as parametric modeling, feature engineering, and machine learning model training. Model training is divided into two tasks: a classification task to predict structural safety (whether safety requirements are met), and a regression task to predict the amount of steel used. Through feature engineering, the extracted input features are standardized and divided into two groups of input data sets. The performance of the model is comprehensively compared through prediction accuracy, training time efficiency, and error distribution, and finally the optimal machine learning model suitable for steel structure selection is selected.

[0028] The present invention provides a steel structure intelligent selection method based on parameterization and neural network fusion, comprising the following steps:

[0029] 1. Parametric modeling and data acquisition:

[0030] Use SAP2000 for secondary development to achieve automated modeling and parametric analysis of steel structures.

[0031] A large number of steel structure schemes are randomly generated within the range of the total length L, total width W, number of floors N and component cross-sectional dimensions (such as H-beam cross-sectional height and flange thickness) of the given building. The safety indicators such as inter-layer displacement angle, component stress ratio, beam deflection, etc. of each scheme are extracted through finite element calculation, and the corresponding steel consumption is recorded. These data are organized into a training sample set, including input features (such as L, W, N, column cross-sectional parameters, beam cross-sectional parameters) and output indicators (safety labels, steel consumption values).

[0032] 2. Neural network model construction and optimization:

[0033] The BP neural network is selected as the basic model and optimized: training strategies such as batch stochastic gradient descent and momentum are introduced to improve the convergence performance and training stability of the model.

[0034] By adjusting the neural network structure, such as increasing the number of hidden layer nodes, using appropriate activation functions and regularization methods, the generalization ability of the model can be improved. Safety prediction is used as a classification task (the output is a label of whether the safety standard is met), and steel consumption prediction is used as a regression task (the output is a continuous value). By training a large number of data samples, the model can quickly output safety judgments and steel consumption prediction values ​​when given input parameters.

[0035] 3. Verification and result comparison

[0036] The final recommended solution is verified by finite element verification. If the prediction result is far from the precise finite element analysis result, the training data set can be expanded or the network structure can be further optimized.

[0037] The beneficial effects of the present invention are:

[0038] (1) High efficiency: The present invention significantly reduces the reliance on multiple rounds of iterations of finite element analysis. Taking the plateau prefabricated steel structure building as an example, the present method can complete the scheme selection in just a few minutes through SAP2000 automatic modeling and neural network prediction, while the traditional method takes several hours.

[0039] (2) Accuracy: By predicting safety through classification tasks and steel usage through regression tasks, the model can quickly lock in the parameter combination that meets the safety constraints and optimize the economic indicators on this basis.

[0040] (3) Scalability: The present invention can realize structural selection of various layout forms such as straight line, L-shaped, U-shaped, etc. through parametric technology, and has wide practicality.

[0041] (4) Improve design efficiency: The selection results provided by the present invention can be used as the initial solution for subsequent high-precision structural optimization analysis, helping engineers complete structural design in a shorter time, improving design efficiency and reducing costs.

[0042] In summary, the present invention can quickly provide an initial solution that meets safety requirements and has better economic efficiency during the structural selection stage.

[0043] In an embodiment of the present invention, a plateau prefabricated steel structure building is taken as the object, and the building shape is a straight line, a floor height of 3000mm, a number of floors of 4, a bay of 3300mm, a number of bays of 10, a depth of 6000mm, 1500mm, 2200mm, 1500mm, 6000mm, and a depth of 5. The section combination parameters are retrieved from the section library obtained by secondary development. The composition layout parameters and the section combination retrieved from the section library together constitute the characteristic parameters representing the simplified dormitory building model. The layout form and the section combination are input as input features into the classification neural network model for safety assessment and the neural network model for steel consumption assessment, and the safety and steel consumption of each combination form are judged, and the results can be predicted immediately. As shown in the following table:.

[0044] Table 1

[0045]

[0046] In Table 1, the safety prediction of 0 means that the safety of the section combination has not passed, and 1 means that the safety has passed. According to the section combinations that have passed the safety, the minimum steel consumption is considered. Through the structure predicted by safety and steel consumption, the section combination of the column HW300X300X10X15, the longitudinal beam HW200X200X12X12, and the cross beam HW250X250X14X14 is selected for comparison between the SAP2000 analysis and the prediction results of the neural network model. Figure 5is the comparison result, where: a is the horizontal deformation diagram; b is the rod stress diagram; c is the vertical deformation diagram. Therefore, the safety index of this section combination has passed, which is consistent with the evaluation result of the momentum method optimization model based on BP neural network.

Claims

1. An intelligent steel structure selection method based on parameterization and neural network fusion, characterized in that: The following steps are involved: (1) Based on the given building layout, component cross-section combination parameters, and building structure parameter range, multiple sets of steel structure schemes are randomly generated using structural analysis software; finite element analysis and design verification are performed on the generated multiple sets of steel structure schemes, and key characteristic data of the steel structure are extracted in batches. The key characteristic data include inter-story displacement angle, component stress ratio, deflection, and steel consumption data; (2) Data preprocessing is performed on the extracted key feature data of the steel structure to improve the input data quality of the neural network model, including data cleaning, denoising, normalization, and feature importance ranking; (3) Use the preprocessed data to build a data set and train a prediction model. The structural safety prediction is used as a classification task to train the model, and the steel consumption prediction is used as a regression task to train the model. (4) Based on the trained prediction model, the safety judgment and steel consumption prediction values ​​are quickly output when the input parameters are given.

2. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (3), a comprehensive evaluation is conducted on the classification and regression models under different training strategies, including the accuracy of safety prediction and the time efficiency and error analysis of steel consumption prediction; the most suitable model for steel structure selection is selected based on the comprehensive comparison results.

3. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (4), if the prediction results show that the scheme does not meet the safety requirements, it is directly discarded; if it is safe but the steel consumption is too high, the cross-sectional dimensions and layout parameters are appropriately adjusted and predicted again until an optimal scheme that meets the comprehensive requirements of safety and economy is obtained.

4. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (4), the safety prediction result of the classification task is used as a mandatory constraint condition. When the safety meets the requirements, the scheme is iteratively adjusted and optimized based on the predicted value of steel consumption of the regression task, and finally a steel structure scheme that meets the engineering design requirements is output.

5. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (3), regularization technology and momentum method are introduced during model training to improve the generalization ability and convergence speed of the model and reduce the overfitting problem.

6. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: If there is a large gap between the final solution and the precise finite element analysis results, the model prediction accuracy and decision quality can be further improved by expanding the training data set or optimizing the network structure.

7. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 6 is characterized in that: Methods for optimizing the network structure include adjusting the number of hidden layer nodes, changing the activation function, and increasing the training data.

8. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: The data preprocessing in step (1) includes cleaning, denoising, normalization and feature importance ranking.

9. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (1), the building layout forms include straight-line, L-shaped and U-shaped; the building structure parameter range includes the total length L, total width W and number of floors N of the building.

10. The intelligent steel structure selection method based on parameterization and neural network fusion according to claim 1 is characterized in that: In step (1), the component section combination parameters include the steel section height and flange thickness.

Images