Blasting construction real-time optimization method for steel frame structure

Abstract

The invention discloses a blasting construction real-time optimization method for a steel frame structure, and the method comprises the steps: constructing a component damage effect prediction model and a steel frame structure mechanical response prediction model based on a graph neural network; predicting the damage effect of the component in the steel frame structure under the blasting impact by using the component damage effect prediction model, and judging a component damage mode according to a component damage judgment standard; predicting the mechanical response of the steel frame structure under the blasting impact by using a mechanical response prediction model of the steel frame structure in combination with the judgment result of the component damage mode; and in combination with a prediction result of the mechanical response prediction model of the steel frame structure, an optimal blasting construction scheme of the steel frame structure is obtained based on a genetic algorithm. The method is convenient to use, efficient in calculation and good in robustness.

Description

Technical Field

[0001] The present invention relates to the field of target damage assessment and decision optimization, and is a real-time optimization method for blasting construction oriented to steel frame structures. Background Art

[0002] In modern building demolition construction, the need for efficient and precise execution poses new challenges for optimizing blasting plans. As projects continue to grow in scale and complexity, there's an urgent need for blasting plan optimization methods that can respond quickly and flexibly to ensure precise control of collapsed structures. Therefore, optimizing blasting plans in real time based on the structural characteristics of the building and precisely controlling the blasting effects to ensure construction quality are crucial.

[0003] At present, there are studies on blasting construction optimization methods, such as the Chinese invention application "A method for optimizing over-excavation and under-excavation in tunnel blasting construction" (publication number: CN117195371A, publication date: 2023.12.08). This patent is based on a fully connected neural network model, establishes a mapping model of initial blasting parameters, rock quality assessment parameters and tunnel over-excavation and under-excavation parameters, and uses an optimization algorithm to obtain the optimal smooth blasting parameters. This method can address the problem of poor blasting results caused by uneven surrounding rock distribution, but the model is based on a basic fully connected structure, which has shortcomings in terms of applicability and robustness in complex working conditions. Research on the assessment of blasting construction in frame buildings includes the Chinese invention application "Method, System, and Electronic Equipment for Determining Building Collapse Due to Blasting" (Publication No.: CN116956589A, Publication Date: October 27, 2023). This method combines the dynamic response of reinforced concrete load-bearing columns in the building under blasting with the instability mechanism of the residual steel skeleton superstructure under gravity to provide a basis for determining whether the building will collapse under blasting loads. However, this research method is computationally complex and time-consuming, making it difficult to meet the real-time, accuracy, and adaptability requirements of modern construction sites. Summary of the Invention

[0004] The purpose of the present invention is to provide a real-time optimization method for blasting construction of steel frame structures, which is easy to use, efficient in calculation and robust.

[0005] The technical solution for achieving the purpose of the present invention is:

[0006] In a first aspect, a real-time optimization method for blasting construction of a steel frame structure is provided, comprising the following steps:

[0007] Based on graph neural networks, a component damage effect prediction model and a steel frame structure mechanical response prediction model are constructed;

[0008] The damage effect prediction model of components is used to predict the damage effect of components in steel frame structures under blasting impact, and the damage mode of components is determined according to the component damage judgment criteria.

[0009] Combined with the results of component damage mode determination, the mechanical response prediction model of steel frame structure is used to predict the mechanical response of the steel frame structure under blasting impact.

[0010] Combined with the prediction results of the steel frame structure mechanical response prediction model, the optimal blasting construction plan of the steel frame structure is obtained based on the genetic algorithm.

[0011] Furthermore, the steps for obtaining the optimized blasting construction plan for the steel frame structure include:

[0012] The blasting construction chromosome is defined as a blasting scenario vector, which is formed by connecting the encoded material properties, explosive equivalent, blasting location, and blasting impact load along one dimension.

[0013] Create an initial population and randomly generate several chromosomes;

[0014] Define the fitness evaluation function:

[0015] ,

[0016] in, is the displacement data at node i, H is the total height of the steel frame structure, is the vertical height at node i, is the total number of nodes in the steel frame structure, ( , , ) is the position coordinate of node i, 、 All nodes in the steel frame structure The maximum and minimum values ​​in the axis coordinates, and All nodes in the steel frame structure The maximum and minimum values ​​in the axis coordinates, Compensate empirical parameters for position weight;

[0017] Calculate the fitness of each chromosome based on the prediction results of the steel frame structure mechanical response prediction model;

[0018] Using the roulette wheel selection method, chromosomes were screened;

[0019] Reproduction of chromosomes through crossover and mutation to generate new populations;

[0020] Repeat the above steps until the set convergence conditions are met, and output the current population as the final population;

[0021] By decoding the chromosomes in the final population, the optimal blasting construction plan is obtained.

[0022] Furthermore, the range of explosive equivalent in the randomly generated chromosomes is between 0 and 10 tons of TNT equivalent, and the range of the blasting location is within the steel frame structure and within 10 meters of its periphery.

[0023] Furthermore, the steps of constructing the component damage effect prediction model include:

[0024] Construct a neural network model, the input of which is the explosion scene vector;

[0025] The neural network model is trained and optimized using the blasting impact characteristics and component damage database including training data sets and validation data sets to obtain a component damage effect prediction model.

[0026] Furthermore, the neural network model includes a graph autoencoder network layer, a multi-stream graph neural network layer, and a multi-stream merging prediction network layer:

[0027] In the encoding layer, the fully connected neural network and the ReLU activation function are cascaded to output the encoding result G n0 ;

[0028] In the multi-flow graph neural network layer, the multi-head attention mechanism in the Transformer encoder model is used. The three heads of the multi-head local attention mechanism are strain attention mechanism, stress attention mechanism and material property attention mechanism. The input is G n0 , the outputs are the graph node abstract strain vector G n1 , graph node abstract stress vector G n2 , graph node abstract material characteristic vector G n3 ;

[0029] In the multi-stream merging prediction network layer, the self-attention mechanism is used to integrate the G output of the multi-stream graph neural network layer. n1 , G n2 and G n3 , and then the component damage effect is obtained through a fully connected neural network and ReLU activation function.

[0030] Furthermore, the steps of constructing the steel frame structure mechanical response prediction model include:

[0031] Construct a graph convolutional neural network, the input of which is the blasting scene vector and the judgment results of the component damage mode;

[0032] The graph convolutional neural network is trained and optimized using the blasting impact characteristics and the steel frame structure mechanical response database including the training data set and the validation data set to obtain a steel frame structure mechanical response prediction model.

[0033] Furthermore, the graph convolutional neural network includes a spatial domain graph convolution layer, a depth graph convolution layer, and a dilated graph convolution layer;

[0034] In the spatial domain graph convolution layer, the graph convolution and ReLU activation functions are cascaded to output potential graph data ;

[0035] In the deep graph convolution layer, a deep structure is formed by stacking multiple ResGCN blocks, and the message passing mechanism of the graph convolution layer is used to connect each ResGCN block to obtain abstract features. ;

[0036] In the dilated graph convolution layer, by stacking multiple dilated graph convolution layers, the structural response data with the displacement of the framework node unit is obtained.

[0037] Furthermore, the steps of obtaining the database of blasting impact characteristics and component damage effects, including the training data set and the validation data set, and the database of blasting impact characteristics and mechanical response of steel frame structures include:

[0038] Set the steel frame structure characteristics according to the target task requirements; set the beam and column characteristics according to the design technical standards; the steel frame structure characteristics include the length, width, height, number of floors, and room layout of each floor; the beam and column characteristics include the beam and column size, cross-sectional shape, and material properties; material properties include elastic modulus, density, and Poisson's ratio;

[0039] Generate various geometric models of steel frame structures based on the structural characteristics of steel frames and beam-column features;

[0040] Set construction plans with different explosive equivalents and different blasting locations, and assign a unique identification code to each construction plan; the construction plan includes the explosive equivalent and blasting location;

[0041] Use meshing tools to discretize the steel frame structure and perform local mesh refinement at the blasting location;

[0042] Abaqus was used to simulate the discretized steel frame structure, and the damage effect data of the components in the steel frame structure and the mechanical response data of the steel frame structure were extracted from the simulation results.

[0043] Correlate the damage effect data of components in the steel frame structure with the corresponding construction plan, steel frame structure characteristics and beam and column characteristics; correlate the mechanical response data of the steel frame structure with the corresponding construction plan, steel frame structure characteristics and beam and column characteristics;

[0044] Extract material properties, explosive equivalent, blasting location, and blasting impact load under different construction schemes, perform data pruning, normalization, and downsampling encoding, and connect the encoded material properties, explosive equivalent, blasting location, and blasting impact load along one dimension to obtain a blasting scenario vector.

[0045] Obtain the blasting scenario vectors for each construction scheme in different blasting construction scenarios. Using the blasting scenario vectors as input features, the corresponding component damage effects and steel frame structure mechanical responses are respectively input into the library system as output results, thereby obtaining a set of blasting impact features and component damage effects, as well as a set of blasting impact features and steel frame structure mechanical responses.

[0046] The sets of blasting impact characteristics and component damage effects, as well as the sets of blasting impact characteristics and steel frame structure mechanical responses are divided respectively, to obtain a database of blasting impact characteristics and component damage effects, including a training data set and a validation data set, as well as a database of blasting impact characteristics and steel frame structure mechanical responses.

[0047] In a second aspect, a computer-readable storage medium storing one or more programs is also provided, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform the method as described above.

[0048] In a third aspect, an electronic device is also provided, comprising one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method described above.

[0049] Compared with the prior art, the present invention has the following significant advantages:

[0050] 1. Easy to use: In the real-time optimization method for blasting construction, the deep learning prediction model uses data consisting of construction plans and steel frame structural characteristics as input, and the corresponding structural mechanical response as output. This end-to-end deep learning model does not require complex professional knowledge or complex formula reasoning, and can directly predict the blasting impact effect, providing support for the iteration of construction plans.

[0051] 2. Computational efficiency: Unlike the traditional layer-by-layer reasoning method of finite element analysis and numerical simulation, the component damage effect prediction model and steel frame structure mechanical response prediction model used in this invention can directly predict the mechanical response results of the steel frame structure in milliseconds through comprehensive training;

[0052] 3. Good robustness: By establishing a multi-scenario steel frame structure blasting impact damage database, the present invention effectively trains the graph neural network model, achieving rapid learning and generalization capabilities for different explosive equivalents, different blasting locations, and different geometric features of steel frame structures. It also realizes the prediction of the mechanical response of the target steel frame structure under multi-scenario and multi-dimensional blasting impact, as well as the iterative optimization of subsequent construction plans.

[0053] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is the main flow chart of a real-time optimization method for blasting construction of steel frame structures according to the present invention.

[0055] Figure 2 It is a schematic diagram of the structure of the component blasting damage effect prediction model.

[0056] Figure 3 It is a flow chart for generating input data for the prediction model of mechanical response of steel frame structures.

[0057] Figure 4 It is a structural diagram of the mechanical response prediction model of steel frame structure. DETAILED DESCRIPTION

[0058] like Figure 1 As shown, the present invention provides a real-time optimization method for blasting construction of a steel frame structure, comprising the following steps:

[0059] S10. Numerical simulation of the mechanical response of steel frame structures: Use Abaqus to set the blasting impact information including the geometric dimensions, material properties, explosive equivalent, blasting location, boundary conditions, etc. of the steel frame structure; import the geometric model of the steel frame structure and divide the model into meshes, and use Abaqus software to simulate and calculate the mechanical response of the steel frame structure; use Python to batch generate the damage effect data of the components in the steel frame structure under different explosive equivalents and different blasting locations, as well as the mechanical response data of the steel frame structure.

[0060] Specifically, the numerical simulation steps of the mechanical response of the steel frame structure include:

[0061] S11. Steel frame structure geometric feature design: Based on the target task requirements, the basic characteristic parameters of the steel frame structure are set, including the length, width, and height parameters of the structure, as well as the number of floors and the room layout of each floor. According to the national civil engineering design technical standards, the characteristics of the beams and columns are set, including the size, cross-sectional shape, and material properties of the beams and columns, which mainly include elastic modulus, density, and Poisson's ratio.

[0062] S12. Generation of multiple frame structures: According to step S11, multiple frame structure geometric models are generated to cover possible frame design types;

[0063] S13. Construction scheme design: Design construction schemes with different explosive equivalents and different blasting locations. The blasting locations are arranged according to the geometric characteristics of the steel frame structure. Each scheme is assigned a unique identification code for subsequent processing;

[0064] S14. Steel frame structure mesh division: Based on the overall characteristic information of the steel frame structure, the steel frame structure is discretized using mesh division tools; based on the divided steel frame structure mesh, the mesh is locally refined at the blasting location;

[0065] S15. Simulation calculation: Abaqus is used to obtain the simulation calculation results of the steel frame structure after blasting impact based on the discretized steel frame structure grid;

[0066] S16. Simulation data collection: Based on the simulation calculation results of the steel frame structure, the damage effect data of the components in the steel frame structure and the mechanical response data of the steel frame structure are extracted respectively, wherein the mechanical response data includes information such as displacement, stress, and strain;

[0067] S17. Simulation and batch data processing: Based on the Python platform, write automated scripts to batch process multiple construction plans, including the equivalent amount of explosives and blasting locations. Automatically generate input files and call Abaqus to extract the damage data of the steel frame structure under each plan, including the damage effect data of the components in the steel frame structure and the mechanical response data of the steel frame structure.

[0068] Abaqus can perform various types of analyses, including static stress / displacement analysis, dynamic stress / displacement analysis, viscoelasticity / viscoelastic response analysis, heat conduction analysis, and annealing process analysis. This software is a commonly used simulation tool in this field and is state-of-the-art, so its description will not be repeated herein.

[0069] The Python platform is an efficient, easy-to-read and widely used programming language environment that supports writing automated scripts to process tasks in batches. This platform is a common mathematical tool in this field and is prior art, so it will not be repeated here.

[0070] S20. Establishment of a multi-scenario steel frame structure blasting impact damage database: Associating the component damage data and the mechanical response data of the steel frame structure with different construction schemes and the geometry and material information of the steel frame structure to obtain steel frame structure blasting impact damage data pairs, performing data feature extraction and fusion on the data, and constructing a multi-scenario steel frame structure blasting impact damage database containing a training data set and a verification data set by stacking the data.

[0071] Specifically, the steps of establishing the steel frame structure blasting impact damage database include:

[0072] S21. Data association: Associating the damage effect data of the components in the steel frame structure with the corresponding construction plan, component geometry, size, material properties and other parameters; associating the mechanical response data of the steel frame structure with the corresponding construction plan, steel frame structure geometry, size, material properties and other parameters;

[0073] Complete the preliminary association of data features to ensure that the subsequent prediction model can accurately predict the mechanical response of the steel frame structure based on the steel frame structure characteristics and construction plan.

[0074] S22. Multi-source data feature extraction and fusion: Extract multi-source data on material properties, explosive equivalent, blasting location, and blasting impact load under different construction plans, fuse multi-source data with heterogeneous features, and achieve effective integration of data of different types and dimensions;

[0075] The step of extracting and fusing multi-source data features in step S22 includes:

[0076] S221. Data pruning: Prune the data of material properties, explosive equivalent, blasting location, and blasting impact load to integrate heterogeneous features of different dimensions and magnitudes into the same dimension and magnitude.

[0077] S222. Data normalization: After data pruning is completed, the data needs to be normalized to the minimum and maximum values. This method scales all features to the range of 0 and 1. The calculation formula can be expressed as:

[0078] ,

[0079] In the formula Represents the original data, Represents the normalized data.

[0080] S223, Feature Extraction: Use a multi-layer convolutional neural network to downsample and encode the normalized data, remove redundant information, and extract the data's potential invariant features.

[0081] S224, multi-source data fusion: connect the encoded material properties, explosive equivalent, blasting location, and blasting impact load vectors along one dimension to obtain a multi-source information fused blasting scene vector;

[0082] S23. Data entry: For the construction plans involved in different blasting construction scenarios, each construction plan is sequentially subjected to the multi-source data fusion method (S22) to obtain the multi-source blasting scenario vector under the corresponding blasting construction plan. All construction plans are stacked and sequentially combined with the corresponding component damage effects and steel frame structure mechanical response data to form a multi-scenario multi-source information fused blasting scenario vector and component damage effect set, as well as a multi-scenario multi-source information fused blasting scenario vector and steel frame structure mechanical response set. These are entered into the library system to obtain a multi-scenario steel frame structure blasting impact damage database including a training data set and a verification data set.

[0083] When entering data into the database system, the blasting scenario vectors obtained by fusion of multi-scenario and multi-source information are used as input features, and the component damage effects and steel frame structure mechanical responses are respectively entered into the database system as output results. This results in a set of blasting impact features and component damage effects obtained by fusion of multi-scenario and multi-source information, as well as a set of blasting impact features and steel frame structure mechanical responses obtained by fusion of multi-scenario and multi-source information. These two sets are then divided into a training dataset and a validation dataset as needed, resulting in a multi-scenario steel frame structure blasting impact damage database comprising both the training and validation datasets.

[0084] S30. Component damage effect prediction model construction: Construct a neural network model structure consisting of a graph autoencoder network, a multi-stream graph neural network, and a multi-stream merging prediction network, and use the training data set and the validation data set to train and optimize the neural network model structure to obtain a component damage effect prediction model.

[0085] Specifically, the steps of constructing the component damage effect prediction model include:

[0086] S31. Neural network model structure construction: The neural network model structure is obtained by constructing a graph autoencoder network, a multi-stream graph neural network, and a multi-stream merging prediction network;

[0087] The neural network model structure construction step includes:

[0088] S311, graph self-encoding layer construction: through the multi-layer downsampling fully connected neural network, the ReLU activation function is introduced for nonlinear processing to construct the graph self-encoding layer. The encoding result obtained after the input data is calculated by this layer can be used express.

[0089] S312, Multi-flow graph neural network construction: Use the multi-head attention mechanism in the Transformer encoder model to output the abstract strain vector of the graph nodes. , graph node abstract stress vector , graph node abstract material characteristic vector .

[0090] S313, multi-stream merging prediction network construction: Use the self-attention mechanism to integrate the output vectors of different attentions in the multi-stream merging prediction network; splice the outputs of each attention together, and finally restore the information through the fully connected upsampling layer and ReLU function to obtain node information data with component unit displacement.

[0091] S32. Component damage effect prediction model training and structural tuning: Utilize the component damage effect training dataset and its verification dataset to train and optimize the graph neural network to obtain a component damage effect prediction model based on the graph neural network.

[0092] Through repeated pre-training and error analysis, the model's training parameters and graph processing layer structure are continuously tuned. The training parameters mainly include training batches, loss functions, and number of iterations. The graph processing layer structure tuning mainly focuses on the number of GN blocks and the number of nodes in each GN block, as well as the size of the hidden layer h in the attention mechanism. Figure 2 shown.

[0093] Specifically, the component damage effect prediction model training and structure optimization steps include:

[0094] S321. Model pre-training: Use the Adam algorithm to pre-train the model in a supervised manner with the help of prepared input data and output data.

[0095] The model parameters are randomly generated at the beginning of training. The backpropagation algorithm is needed to transfer the value of the loss function in reverse and calculate the gradient of the model parameters relative to the loss value to update the model parameters.

[0096] S322. Select training parameters: Monitor the model pre-training status, such as computing memory usage, GPU memory usage, and model convergence performance information. When model training cannot be carried out normally, reselect the model training parameters;

[0097] S323. Adjust the graph data processing layer structure: Input the validation set into the prediction model after iterative training, check the mean squared error (MSE) and mean absolute error (MAE) of the model output results, and determine whether both have converged and decreased.

[0098] S324. Repeat steps S341, S342 and S343 until the MSE and MAE converge and decrease.

[0099] In this example, the training parameters of the component damage effect prediction model finally optimized are: training batch size of 32, loss function of mean absolute error (MAE), and training iteration steps of 10,000 times; the processing layer uses a total of 8 GN blocks, each GN block contains 128 nodes, and the hidden layer h in the attention mechanism is 128.

[0100] S40. Component damage mode determination: Based on the output result of the component damage effect prediction model based on the graph neural network and according to the component damage determination standard, the component damage mode is determined.

[0101] The damage mode of components under the influence of blasting impact is judged, and the relevant results can assist in the establishment of a prediction model for the mechanical response mode of steel frame structures.

[0102] Specifically, the component damage mode determination step includes:

[0103] S41. Predicting component damage effects: The blasting scenario vector fused with multi-source information is input into the component damage effect prediction model based on the graph neural network, and the output is the component damage effect. Based on the safety standards and design specifications in actual engineering, the corresponding component damage judgment criteria are shown in Table 1.

[0104] Table 1 Damage mode discrimination table

[0105] Damage Mode Damage factor range Minor damage 0-1 Minor damage 1-10 Moderate damage 10-30 Severe injuries 30-60 Seriously injured 60-100

[0106] S42. Determine component damage mode: Determine the component damage mode based on the component damage effect according to the damage determination criteria. The component damage mode determination results include minor damage, slight damage, moderate damage, severe damage, and serious damage. These will serve as input features to assist in model training for the prediction model of the mechanical response mode of the steel frame structure. The process of generating input data for the mechanical response model of the steel frame structure is as follows: Figure 3 shown.

[0107] S50, constructing a mechanical response prediction model for a steel frame structure: the model is based on a graph convolutional neural network, and the training data set and the validation data set are used to train and optimize the prediction model to obtain a mechanical response prediction model for a steel frame structure;

[0108] Specifically, the steps of establishing the steel frame structure mechanical response prediction model include:

[0109] S51. Construction of graph convolutional neural network framework: By constructing spatial domain graph convolution layer, depth graph convolution layer and dilated graph convolution layer, a graph convolutional neural network framework is obtained.

[0110] Specifically, the graph convolutional neural network framework construction steps include:

[0111] S511, spatial domain graph convolution layer construction: Using the idea of ​​traditional convolutional network, construct a local neighborhood graph, and apply edge convolution on the connection between adjacent points to extract local geometric features. The ReLU activation function is introduced for nonlinear processing to ensure that the nonlinear characteristics of the original input data are effectively preserved. The potential graph data of the input data can be obtained through the spatial domain graph convolution layer. .

[0112] S512. Construction of deep graph convolutional layer: ResGCN is used as the backbone block of the graph convolutional network. A deep structure is formed by stacking multiple ResGCN blocks. The message passing mechanism of the graph convolutional layer is used to connect each ResGCN block to construct a deep graph convolutional layer.

[0113] In the depth graph convolution layer, the potential graph data obtained by the spatial domain graph convolution layer Iterative processing is performed through multiple ResGCN structures with the same structure to achieve in-depth analysis and feature extraction of data; each ResGCN structure iteration includes the update of two key information, namely edge information and point information, and finally obtains the abstract features .

[0114] S513, dilated graph convolution layer construction: adopt a segmentation network based on dilated convolution, use dilated convolution operation to increase the receptive range to fully extract local features. The input of the dilated graph convolution layer is the abstract features output by the depth graph convolution layer. , after calculation and restoration, the structural response data with frame node unit displacement can be obtained.

[0115] S514, Graph Convolutional Neural Network Framework Synthesis: Figure 4 As shown in the figure, a graph convolutional neural network framework is formed by splicing the spatial domain graph convolution layer, the depth graph convolution layer and the dilated graph convolution layer.

[0116] S52. Steel frame structure mechanical response prediction model training and structural tuning: Use the steel frame structure mechanical response training dataset and its verification dataset to train and optimize the model to obtain a steel frame structure mechanical response prediction model based on graph neural network.

[0117] Through repeated pre-training and error analysis, the model's training parameters and model structure are continuously tuned. The training parameters mainly include training batches, loss functions, number of iterations, number of layers of the graph convolutional neural network, node aggregation method, and convolution method.

[0118] Specifically, the model training and structure tuning steps include:

[0119] S521. Model pre-training: Use the Adam algorithm to pre-train the model in a supervised manner with the help of prepared input data and output data.

[0120] The model parameters are randomly generated at the beginning of training. The backpropagation algorithm is needed to transfer the value of the loss function in reverse and calculate the gradient of the model parameters relative to the loss value to update the model parameters.

[0121] S522, select training parameters: monitor the model pre-training status, such as computing memory usage, graphics processor memory usage, and model convergence performance information. When model training cannot be carried out normally, reselect the model training parameters;

[0122] S523. Adjust the graph data processing layer structure: Input the validation set into the prediction model after iterative training, check the mean square error (MSE) and mean absolute error (MAE) of the model output results, and determine whether both have converged and decreased.

[0123] S524. Repeat steps S521, S522, and S523 until the MSE and MAE converge and decrease, and until the error evaluation standard is met, thereby obtaining an optimized steel frame structure mechanical response prediction model.

[0124] In this example, the training parameters of the finally optimized steel frame structure mechanical response prediction model are: training batch size is 32, loss function is Mean Absolute Error (MAE), and training iteration steps are 10,000 times; the number of layers of the graph convolutional neural network is 8, the node aggregation method is weighted aggregation, and the convolution method is cross convolution.

[0125] S60. Optimization of blasting construction plan: Use genetic algorithm to optimize the blasting construction plan, define blasting impact chromosomes, constraints, and evaluation functions, combine the component damage effect prediction model and the steel frame structure mechanical response prediction model, establish a genetic algorithm, and obtain the optimal blasting construction plan for the steel frame structure through multiple genetic algorithm iterations.

[0126] Specifically, the blasting construction plan optimization steps include:

[0127] S61. Definition of blasting impact chromosome: In this optimization scheme, a chromosome (hereinafter referred to as chromosome vector for ease of understanding) will contain two genes: one represents the equivalent amount of explosives, and the other represents the blasting position.

[0128] The blasting scene vector formed after step S22 is used as the chromosome vector, the explosive equivalent segment code in the blasting scene vector is used as the genetic code of the explosive equivalent, and the blasting position segment code is used as the genetic code of the blasting position.

[0129] S62. Initial population creation: Create an initial population with a population size of 100; randomly generate 100 chromosome vectors, that is, the explosive equivalent segment and blasting location segment in the blasting scene vector are randomly generated; the range of explosive equivalent is between 0 and 10 tons of TNT equivalent, and the range of blasting location is within the steel frame structure and within 10 meters of its periphery.

[0130] S63, fitness evaluation function definition: perform weighted summation on the displacement data at each node unit of the frame. The larger the weighted summation value, the closer it is to the optimal solution. The fitness evaluation formula for the weighted summation is:

[0131] ,

[0132] Among them, the displacement data at the node element is , the total height of the steel frame structure is , the vertical height from the ground at the node element , is the total number of nodes in the frame structure, and the node unit position coordinates are ( , , ), 、 is a node The maximum and minimum values ​​in the axis coordinates, and is a node The maximum and minimum values ​​in the axis coordinates, It is the position weight compensation parameter, which is summarized here after experiments. The value is 1.

[0133] This improvement to the fitness function assigns more weight to node displacements near the bottom of the structure, while giving it less weight, given that displacements at the top of the building have little impact on the damage caused by collapse. The addition of position weighting is based on the idea that blasting at the center of a frame structure will not increase fitness, while blasting near the edge of the frame structure will accelerate collapse due to loss of balance.

[0134] S64. Genetic fitness evaluation: The fitness function will evaluate the fitness of each chromosome based on the given explosive equivalent and blasting position.

[0135] The chromosome vector is input into the component damage effect prediction model to obtain the component damage effect; the component damage effect is combined with the component damage judgment standard to obtain the component damage mode, and the component damage mode is combined with the chromosome vector and input into the steel frame structure mechanical response prediction model to obtain the structural response data of the frame node unit displacement; the displacement data at each node unit of the frame are weighted summed to obtain the evaluation function value of the chromosome vector, which represents the fitness of the given chromosome vector.

[0136] S65. Efficient selection of chromosome vectors: Using the roulette wheel selection method, manually select 40 chromosome vectors from the top 40 chromosome vectors in descending fitness order. Then manually select 10 chromosome vectors from the bottom 60 chromosome vectors in descending fitness order. The total of 50 selected chromosome vectors are copied and expanded to the original number of 100. Chromosome vectors are efficiently screened using this method.

[0137] S66. Chromosome vector reproduction: The reproduction process includes crossover and mutation.

[0138] In the crossover step, the selected chromosome vectors are paired to create a new chromosome vector, and a crossover point is randomly selected using single-point crossover to exchange genes; the two genotypes in the chromosome vector are slightly changed with a probability of 1% during the mutation process;

[0139] S67, New population generation: Generate a new population using the chromosome vector after crossover and mutation;

[0140] S68, genetic iteration: repeat steps S63 to S68 until the fitness of the new population tends to be stable;

[0141] S69. Generation of the optimal blasting construction plan: By decoding the chromosome vectors in the new population whose fitness tends to be stable, the optimal blasting construction plan can be obtained.

[0142] Through the above steps, the real-time optimization of the blasting construction plan of the steel frame structure can be completed, which can provide more efficient and reliable decision support for the blasting construction.

[0143] Based on a graph neural network, the present invention utilizes a steel frame structure blasting damage database to train two prediction models. This model predicts the mechanical response of the steel frame structure under different blasting locations and with different explosive equivalents. Furthermore, a genetic algorithm combined with a deep learning prediction model is used for rapid iteration to obtain an optimized blasting construction plan. First, Abaqus finite element software is used to generate steel frame structure blasting damage data. A steel frame structure blasting impact damage database is established, and multi-source data feature extraction and fusion are performed on the data. The processed data in the database is then used to construct a dataset (training set and validation set) with a graph data structure. A component damage effect prediction model and a steel frame structure mechanical response prediction model are then sequentially constructed, and the model structure and parameters within the model are determined using the training set and validation set. Once the model training is complete, the graph neural network-based steel frame structure mechanical response prediction model is constructed. An initial blasting impact chromosome vector is generated, and the optimized chromosome vector is calculated through multiple iterations of the genetic algorithm. The chromosome vector is decoded to obtain the optimized blasting construction plan.

[0144] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute the above-mentioned real-time optimization method for blasting construction for steel frame structures.

[0145] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned real-time optimization method for blasting construction for steel frame structures.

[0146] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0150] The above embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A real-time optimization method for blasting construction of steel frame structure, characterized in that: The steps include: Based on graph neural network, a component damage effect prediction model and a steel frame structure mechanical response prediction model are constructed; The damage effect prediction model of the component damage effect is used to predict the damage effect of the components in the steel frame structure under the blasting impact, and the component damage mode is determined according to the component damage judgment standard; Combined with the results of component damage mode determination, the mechanical response prediction model of steel frame structure is used to predict the mechanical response of the steel frame structure under blasting impact. Combined with the prediction results of the steel frame structure mechanical response prediction model, the optimal blasting construction plan of the steel frame structure is obtained based on the genetic algorithm.

2. The method according to claim 1, characterized in that The steps for obtaining the optimal blasting construction plan for steel frame structures include: The blasting construction chromosome is defined as a blasting scenario vector, which is formed by connecting the encoded material properties, explosive equivalent, blasting location, and blasting impact load along one dimension. Create an initial population and randomly generate several chromosomes; Define the fitness evaluation function: ， in, is the displacement data at node i, H is the total height of the steel frame structure, is the vertical height at node i, is the total number of nodes in the steel frame structure, ( , , ) is the position coordinate of node i, , is the value of all nodes in the steel frame structure. The maximum and minimum values ​​in the axis coordinates, and is the value of all nodes in the steel frame structure. The maximum and minimum values ​​in the axis coordinates, Compensate empirical parameters for position weight; According to the prediction results of the steel frame structure mechanical response prediction model, the fitness of each chromosome is calculated; Using the roulette wheel selection method, chromosomes were screened; Reproduction of chromosomes through crossover and mutation to generate new populations; Repeat the above steps until the set convergence conditions are met, and output the current population as the final population; By decoding the chromosomes in the final population, the most optimized blasting construction plan can be obtained.

3. The method according to claim 1, characterized in that The explosive equivalent in the randomly generated chromosomes ranges from 0 to 10 tons of TNT equivalent, and the blasting location is within 10 meters inside the steel frame structure and outside it.

4. The method according to claim 1, characterized in that: The steps of building the component damage effect prediction model include: Construct a neural network model, the input of which is a blasting scene vector; The neural network model is trained and optimized using the blasting impact characteristics and component damage database including training data sets and validation data sets to obtain a component damage effect prediction model.

5. The method according to claim 4, characterized in that The neural network model includes a graph autoencoder network layer, a multi-stream graph neural network layer, and a multi-stream merge prediction network layer: In the image from the encoding layer, the fully connected neural network and the ReLU activation function are cascaded to output the encoding result G n0 ; In the multi-flow graph neural network layer, the multi-head attention mechanism in the Transformer encoder model is used. The three heads of the multi-head local attention mechanism are strain attention mechanism, stress attention mechanism and material property attention mechanism. The input is G n0 , and the outputs are the graph node abstract strain vector G n1 , graph node abstract stress vector G n2 、Graph node abstract material characteristic vector G n3 ; In the multi-stream merging prediction network layer, the self-attention mechanism is used to integrate the G output of the multi-stream graph neural network layer. n1 , G n2 and G n3 , and then the component damage effect is obtained through a fully connected neural network and ReLU activation function.

6. The method according to claim 1, characterized in that The steps for building a mechanical response prediction model for steel frame structures include: Construct a graph convolutional neural network, the input of which is the blasting scene vector and the judgment result of the component damage mode; The graph convolutional neural network is trained and optimized using the blasting impact characteristics and the steel frame structure mechanical response database including the training data set and the validation data set to obtain a steel frame structure mechanical response prediction model.

7. The method according to claim 6, characterized in that The graph convolutional neural network includes spatial domain graph convolution layer, depth graph convolution layer and dilated graph convolution layer; In the spatial domain graph convolution layer, the graph convolution and ReLU activation functions are cascaded to output the potential graph data ; In the deep graph convolution layer, multiple ResGCN blocks are stacked to form a deep structure, and the message passing mechanism of the graph convolution layer is used to connect each ResGCN block to obtain abstract features. ; In the dilated graph convolution layer, by stacking multiple dilated graph convolution layers, the structural response data with the displacement of the frame node unit is obtained.

8. The method according to claim 4 or 6, characterized in that: The steps for obtaining the blasting impact characteristics and component damage effect database including the training data set and the verification data set, as well as the blasting impact characteristics and steel frame structure mechanical response database include: According to the target task requirements, set the characteristics of the steel frame structure; according to the design technical standards, set the characteristics of the beam and column; the characteristics of the steel frame structure include the length, width, height, number of floors, and room layout of each floor; the characteristics of the beam and column include the size, cross-sectional shape, and material properties of the beam and column; the material properties include elastic modulus, density, and Poisson's ratio; Generate various geometric models of steel frame structures according to the characteristics of steel frame structures and beam and column characteristics; Set construction plans with different explosive equivalents and different blasting locations, and assign a unique identification code to each construction plan; the construction plan includes the explosive equivalent and the blasting location; The steel frame structure is discretized using meshing tools, and the mesh is locally refined at the blasting location; Abaqus is used to simulate the discretized steel frame structure, and the damage effect data of the components in the steel frame structure and the mechanical response data of the steel frame structure are extracted from the simulation results. Associating the damage effect data of the components in the steel frame structure with the corresponding construction scheme, steel frame structure characteristics and beam-column characteristics; associating the mechanical response data of the steel frame structure with the corresponding construction scheme, steel frame structure characteristics and beam-column characteristics; Extract material properties, explosive equivalent, blasting position and blasting impact load under different construction schemes, perform data pruning, normalization and down-sampling encoding, connect the encoded material properties, explosive equivalent, blasting position and blasting impact load along one dimension to obtain the blasting scene vector; The blasting scene vectors under each construction scheme in different blasting construction scenes are obtained, and the blasting scene vectors are used as input features. The corresponding component damage effects and steel frame structure mechanical responses are respectively entered into the library system as output results to obtain a set of blasting impact features and component damage effects, and a set of blasting impact features and steel frame structure mechanical responses; The sets of blasting impact characteristics and component damage effects, as well as the sets of blasting impact characteristics and steel frame structure mechanical responses are divided respectively, to obtain a blasting impact characteristic and component damage effect database including a training data set and a validation data set, as well as a blasting impact characteristic and steel frame structure mechanical response database.

9. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 8.

10. An electronic device, characterized in that: The method comprises one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as described in any one of claims 1 to 8.

Images