BIM model bridge structure optimization method based on neural network

Abstract

The invention provides a BIM (Building Information Modeling) bridge structure optimization method based on a neural network. The method comprises the following steps: establishing a bridge structure BIM; defining an optimization target and constraint conditions; generating a training data set; training a neural network model; fitting a curved surface; random forest screening of important design parameters; and the bridge structure is optimized. According to the method, the bridge structure can be better optimized, and the efficiency and quality of bridge structure design are improved.

Description

Technical Field

[0001] The present invention relates to the technical field, and particularly to a BIM model bridge structure optimization method based on a neural network. Background Art

[0002] Building Information Modeling (BIM) is a new tool in architecture, engineering, and civil engineering. The core of BIM is to establish a virtual three-dimensional model of a building project and use digital technology to provide a complete building project information database that is consistent with the actual situation. This information database not only contains geometric information, professional attributes, geometric dimensions, material attributes, load conditions, and status information describing building components, but also contains status information of non-component objects (such as spaces and movement behaviors). With the help of this three-dimensional model containing building project information, the degree of information integration of building projects is greatly improved, thus providing a platform for the exchange and sharing of project information for relevant stakeholders in building projects.

[0003] Bridge structure design is a complex process that requires considering various factors such as loads, materials, and construction. Traditional bridge structure design methods mainly rely on the experience and specifications of engineers, and have problems such as low efficiency, high cost, and difficulty in finding the optimal solution. Summary of the Invention

[0004] The present invention provides a BIM model bridge structure optimization method based on a neural network to better optimize bridge structures and improve the efficiency and quality of bridge structure design.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] This specification discloses a BIM model bridge structure optimization method based on a neural network, including:

[0007] Establish a BIM model of the bridge structure: Use BIM software to establish a geometric model, a material model, and a load model of the bridge structure;

[0008] Define the optimization objective and constraints: According to the design requirements of the bridge structure, define the optimization objective and constraints;

[0009] Generate a training data set: Use parametric modeling methods to generate a large number of BIM models of bridge structures with different design parameters, and use finite element analysis software to calculate the response values of each model to form a training data set;

[0010] Train a neural network model: Use the training data set to train the neural network model so that it can predict the response value of the bridge structure according to the input design parameters;

[0011] Surface fitting: Using the trained neural network model, a large number of predicted data points are generated, and a smooth response surface is fitted using the surface fitting method;

[0012] Random forest screening of important design parameters: Using the random forest method, the design parameters that have the greatest impact on the bridge structure response value are screened out;

[0013] Optimizing the bridge structure: Using the fitted response surface and the screened important design parameters, combined with the genetic algorithm and the particle swarm algorithm, the bridge structure is optimized to obtain a design scheme that meets the constraint conditions and has the smallest optimal objective value.

[0014] This specification discloses a BIM model bridge structure optimization method based on a neural network, including:

[0015] S1. Define the bridge design variable set through BIM parametric modeling, and generate the sample space using Latin hypercube sampling;

[0016] S2. Call the finite element solver to batch calculate the structural response data of the samples;

[0017] S3. Construct a deep neural network surrogate model, and train it with the design variables as the input and the structural response as the output;

[0018] S4. Use the random forest algorithm to evaluate the feature importance of the design variables and screen the key parameters;

[0019] S5. Based on the screened key parameters, construct a quadratic response surface model, and use the NSGA-II algorithm for multi-objective optimization to obtain the optimized parameters;

[0020] S6. Reverse map the optimized parameters to the BIM model, and output the final scheme through finite element secondary verification.

[0021] In this specification, the design scheme and the final scheme that meet the constraint conditions and have the smallest optimal objective value are modeled and compared in the BIM model, and the better scheme is selected.

[0022] In this specification, the deep neural network surrogate model described in S3 includes 3 hidden layers, the activation function is LeakyReLU, α = 0.01, and the loss function L is defined as:

[0023] L = 0.7MSE + 0.3MAE + 0.1‖W‖2;

[0024] MSE is the mean square error, MAE is the mean average error, and ‖W‖2 is the L2 regularization term.

[0025] In this specification, parameter screening is achieved through the Gini importance index in S4, the threshold θ = 0.15 is set, and the importance calculation formula is:

[0026]

[0027] I k is the feature importance score of the k-th design variable, T is the total number of decision trees in the random forest, is the reduction in impurity caused by splitting the k-th feature in the t-th tree, N t is the number of samples in the t-th tree, and N is the total number of samples.

[0028] In this specification, the quadratic response surface model described in S5 is:

[0029]

[0030] is the predicted structural response value; β0 is the intercept term; represents the baseline response value when all design variables are zero; β i is the linear term coefficient of the i-th design variable, reflecting the linear influence strength of this variable on the response; β ij is the interaction term coefficient between the i-th and j-th design variables, and is the quadratic term coefficient when i = j; x i is the i-th design variable, x j is the j-th design variable, p is the total number of key design variables after screening, determined by random forest feature importance analysis, p ≤ n, where n is the total number of original variables; ∈ is the model error term, following a normal distribution with a mean of zero;

[0031] Use ridge regression to fit the coefficients, normalize the design variables, and use the least squares method with L2 regularization to fit the coefficients:

[0032]

[0033] The regularization parameter λ = 0.01, X: the design matrix containing linear terms, interaction terms, and quadratic terms.

[0034] In this specification, the improved NSGA-II algorithm is used in S5, and an adaptive crossover probability is introduced:

[0035]

[0036] p c (g) is the crossover probability during the g-th generation of evolution, G max is the preset maximum number of generations of evolution, representing the maximum number of iterations of the optimization algorithm; g is the current generation of evolution, representing the number of iterations that the algorithm has executed.

[0037] In this specification, in the construction of the deep neural network proxy model, the number of neurons in the input layer is equal to the dimension of the design parameters. The hidden layer adopts a fully connected layer structure and applies the ReLU activation function. The output layer linearly maps to predict the structural response. The mathematical expression is as follows:

[0038] h1 = ReLU(W1X + b1);

[0039] h2 = ReLU(W2h1 + b2);

[0040]

[0041] ReLU is the rectified linear unit activation function. W1, W2, and W3 are weight matrices, and b1, b2, and b3 are bias terms. is the predicted structural response value, and h1 and h2 are transition parameters.

[0042] In this specification, the function of the multi-objective optimization is as follows:

[0043]

[0044] minf1(X) is the optimization objective function of the structural weight, minf2(X) is the optimization objective function of the peak equivalent stress, and X is the set of design variables; ρ is the material density; A i is the cross-sectional area of the i-th member, and L i is the length of the i-th member, σ um is the peak equivalent stress; u max is the maximum displacement response of the bridge structure, L is the main span of the bridge, and λ 屈曲 is the buckling safety factor.

[0045] In this specification, the termination condition of the improved NSGA-II algorithm is to reach the maximum number of generations G max , or the improvement rate of the Pareto front for 20 consecutive generations is <0.1%. The calculation formula for the improvement rate is as follows:

[0046]

[0047] is the average value of the objective function of the i-th generation population.

[0048] In summary, the present invention has at least the following beneficial effects:

[0049] 1. Improve design efficiency: By using neural network models, surface fitting methods, random forest methods, genetic algorithms, and particle swarm algorithms, the number of finite element analyses can be significantly reduced, and the efficiency of bridge structure optimization can be improved.

[0050] 2. Reduce design costs: By reducing the number of finite element analyses, the cost of bridge structure design can be lowered.

[0051] 3. Improve design quality: By leveraging the powerful prediction and optimization capabilities of neural network models, surface fitting methods, random forest methods, genetic algorithms, and particle swarm algorithms, a more optimal bridge structure design solution can be found. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 It is a schematic diagram of the BIM model bridge structure optimization method based on neural network involved in the present invention.

[0054] Figure 2 It is a schematic diagram of another BIM model bridge structure optimization method based on neural network involved in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In the following text, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0056] The following disclosure provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. To simplify the disclosure of the embodiments of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0057] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0058] As Figure 1 shown, this embodiment provides a BIM model bridge structure optimization method based on neural network, including:

[0059] Step 1: Establish a BIM model of the bridge structure

[0060] Use BIM software (such as Revit, Tekla, etc.) to establish geometric models, material models, load models, etc. of the bridge structure.

[0061] Step 2: Define the optimization objectives and constraints

[0062] According to the design requirements of the bridge structure, define the optimization objectives (such as minimizing weight, minimizing cost, etc.) and constraints (such as stress constraints, displacement constraints, etc.).

[0063] Step 3: Generate the training dataset

[0064] Using the parametric modeling method, generate a large number of BIM models of bridge structures with different design parameters, and use finite element analysis software (such as ANSYS, ABAQUS, etc.) to calculate the response values (such as stress, displacement, etc.) of each model to form a training dataset.

[0065] Step 4: Train the neural network model

[0066] Use the training dataset to train the neural network model (such as multi-layer perceptron, convolutional neural network, etc.) so that it can predict the response values of the bridge structure according to the input design parameters. Training process: Use the training dataset to train the neural network model, and improve the prediction accuracy of the model by adjusting parameters such as network structure, learning rate, activation function, etc.

[0067] Step 5: Surface fitting

[0068] Use the trained neural network model to generate a large number of predicted data points, and use the surface fitting method (such as polynomial fitting, spline fitting, radial basis function fitting, etc.) to fit a smooth response surface.

[0069] Step 6: Use random forest to screen important design parameters

[0070] Use the random forest method (such as classification random forest, regression random forest, etc.) to screen out the design parameters (such as beam height, beam width, concrete strength, etc.) that have the greatest impact on the response values of the bridge structure.

[0071] Step 7: Optimize the bridge structure

[0072] Use the fitted response surface and the screened important design parameters, combined with the genetic algorithm and particle swarm algorithm, to optimize the bridge structure and find a design scheme that meets the constraint conditions and has the smallest optimal objective value (optimization objective).

[0073] In some embodiments, genetic algorithm optimization: Use the genetic algorithm to perform preliminary optimization on the bridge structure, and generate a set of better design schemes through operations such as selection, crossover, and mutation.

[0074] In some embodiments, particle swarm optimization: The design scheme optimized by genetic algorithm is further optimized using particle swarm algorithm. By updating the particle position and velocity, the optimal design scheme is found.

[0075] In some embodiments, the optimization objectives include minimizing weight, maximizing stiffness, and minimizing cost.

[0076] In some embodiments, the constraint conditions include stress constraint, stability constraint, and displacement constraint.

[0077] In some embodiments, the response values include stress, vibration frequency, and displacement.

[0078] In some embodiments, a geometric model of the bridge is accurately established using BIM software, including bridge piers, abutments, and girders, to ensure the accuracy and integrity of the model.

[0079] In some embodiments, the material properties of each component of the bridge are defined in the BIM model, including concrete strength and steel bar grade, to ensure the accuracy of the material parameters.

[0080] In some embodiments, according to the design requirements of the bridge, various load conditions are defined, including dead load, live load, wind load, and seismic load, to ensure the comprehensiveness of the load model.

[0081] In some embodiments:

[0082] Taking a simply supported beam bridge with a span of 100 meters as an example, the optimization design is carried out using the method of the present invention.

[0083] 1. Establish a BIM model of the bridge structure: Use Revit software to establish a geometric model, material model, load model, etc. of the simply supported beam bridge.

[0084] 2. Define the optimization objectives and constraint conditions: The optimization objective is to minimize the weight of the bridge, and the constraint conditions are that the maximum stress does not exceed 200 MPa and the maximum displacement does not exceed 10 mm.

[0085] 3. Generate a training data set: Using parametric modeling method, generate 1000 BIM models of simply supported beam bridges with different design parameters, and use ANSYS software to calculate the response values such as stress and displacement of each model to form a training data set.

[0086] 4. Train a neural network model: Use the training data set to train a neural network model with 3 hidden layers, and each hidden layer contains 100 neurons.

[0087] 5. Surface fitting: Using the trained neural network model, generate 10000 predicted data points, and use polynomial fitting method to fit out a smooth response surface.

[0088] 6. Random forest screening of important design parameters: Using the random forest method, screen out the design parameters that have the greatest impact on the bridge structure response values, such as beam height, beam width, concrete strength, etc.

[0089] 7. Optimize the bridge structure: Utilize the fitted response surface and the screened important design parameters, combine the genetic algorithm and the particle swarm algorithm to optimize the simply supported beam bridge, and finally obtain an optimized design scheme with a weight of 500 tons, a maximum stress of 195 MPa, and a maximum displacement of 9.8 mm.

[0090] As Figure 2 shown, this embodiment discloses a BIM model bridge structure optimization method based on a neural network, including:

[0091] S1. Define the bridge design variable set through BIM parametric modeling, and generate the sample space by Latin hypercube sampling;

[0092] S2. Call the finite element solver to batch calculate the structural response data of the samples;

[0093] S3. Construct a deep neural network surrogate model, and train it with the design variables as the input and the structural response as the output;

[0094] S4. Use the random forest algorithm to evaluate the feature importance of the design variables and screen out the key parameters;

[0095] S5. Build a quadratic response surface model based on the screened key parameters, and use the NSGA-II algorithm for multi-objective optimization to obtain the optimized parameters;

[0096] S6. Inverse map the optimized parameters to the BIM model, and output the final scheme through finite element secondary verification.

[0097] In some embodiments, model the design scheme and the final scheme that meet the constraint conditions and have the minimum optimal objective value in the BIM model, and select the better scheme.

[0098] In some embodiments, the deep neural network surrogate model in S3 includes 3 hidden layers, the activation function is LeakyReLU, α = 0.01, and the loss function L is defined as:

[0099] L = 0.7MSE + 0.3MAE + 0.1‖W‖2;

[0100] MSE is the mean square error, MAE is the mean average error, and ‖W‖2 is the L2 regularization term.

[0101] In some embodiments, in S4, parameter screening is achieved through the Gini importance index, the threshold θ = 0.15 is set, and the importance calculation formula is:

[0102]

[0103] I k is the feature importance score of the k-th design variable, T is the total number of decision trees in the random forest, is the reduction in impurity caused by splitting the k-th feature in the t-th tree, N t is the number of samples in the t-th tree, and N is the total number of samples.

[0104] In some embodiments, the quadratic response surface model in S5 is:

[0105]

[0106] is the predicted structural response value; β0 is the intercept term; represents the baseline response value when all design variables are zero; β i is the linear term coefficient of the i-th design variable, reflecting the linear influence intensity of this variable on the response; β ij is the interaction term coefficient of the i-th and j-th design variables, and is the quadratic term coefficient when i = j; x i is the i-th design variable, x j is the j-th design variable, p is the total number of key design variables after screening, determined by random forest feature importance analysis, p ≤ n, where n is the total number of original variables; ∈ is the model error term, following a normal distribution with a mean of zero;

[0107] Use ridge regression to fit the coefficients, normalize the design variables, and use the least squares method with L2 regularization to fit the coefficients:

[0108]

[0109] The regularization parameter λ = 0.01, X: the design matrix containing linear terms, interaction terms, and quadratic terms.

[0110] In some embodiments, the improved NSGA-II algorithm is used in S5, and an adaptive crossover probability is introduced:

[0111]

[0112] p c (g) is the crossover probability during the g-th generation of evolution, G max is the preset maximum number of generations of evolution, representing the maximum number of iterations of the optimization algorithm; g is the current generation of evolution, representing the number of iterations that the algorithm has executed.

[0113] In some embodiments, in constructing the deep neural network surrogate model, the number of neurons in the input layer is equal to the dimension of the design parameters. The hidden layer adopts a fully connected layer structure and applies the ReLU activation function. The output layer linearly maps to predict the structural response. The mathematical expression is as follows:

[0114] h1 = ReLU(W1X + b1);

[0115] h2 = ReLU(W2h1 + b2);

[0116]

[0117] ReLU is the rectified linear unit activation function. W1, W2, and W3 are weight matrices, and b1, b2, and b3 are bias terms. is the predicted structural response value, and h1 and h2 are intermediate parameters.

[0118] In some embodiments, the function of multi-objective optimization is as follows:

[0119]

[0120] minf1(X) is the optimization objective function of the structural weight, minf2(X) is the optimization objective function of the peak equivalent stress, and X is the set of design variables; ρ is the material density; A i is the cross-sectional area of the i-th component, and L i is the length of the i-th component, and σ um is the peak equivalent stress; u max is the maximum displacement response of the bridge structure, L is the main span of the bridge, and λ 屈曲 is the buckling safety factor.

[0121] In some embodiments, the termination condition of the improved NSGA-II algorithm is to reach the maximum number of evolutionary generations G max , or the improvement rate of the Pareto front for 20 consecutive generations is < 0.1%. The calculation formula for the improvement rate is as follows:

[0122]

[0123] is the average value of the objective function of the i-th generation population.

[0124] In some embodiments, the parametric BIM model is established: Define the set of bridge design variables X = {x1, x2,..., x n} as the n-dimensional design space, including key parameters such as cross-sectional dimensions and prestressed tendon layout. Establish the constraint condition g j(X) ≤ 0 (such as stress limits, displacement limits) and a multi-objective optimization function min f(X) = α·C + β·D, where C is the construction cost, D is the maximum displacement response, and α, β are weight coefficients.

[0125] In some embodiments, the Latin Hypercube Sampling method is used to generate uniformly distributed sample points within the design space to ensure that each design parameter is stratified sampled within its value range. For a bridge model containing 5 key parameters, 1000 groups of sample points are generated to form a training dataset.

[0126] In some embodiments, the sample set is imported into the finite element analysis software through a parameter-driven interface, and the structural static and dynamic analysis is automatically performed to extract response data such as the stress field, displacement field, and buckling coefficient corresponding to each sample.

[0127] In some embodiments, the model training strategy for the deep neural network: The Adaptive Moment Estimation (Adam) optimizer is used for training, the initial learning rate is set to 0.001, and an early stopping mechanism is implemented: when the validation set loss does not decrease for 50 consecutive iterations, the training is terminated to prevent overfitting.

[0128] In some embodiments, a random forest regression model containing 500 decision trees is constructed, and when each tree is split, features are randomly selected for node division.

[0129] In some embodiments, when screening key parameters, an importance threshold is set, and only the design parameters that have a significant impact on the structural response are retained. For example, when the importance score of a certain parameter exceeds the preset threshold, it is determined as a key design variable to achieve dimensionality reduction of the design space.

[0130] In some embodiments, parameter inverse mapping: The optimized design parameters are used to update the bridge model in reverse through the BIM software API interface, and parameters such as the geometric dimensions and material properties of the components are automatically adjusted to ensure the consistency of the model and the optimization results.

[0131] In some embodiments, based on the updated BIM model, the built-in script engine is called to automatically generate construction drawings that comply with engineering specifications, and a bill of quantities containing information such as the volume of concrete and the amount of steel bars is synchronously exported.

[0132] In some embodiments, parameter inverse mapping is achieved through the ElementParameter mechanism of Revit API, and the mapping error control satisfies:

[0133]

[0134] X opt : The optimized design parameter vector; X BIM: The actual parameter vector in the BIM model; ‖·‖2: Euclidean norm (i.e., the modulus of the vector).

[0135] In some embodiments, when the prediction error δ of the deep neural network surrogate model is > 5%, an additional sampling strategy is automatically triggered:

[0136] X new = X opt ±ΔX·N(0, 0.1);

[0137] X new is the sampling sample, and ΔX is the allowable fluctuation range of the design variable.

[0138] In some embodiments, the improved NSGA-II algorithm adopted has an adaptive crossover probability adjustment mechanism, and its crossover probability is dynamically adjusted according to the number of evolutionary generations:

[0139] Within the first 50% of the evolutionary generations, the crossover probability linearly decreases from the initial value of 0.9 to 0.5;

[0140] Within the last 50% of the evolutionary generations, the crossover probability remains constant at 0.5;

[0141] The mathematical expression of the crossover probability is defined as:

[0142] When the current evolutionary generation is less than or equal to half of the maximum evolutionary generation, the crossover probability changes according to the linear decreasing rule; when the evolutionary generation exceeds half of the maximum evolutionary generation, the crossover probability remains constant at 0.5.

[0143] When G max = 100:

[0144] The p of the 0th generation c = 0.9;

[0145] The p of the 50th generation c = 0.9 - 0.4×0.5 = 0.7;

[0146] After the 50th generation, p is maintained c = 0.5.

[0147] This embodiment breaks through the limitation of the fixed crossover probability of the traditional NSGA-II and avoids premature convergence through the time decay mechanism. For example: taking the maximum evolutionary generation G max = 200 as an example, the crossover probability drops to 0.5 at the 100th generation and remains constant.

[0148] The above-described embodiments are used to illustrate the present invention and are not used to limit the present invention. Therefore, changes in the example numerical values or replacement of equivalent elements should still fall within the scope of the present invention.

[0149] From the above detailed description, those of ordinary skill in the art can clearly understand that the present invention can indeed achieve the aforementioned objectives and actually comply with the provisions of the Patent Law.

[0150] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention. The above description is only the preferred embodiments of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

[0151] It should be noted that the above description of the process is only for illustration and explanation and does not limit the scope of application of this specification. For those skilled in the art, various corrections and changes can be made to the process under the guidance of this specification. However, these corrections and changes are still within the scope of this specification.

[0152] The basic concept has been described above. Obviously, for those of ordinary skill in the art after reading this application, the above invention disclosure is only for illustration and does not constitute a limitation to this application. Although not explicitly stated here, those of ordinary skill in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, so such modifications, improvements, and corrections still fall within the spirit and scope of the exemplary embodiments of this application.

[0153] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "an embodiment" or "one embodiment" or "an alternative embodiment" mentioned twice or more at different positions in this specification is not necessarily the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application can be appropriately combined.

[0154] In addition, those of ordinary skill in the art can understand that various aspects of the present application can be illustrated and described by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Therefore, various aspects of the present application can be implemented entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software. The above hardware or software can all be referred to as "units", "modules", or "systems". In addition, various aspects of the present application can take the form of a computer program product embodied in one or more computer-readable media, in which computer-readable program code is included.

[0155] The computer program code required for the operation of each part of the present application can be written in any one or more of the above programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages such as Python, Ruby, and Groovy, or other programming languages. This program code can run entirely on the user's computer, or run as an independent software package on the user's computer, or run partially on the user's computer and partially on a remote computer, or run entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer in any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (for example, through the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0156] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numbers and letters, or the use of other names in the present application are not used to limit the order of the processes and methods of the present application. Although some currently considered useful embodiments of the invention are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of the present application. For example, although the implementation of the above various components can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installed on an existing server or mobile device.

[0157] Similarly, it should be noted that, in order to simplify the description disclosed in the present application and thus help in understanding one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, sometimes multiple features are grouped into one embodiment, drawing, or description thereof. However, this method of the present application should not be construed as reflecting an intention that the claimed subject matter requires more features than are expressly recited in each claim. On the contrary, the subject matter of the invention should have fewer features than the above single embodiment.

Claims

1. A method for optimizing the bridge structure of a BIM model based on a neural network, characterized in that, Including: Establish a BIM model of the bridge structure: Use BIM software to establish the geometric model, material model, and load model of the bridge structure; Define the optimization objectives and constraints: According to the design requirements of the bridge structure, define the optimization objectives and constraints; Generate a training data set: Use the parametric modeling method to generate a large number of BIM models of the bridge structure with different design parameters, and use finite element analysis software to calculate the response values of each model to form a training data set; Train the neural network model: Use the training data set to train the neural network model so that it can predict the response value of the bridge structure according to the input design parameters; Surface fitting: Use the trained neural network model to generate a large number of predicted data points, and use the surface fitting method to fit a smooth response surface; Random forest to screen important design parameters: Use the random forest method to screen out the design parameters that have the greatest impact on the response value of the bridge structure; Optimize the bridge structure: Use the fitted response surface and the screened important design parameters, combined with the genetic algorithm and the particle swarm algorithm, to optimize the bridge structure and obtain a design scheme that meets the constraint conditions and has the smallest optimal objective value.

2. A BIM model bridge structure optimization method based on a neural network, characterized in that Including: S1. Define the bridge design variable set through BIM parametric modeling, and use Latin hypercube sampling to generate the sample space; S2. Call the finite element solver to batch calculate the structural response data of the samples; S3. Construct a deep neural network surrogate model and train it with the design variables as the input and the structural response as the output; S4. Use the random forest algorithm to evaluate the feature importance of the design variables and screen out the key parameters; S5. Based on the screened key parameters, construct a quadratic response surface model and use the NSGA-II algorithm for multi-objective optimization to obtain the optimized parameters; S6. Inversely map the optimized parameters to the BIM model and output the final scheme through finite element secondary verification.

3. The BIM model bridge structure optimization method based on a neural network according to claims 1 and 2, characterized in that, Compare the BIM modelings of the design scheme that meets the constraint conditions and has the smallest optimal objective value and the final scheme, and select the better scheme.

4. The BIM model bridge structure optimization method based on a neural network according to claim 2, wherein, The deep neural network surrogate model described in S3 contains 3 hidden layers, the activation function is LeakyReLU, α = 0.01, and the loss function L is defined as: L = 0.7MSE + 0.3MAE + 0.1‖W‖2; MSE is the mean square error, MAE is the mean absolute error, and ‖W‖2 is the L2 regularization term.

5. The BIM model bridge structure optimization method based on a neural network according to claim 2, wherein In S4, parameter screening is achieved through the Gini importance index, the threshold θ = 0.15 is set, and the importance calculation formula is: I k is the feature importance score of the k-th design variable, T is the total number of decision trees in the random forest, is the reduction in impurity caused by splitting the k-th feature in the t-th tree, N t is the number of samples in the t-th tree, and N is the total number of samples.

6. The BIM model bridge structure optimization method based on a neural network according to claim 2, characterized in that The quadratic response surface model described in S5 is: is the predicted structural response value; β0 is the intercept term; represents the baseline response value when all design variables are zero; β i is the linear term coefficient of the i-th design variable, reflecting the strength of the linear influence of this variable on the response; β ij is the interaction term coefficient between the i-th and j-th design variables, and is the quadratic term coefficient when i = j; x i is the i-th design variable, x j is the j-th design variable, p is the total number of key design variables after screening, determined by random forest feature importance analysis, p ≤ n, where n is the total number of original variables; ∈ is the model error term, following a normal distribution with a mean of zero; Use ridge regression to fit the coefficients, normalize the design variables, and use the least squares method with L2 regularization to fit the coefficients: The regularization parameter λ = 0.01, X: the design matrix containing linear terms, interaction terms, and quadratic terms.

7. The BIM model bridge structure optimization method based on a neural network according to claim 2, wherein In S5, the improved NSGA-II algorithm is used, and an adaptive crossover probability is introduced: p c (g) is the crossover probability during the g-th generation of evolution, G max is the preset maximum number of generations of evolution, representing the maximum number of iterations of the optimization algorithm; g is the current generation of evolution, representing the number of iterations that the algorithm has executed.

8. The BIM model bridge structure optimization method based on a neural network according to claim 2, characterized in that, In constructing the deep neural network surrogate model, the number of neurons in the input layer is equal to the dimension of the design parameters, the hidden layer adopts a fully connected layer structure and applies the ReLU activation function, and the output layer linearly maps to predict the structural response. The mathematical expression is: h1 = ReLU(W1X + b1); h2 = ReLU(W2h1 + b2); ReLU is the rectified linear unit activation function, W1, W2, and W3 are weight matrices, and b1, b2, and b3 are bias terms. is the predicted structural response value, and h1 and h2 are transition parameters.

9. The BIM model bridge structure optimization method based on a neural network according to claim 2, characterized in that, The function for multi-objective optimization is: minf1(X) is the optimization objective function of the structural weight, minf2(X) is the optimization objective function of the peak equivalent stress, X is the set of design variables; ρ is the material density; A i is the cross-sectional area of the i-th member, L i is the length of the i-th member, σ um is the peak equivalent stress; u max is the maximum displacement response of the bridge structure, L is the main span of the bridge, λ 屈曲 is the buckling safety factor.

10. The BIM model bridge structure optimization method based on a neural network according to claim 7, characterized in that, The termination condition of the improved NSGA-II algorithm is to reach the maximum number of generations G max , or the improvement rate of the Pareto front is less than 0.1% for 20 consecutive generations, where the calculation formula for the improvement rate is: is the average value of the objective function of the i-th generation population.

Images