Building structure design optimization method based on artificial intelligence

Abstract

The invention discloses a building structure design optimization method based on artificial intelligence, and the method comprises the steps: extracting the mechanical characteristics of a building structure through a convolutional neural network, achieving the multi-target dynamic optimization of the building structure through the combination of reinforcement learning and a genetic algorithm, and integrating CAE verification, CAD drawing and BIM construction feedback. According to the design method, the building structure design efficiency and the economical efficiency of the design scheme are remarkably improved, complex building engineering constraints and green building authentication are supported, and the design method is suitable for newly-built and existing building structure transformation scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction engineering, and in particular relates to an artificial intelligence-based building structure design optimization method. Background Art

[0002] While architectural structural design theories, methods, techniques, and specifications have become increasingly sophisticated, the following issues remain widespread: Design specifications for complex structures are difficult to fully cover, leaving engineers with a lack of design basis. While conventional structures often involve limited creative work, they often involve a high volume of repetitive tasks, leaving production efficiency in a critical need. Therefore, improving the design capabilities of complex structures and increasing the efficiency of conventional structures have become crucial requirements for current architectural structural design. Advances in computer analysis capabilities and advancements in intelligent algorithms offer new solutions to these challenges, and deep learning-based intelligent design will be a key technology for addressing these challenges.

[0003] Existing AI-based building structure technology approaches often focus on single-objective optimization, lack full-process automation capabilities, and struggle to collaboratively address multi-dimensional factors such as material properties, regulatory constraints, and construction feasibility. Furthermore, issues such as low historical data utilization and difficulties in cross-disciplinary collaboration further constrain design efficiency. This invention systematically addresses these issues by systematically integrating multimodal AI technologies to propose a full-chain solution encompassing data-driven modeling, dynamic optimization decision-making, and engineering verification. Summary of the Invention

[0004] The purpose of the present invention is to provide an artificial intelligence-based building structure design optimization method to solve the problems raised in the above background technology.

[0005] The objectives of the present invention are achieved through the following technical solutions.

[0006] An artificial intelligence-based building structure design optimization method, the method comprising:

[0007] Use convolutional neural networks (CNNs) to extract mechanical features from historical engineering data;

[0008] Generate multi-objective optimization solutions through collaborative reinforcement learning and genetic algorithms;

[0009] Integrate CAE simulation verification and CAD automated drawing.

[0010] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that: before the CNN extracts the mechanical features, the historical engineering data is preprocessed, including:

[0011] Remove noisy data through data cleaning;

[0012] Standardize coding of unstructured data such as design drawings and test reports;

[0013] Normalization is used to unify the dimensional differences.

[0014] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that: the genetic algorithm adopts the NSGA-II algorithm framework, and the optimization objectives include at least material cost, structural stiffness and carbon emissions.

[0015] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that: the method further includes an integrated interface with the BIM platform, supporting real-time data feedback to the design optimization module during the construction phase.

[0016] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that the integrated interface of the BIM platform collects construction data in real time through Internet of Things sensors, including material deformation, environmental temperature and humidity, and load changes, and dynamically adjusts design parameters to match actual construction conditions.

[0017] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that the reward function of reinforcement learning includes a weighted sum of a safety index and an economic index, and the weights are dynamically adjusted by technical personnel.

[0018] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that: the safety indicators of the reward function include seismic performance coefficient, structural failure probability and key node redundancy, the economic indicators include material cost, construction cost and operation and maintenance cost, and the weight adjustment method supports manual input and automatic preset templates based on project type matching.

[0019] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that the optimization objectives of the genetic algorithm further include a construction feasibility score and a historical building protection level, wherein the construction feasibility score is evaluated by simulating the feasibility of the scheme through the Monte Carlo method.

[0020] Furthermore, an artificial intelligence-based building structure design optimization method is characterized in that: the method also includes a human-computer collaborative decision-making module, which provides a visual interface for engineers to manually correct the AI-generated solution. The corrected solution is fed back to the CNN and reinforcement learning model as new training data to realize online model updating.

[0021] Further, an artificial intelligence-based building structure design optimization method is characterized in that the CAE simulation verification adopts multi-physical field coupling analysis, including statics, dynamics and thermodynamics simulation, and when the simulation result deviates from the AI prediction value by more than a preset threshold, local parameter re-optimization is automatically triggered and a difference report is generated.

[0022] The artificial intelligence-based building structure design optimization method extracts building structure mechanical characteristics through a convolutional neural network, realizes multi-objective dynamic optimization of the building structure by combining reinforcement learning and a genetic algorithm, and integrates CAE verification, CAD drawing and BIM construction feedback. The design method significantly improves the efficiency of building structure design and the economy of the design scheme, supports complex building engineering constraints and green building certification, and is suitable for new building structure reconstruction scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a flowchart of the artificial intelligence-based building structure design optimization method. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0025] As shown in Figure 1 , an artificial intelligence-based building structure design optimization method includes:

[0026] Data acquisition and preprocessing module: used for collecting, cleaning, encoding and normalizing historical engineering data;

[0027] Feature extraction module: using a convolutional neural network (CNN) to extract mechanical characteristics in historical engineering data;

[0028] Reinforcement decision module: generating a multi-objective optimization scheme through reinforcement learning and genetic algorithm cooperation;

[0029] Verification and output module: CAE simulation verification and CAD automatic drawing;

[0030] Closed-loop feedback module: BIM integrated interface and dynamic adjustment of construction data.

[0031] The implementation process of the artificial intelligence-based building structure design optimization method specifically includes the following steps.

[0032] Step 1: Collect engineering data and perform preprocessing.

[0033] The engineering data to be collected includes:

[0034] Architectural structural parameters of historical projects, such as component cross-sectional dimensions, material performance parameters, load types and values, etc.;

[0035] Environmental data such as geological data, seismic intensity zoning, climate characteristics, wind loads, etc.;

[0036] Economic indicators such as cost, construction period, etc.;

[0037] Unstructured data includes design drawings, test reports, etc.

[0038] The data preprocessing process includes:

[0039] First, data cleaning is performed to remove sensor outliers and fill in missing data;

[0040] Secondly, standardized coding is adopted to vectorize drawings and convert test report text into structured tags;

[0041] Finally, normalization processing is performed to normalize and remove dimension of numerical data such as load and cost.

[0042] Step 2: Extract the structural mechanics characteristics of the building project and establish a data model.

[0043] For the construction engineering data preprocessed in step 1, a convolutional neural network (CNN) is used to extract the mechanical characteristic parameters of the above construction engineering data and convert them into building structure topology maps and load distribution maps; the mechanical characteristic parameters mainly include building structure composition, component force form, load distribution, stress distribution, stiffness, deformation, etc.

[0044] The convolutional neural network uses ResNet-50 as its backbone network, taking as input the structural topology and load distribution map, and outputting the stress concentration factor and displacement gradient of key areas. As a classic deep residual network, ResNet-50's core advantages lie in: its residual connection mechanism effectively mitigates the vanishing gradient problem, allowing for the construction of deeper 50-layer networks; its pre-trained parameters provide powerful feature extraction capabilities, excelling in cross-domain tasks; and its multi-stage downsampling structure is suitable for capturing the hierarchical features of structural images.

[0045] The structural topology image and the load distribution image are spliced ​​into a dual-channel input. A dual-branch network is used to extract features from each image, then perform feature splicing. Histogram equalization is then performed to enhance contrast. A deconvolution network generates a displacement gradient field with the same resolution as the original image, and outputs stress concentration factor and displacement gradient vector field model data in parallel.

[0046] Step 3: Train the model data.

[0047] Using transfer learning, we trained stress concentration factors and displacement gradient vector field data based on an ImageNet pre-trained model. By converting engineering drawings to ImageNet images and using a pre-trained model on the ImageNet dataset, we achieved superior results compared to random initialization.

[0048] Transfer learning employs a gradual unfreezing strategy to effectively avoid overfitting and training instability during fine-tuning. Differentiated learning rates are set for different layers of the deep neural network to optimize model training. A teacher network is introduced to guide the training of the student network, enabling knowledge transfer and reducing training costs.

[0049] The model data enhancement strategies for the characteristics of construction engineering data include: using elastic deformation to simulate material deformation, using random holes to simulate structural defects; spatial transformation of load distribution while maintaining mechanical conservation conditions; using the CutMix strategy to generate new training samples by fusing parts of different images, thereby improving the model's perception of local features.

[0050] Step 4: Multi-objective dynamic optimization of model data. The multi-objective dynamic optimization of building engineering structural design model data is achieved through the synergy of reinforcement learning and genetic algorithm.

[0051] The reinforcement learning module, through interactive learning with external environmental construction data, enables the entire model system to adapt to changes in external environmental construction data in real time, thereby achieving optimal control over the building structure design. The reinforcement learning module establishes a reward function, R, to evaluate the structural design corresponding to control decisions made in response to changes in external environmental construction data, ensuring that the model system's control effect achieves the maximum reward within the current environmental state.

[0052] The reward function R is the weighted sum of the safety index S and the economic index E. The weight adjustment method can be manually input by technicians or automatically preset based on project type matching templates.

[0053] The safety index S is calculated using the following formula: S = 0.6 x K1 + 0.3 x K2 + 0.1 x K3;

[0054] Where K1 is the seismic coefficient calculated by the response spectrum method;

[0055] K2 is the component failure probability calculated based on reliability analysis;

[0056] K3 is the redundancy score of key nodes.

[0057] The economic index E is calculated using the following formula: E = 1 / (0.5 x N1 + 0.35 x N2 + 0.15 x N3);

[0058] Where E1 is the material cost;

[0059] E2 is the construction cost;

[0060] E3 is the maintenance cost.

[0061] The genetic algorithm adopts the NSGA-II algorithm framework to solve the trade-off problem of multiple optimization objectives through non-dominated sorting, uses the congestion comparison mechanism to maintain the diversity of solutions, and finally outputs a set of Pareto optimal solutions.

[0062] Optimization goals include:

[0063] Economic goal: Minimize material costs, which is directly related to the economic benefits of the project;

[0064] Engineering performance goals: maximize structural stiffness to ensure building safety and stability;

[0065] Environmental goals: Minimize carbon emissions and respond to green building and sustainable development needs;

[0066] Construction practice objectives: maximize the construction feasibility score and evaluate construction risks through Monte Carlo simulation;

[0067] Cultural protection goal: to maximize the protection level of historical buildings and reflect respect for historical and cultural heritage.

[0068] To address the complex issue of construction feasibility, we employ a Monte Carlo method for probabilistic assessment. By randomly sampling and simulating uncertainties in the construction process, such as weather changes, fluctuations in resource supply, and differences in labor efficiency, we generate a large number of possible construction scenarios. The statistical probability of successful implementation of each scenario serves as the basis for scoring, effectively quantifying construction risks.

[0069] Genetic algorithms must simultaneously address conflicting demands across five dimensions: economic, technological, environmental, practical, and cultural. For example, reducing material costs may compromise structural rigidity, while preserving historic buildings may increase carbon emissions. NSGA-II uses a non-dominated sorting mechanism to output a set of candidate solutions that optimally balance these multiple objectives, providing a scientific basis for decision makers.

[0070] Genetic algorithms and reinforcement learning work together. The genetic algorithm is responsible for generating initial candidate solutions and screening the optimal solutions, while reinforcement learning makes local fine-tuning of these solutions based on dynamic environmental feedback. The learning results of reinforcement learning can be fed back into the evolutionary process of the genetic algorithm, improving the search efficiency of subsequent iterations, forming a multi-objective dynamic optimization solution that combines global exploration of the genetic algorithm and local optimization of reinforcement learning.

[0071] Step 5: Solution verification and output.

[0072] The system integrates CAE simulation verification and uses multi-physics field coupling to analyze the statics, dynamics, and thermodynamics data of building engineering structures. If the simulation results deviate from the system prediction by more than 5%, the local parameter re-optimization mechanism is automatically triggered and a difference report is generated.

[0073] The system integrates CAD to automatically generate drawings, imports optimized parameters into Revit through API, and generates BIM models and construction drawings.

[0074] Step 6: Construction closed-loop feedback.

[0075] The integrated interface of the BIM platform collects construction data in real time through IoT sensors, including material deformation, ambient temperature and humidity, and load changes. If the monitoring data exceeds the preset threshold, the plan will be automatically adjusted and the design parameters will be updated.

[0076] The BIM platform integrates multi-source data, including design models, construction plans, and sensor data, through standardized interfaces to form a unified digital twin environment. IoT sensors are deployed at key locations on the construction site, such as structural support points and environmentally sensitive areas, to monitor material deformation, ambient temperature and humidity, and load change data at key nodes in real time. Based on engineering design specifications, historical data, and safety margins, reasonable thresholds for each parameter are pre-defined in the BIM platform. Sensor data is uploaded to the BIM platform at a frequency of milliseconds, and the system automatically compares the measured data with the threshold. If a parameter continues to exceed the standard, such as deformation exceeding the threshold by 10% for 30 seconds, an early warning mechanism is triggered.

[0077] When the monitoring data exceeds the limit, the system implements closed-loop control through the following paths: combining the BIM model to locate the abnormal position and analyze the cause of the exceedance; calling the built-in optimization algorithm library to generate an adjustment plan; updating the adjusted design parameters to the BIM model in real time, and synchronously pushing them to on-site managers through mobile terminals.

[0078] Step 7: Human-machine collaborative decision-making.

[0079] The system is equipped with a visual human-computer dialogue interface that can simultaneously display the AI ​​solution and the technician's revised version for comparison on one screen, and supports real-time adjustment of component size by dragging and dropping; the revised solution is fed back to the CNN and reinforcement learning model as a new sample to realize online updating of model data.

[0080] The various technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A building structure design optimization method based on artificial intelligence, characterized in that: The method comprises: Use convolutional neural networks (CNNs) to extract mechanical features from historical engineering data; Generate multi-objective optimization solutions through collaborative reinforcement learning and genetic algorithms; Integrate CAE simulation verification and CAD automated drawing.

2. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: Before the CNN extracts mechanical features, the historical engineering data is preprocessed, including: Remove noisy data through data cleaning; Standardize coding of unstructured data such as design drawings and test reports; Normalization is used to unify the dimensional differences.

3. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: The genetic algorithm adopts the NSGA-II algorithm framework, and the optimization objectives include at least material cost, structural stiffness and carbon emissions.

4. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: The method further includes an integrated interface with the BIM platform to support real-time data feedback from the construction phase to the design optimization module.

5. The method for optimizing building structure design based on artificial intelligence according to claim 4, characterized in that: The integrated interface of the BIM platform collects construction data in real time through IoT sensors, including material deformation, ambient temperature and humidity, and load changes, and dynamically adjusts design parameters to match actual construction conditions.

6. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: The reward function of the reinforcement learning includes a weighted sum of a safety index and an economic index, and the weights are dynamically adjusted by technical personnel.

7. The method for optimizing building structure design based on artificial intelligence according to claim 6, characterized in that: The safety indicators of the reward function include seismic performance coefficient, structural failure probability and redundancy of key nodes, and the economic indicators include material cost, construction cost and operation and maintenance expenses. The weight adjustment method supports manual input and automatic preset templates based on project type matching.

8. The method for optimizing building structure design based on artificial intelligence according to claim 3, characterized in that: The optimization objectives of the genetic algorithm further include a construction feasibility score and a historical building protection grade, wherein the construction feasibility score is evaluated by simulating the feasibility of the scheme through the Monte Carlo method.

9. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: The method also includes a human-machine collaborative decision-making module, which provides a visual interface for engineers to manually correct the AI-generated solution. The corrected solution is fed back to the CNN and reinforcement learning model as new training data to achieve online model updates.

10. The method for optimizing building structure design based on artificial intelligence according to claim 1, characterized in that: The CAE simulation verification adopts multi-physics field coupling analysis, including statics, dynamics and thermodynamics simulation, and when the deviation between the simulation result and the AI ​​prediction value exceeds a preset threshold, it automatically triggers local parameter re-optimization and generates a difference report.

Images