Building design optimization method and system based on artificial intelligence
Through the architectural design optimization method based on artificial intelligence, using technologies such as convolutional neural networks and variational autoencoders to build a multi-objective comprehensive model, solving the problems of inefficient architectural design and difficulty in balancing multiple goals in the existing technology, and achieving efficient and accurate architectural design optimization.
Patent Information
- Application Number
- CN202510206339.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-03-25
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing architectural design technology is inefficient and susceptible to human factors. It is difficult to fully consider multiple aspects of building performance, and it is difficult to find the optimal balance between multiple goals.
Using an artificial intelligence-based architectural design optimization method, we use a convolutional neural network to build a mapping model, combine a variational autoencoder and a long and short-term memory network to build a multi-objective comprehensive model, and use a multi-objective particle swarm optimization algorithm and hierarchical analysis method to optimize the design scheme.
It greatly shortens the time for design evaluation and adjustment, and can consider multiple performance indicators of the building at the same time, find the optimal balance between multiple goals, and improves the efficiency and accuracy of the design process.
Smart Images

Figure CN119691879A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of architectural design, and in particular to an architectural design optimization method and system based on artificial intelligence. Background Art
[0002] In the field of architecture, the design process needs to take into account many factors to achieve ideal performance and quality. However, there are many problems that need to be solved in the existing technology in terms of architectural design optimization, as detailed below: Traditional architectural design mainly relies on the designer's experience and manual calculations. Designers need to initially conceive a design plan based on past project experience, and then use complex manual calculations to evaluate the feasibility and performance indicators of the plan. However, this method is inefficient, and designers often need to spend a lot of time on repeated calculations and modifications. Moreover, manual calculations are easily interfered by human factors, and accuracy is difficult to guarantee. Especially when dealing with complex building structures and diverse performance requirements, calculation errors are prone to occur, resulting in the design plan may not meet actual needs.
[0003] At present, some building designs focus on a single performance indicator when evaluating, such as focusing only on the structural safety or energy-saving effect of the building. But in fact, building performance is a comprehensive concept, covering multiple aspects such as structural safety, energy saving, lighting, ventilation, etc. Focusing only on a single performance indicator will lead to defects in other performance aspects. For example, over-emphasizing energy saving may sacrifice the lighting and ventilation effects of the building and affect the comfort of the residents; while focusing only on structural safety may increase construction costs and material consumption, which does not meet environmental and economic requirements.
[0004] Architectural design usually faces multi-objective optimization problems, which require considering multiple conflicting performance indicators at the same time, such as improving building comfort while reducing building energy consumption, and reducing building costs while ensuring structural safety. Existing design methods and tools have limited capabilities in dealing with multi-objective optimization problems, and it is difficult to find the optimal balance between multiple objectives. They often optimize each objective one by one. This method cannot fully consider the relationship between each objective, and is prone to falling into a local optimal solution, and cannot obtain a global optimal design solution.
[0005] A large amount of data is generated during the architectural design process, including design parameters, performance test data, past project cases, etc. However, the existing design process fails to fully explore and utilize the value of this data. On the one hand, the lack of effective data integration and management leads to scattered information and difficulty in comprehensive analysis; on the other hand, the lack of advanced data processing technology and models makes it impossible to extract valuable information from massive data to support design decisions and optimization processes.
[0006] Although simulation technologies such as finite element analysis have been applied to building performance evaluation, these technologies have certain limitations. Finite element analysis requires the establishment of complex models, and has high requirements for the accuracy of the models and the setting of boundary conditions. Once the model is inaccurate or the boundary conditions are set unreasonably, the simulation results will deviate greatly from the actual situation. Moreover, the simulation analysis process is usually computationally intensive and time-consuming, and cannot meet the needs of rapid design and optimization. Summary of the invention
[0007] The purpose of the present invention is to provide an artificial intelligence-based architectural design optimization method and system to solve the problems raised in the above-mentioned background technology.
[0008] To achieve the above object, the present invention provides the following technical solution: an architectural design optimization method based on artificial intelligence, the method comprising: S1. Collect data samples containing building design parameters and building performance. These data samples come from building design documents and measured data from past building projects.
[0009] Preferably, the collected data samples are subjected to Kalman filtering for denoising. Kalman filtering is a recursive optimal estimation algorithm that can effectively remove noise interference in the data and improve the accuracy and reliability of the data through the prediction and update process of the data. Specifically, it estimates and corrects the true state of the data based on the historical information and current measurement values of the data, using the state transfer equation and the observation equation, so as to obtain a purer data sample and provide a better foundation for subsequent model training.
[0010] S2. Use the Convolutional Neural Network (CNN) in deep learning to build a mapping model that can map architectural design parameters to building performance. The input of the mapping model is the architectural design parameter vector , the output is the building performance vector , its mathematical expression is ,in are the parameters of the convolutional neural network.
[0011] Preferably, a convolutional neural network is used to fit the collected data samples, and the data samples are divided into training data and test data according to a certain ratio (e.g., 7:3 or 8:2). The purpose of dividing the training data and test data is to evaluate the generalization ability of the model and avoid overfitting of the model. The training data is used to train the parameters of the convolutional neural network so that the model learns the mapping relationship between the building design parameters and the building performance; the test data is used to verify the prediction accuracy of the model on unseen data after the model training is completed.
[0012] S3. Use variational autoencoder (VAE) to build a distribution expression model that can map the distribution of building design parameters and building performance. The variational autoencoder consists of two parts: the encoder and the decoder. The encoder maps the input building design parameters and building performance data to the latent space, and the decoder reconstructs the latent space vector into building design parameters and building performance data.
[0013] Preferably, a variational autoencoder that can characterize the distribution of the original data samples is configured, and network parameter training is performed based on the training data. During the training process, the variational autoencoder learns the distribution characteristics of the data by minimizing the reconstruction error and the KL divergence between the latent space distribution and the prior distribution. When the preset training end conditions are met (for example, the maximum number of training rounds is reached or the reconstruction error converges to a certain threshold), the encoder in the variational autoencoder is used as a distribution expression model. The distribution expression model can map the building design parameters and building performance data to a low-dimensional latent space, which is convenient for subsequent multi-objective comprehensive model processing.
[0014] S4. Incorporate the outputs of the distribution expression model and the mapping model into at least one long short-term memory network layer (LSTM) to obtain a multi-objective comprehensive model.
[0015] Preferably, when learning and training the network parameters of the multi-objective comprehensive model based on the training data, the mean square error loss function and the cross entropy loss function are used as the two objectives in the dual-objective optimization problem. The mean square error loss function is used to measure the error between the model prediction value and the true value, and the cross entropy loss function is often used for classification problems or comparison of probability distributions. By optimizing these two loss functions at the same time, the multi-objective comprehensive model can achieve better results in both prediction accuracy and classification performance.
[0016] Preferably, a multi-objective particle swarm optimization algorithm (MOPSO) is used to obtain the non-dominated solution set boundary of the corresponding mapping model training error. In the multi-objective particle swarm optimization algorithm, the position of each particle i represents a set of architectural design parameters, the speed of each particle i The update formula is: ; in, Indicates that particle i is The speed of the moment; Indicates that particle i is The speed at the moment, w is the inertia weight, and is the learning factor, and is a random number between [0, 1], is the historical optimal position of the particle, is the global optimal position; position The update formula is: ; in, Indicates that particle i is The location at the moment, Represents particle i in By continuously iteratively updating the position and velocity of the particle, the algorithm can search for the boundary of the non-dominated solution set that maps the training error of the model.
[0017] Preferably, the Analytic Hierarchy Process (AHP) is used to select a specified number of candidate solution sets within the boundary of the mapping model training error. The Analytic Hierarchy Process is a decision-making method that combines qualitative and quantitative analysis. It establishes a hierarchical model, compares each factor pairwise, and determines its relative importance, thereby selecting the candidate solution set that best matches the target demand from the non-dominated solution set. Then, the network parameters of the mapping model are determined based on the final performance of the candidate solution set on the predicted data.
[0018] Preferably, according to the performance of the multi-objective comprehensive model corresponding to different solution sets on the test data, a set of network parameters with the best performance is finally selected to obtain a trained mapping model. By evaluating the model performance corresponding to different solution sets on the test data, it can be ensured that the selected model has good generalization ability and prediction accuracy.
[0019] S5. Solve the boundary of the non-dominated solution set of the multi-objective comprehensive model.
[0020] Preferably, the Fast Non-Dominated Sorting Algorithm is used to solve the boundary of the non-dominated solution set. In the Fast Non-Dominated Sorting Algorithm, for each individual q, its number of dominated and dominating set .like , then add it to the first non-dominated layer . Then remove The domination relationship of individuals in the network to other individuals is updated, and the number of dominated individuals is updated. This process is repeated to obtain all non-dominated layers. The non-dominated solution set boundary of the multi-objective comprehensive model can be efficiently found through the fast non-dominated sorting algorithm.
[0021] Preferably, a candidate solution set matching the target requirement is selected in the boundary of the non-dominated solution set according to the hierarchical analysis method. By considering the importance of different building performance indicators, the hierarchical analysis method is used to sort and screen the individuals in the non-dominated solution set to select the candidate solution set that best meets the design goal.
[0022] S6. A new building design sample is obtained on the candidate solution set by a building performance simulation method based on finite element analysis. If the specified performance of the new building design sample reaches the expected target performance value of the design, the simulation is stopped. Otherwise, the new building design sample is used as a data sample and the matching information is input into the multi-objective comprehensive model to continue the solution set processing and simulation optimization operation.
[0023] Preferably, the building performance simulation method based on finite element analysis can perform detailed simulation analysis on the structural mechanical properties, thermal conductivity properties, fluid mechanical properties, etc. of the building. By establishing a finite element model of the building, the building is divided into multiple small units, and the mechanical, thermal and other equations of each unit are solved according to the laws of physics and boundary conditions to obtain the performance indicators of the building. This simulation method can provide accurate prediction of building performance and provide strong support for building design optimization.
[0024] The present invention also provides an artificial intelligence-based architectural design optimization system, which includes the following modules: The data collection module is used to collect data samples containing building design parameters and building performance based on building design documents and measured data from past building projects, and perform Kalman filtering and denoising on the collected data samples.
[0025] Preferably, the module can be equipped with a data interface to facilitate the acquisition of architectural design documents and measured data from different data sources. At the same time, the implementation code of the Kalman filter algorithm is integrated to automatically denoise the collected data and improve the data quality.
[0026] The mapping model construction module uses the convolutional neural network in deep learning to construct a mapping model that can map building design parameters to building performance. Its input, output and mathematical expression are as described in the above method.
[0027] Preferably, the module can use a deep learning framework (such as TensorFlow or PyTorch) to build and train a convolutional neural network. It provides a visual interface to facilitate users to set the network structure, training parameters, etc., while monitoring the loss function value and model performance indicators in real time during the training process.
[0028] The distribution expression model construction module uses a variational autoencoder to construct a distribution expression model that can map the distribution relationship between building design parameters and building performance. The functions of the encoder and decoder are as described in the above method.
[0029] Preferably, this module also implements the construction and training of variational autoencoders based on a deep learning framework. It supports users to adjust the network structure, learning rate and other parameters of the encoder and decoder to optimize the performance of the distribution expression model. At the same time, it provides a visualization function for the latent space to help users understand the distribution characteristics of data in the latent space.
[0030] The multi-objective comprehensive model construction module incorporates the outputs of the distribution expression model and the mapping model into at least one long short-term memory network layer to obtain a multi-objective comprehensive model. The construction process is as described in the above method.
[0031] Preferably, the module integrates the implementation code of the multi-objective particle swarm optimization algorithm and the hierarchical analysis method, automatically completing the network parameter training and candidate solution set selection of the multi-objective comprehensive model. A parameter setting interface is provided to allow users to adjust the parameters of the optimization algorithm, such as inertia weight, learning factor, etc., to adapt to different building design optimization problems.
[0032] The solution set processing module solves the non-dominated solution set boundary of the multi-objective comprehensive model and selects a candidate solution set matching the target requirement in the non-dominated solution set boundary according to the hierarchical analysis method.
[0033] Preferably, the module implements a fast non-dominated sorting algorithm and a hierarchical analysis method to efficiently solve the boundary of the non-dominated solution set and select the candidate solution set. A result display function is provided to present the non-dominated solution set and the candidate solution set to the user in the form of a chart or table, which is convenient for the user to analyze and make decisions.
[0034] The simulation optimization module obtains new building design samples on the candidate solution set through the building performance simulation method based on finite element analysis, and decides whether to stop the simulation or continue the iterative optimization according to the performance of the new samples. The operation is as described in the above method.
[0035] Preferably, the module integrates the interface of finite element analysis software, automatically generates finite element models of buildings and performs performance simulation. The performance indicators of new building design samples are monitored in real time. When the expected target performance value is reached, the simulation is automatically stopped; otherwise, the new sample data is fed back to the multi-objective comprehensive model to continue optimization iteration. At the same time, a visual display of the simulation results is provided to help users intuitively understand the performance characteristics of the building.
[0036] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses a convolutional neural network to construct a mapping model, which can quickly and accurately map building design parameters to building performance. Compared with the traditional method that relies on designer experience and manual calculations, it greatly shortens the time for design scheme evaluation and adjustment. Designers do not need to perform a large number of tedious manual calculations. The model can give the predicted results of building performance in a short time, making the design process more efficient and effectively shortening the entire building design cycle.
[0037] The present invention adopts a multi-objective comprehensive model, and incorporates the output of the distribution expression model and the mapping model into the long short-term memory network layer, which can simultaneously consider multiple performance indicators of the building, such as structural safety, energy saving, lighting, ventilation, etc. The boundary of the non-dominated solution set is solved by the multi-objective particle swarm optimization algorithm, and the candidate solution set that matches the target demand is selected from it by the hierarchical analysis method, which effectively solves the problem that the traditional method is difficult to balance multiple conflicting goals. It can improve the comfort level while reducing the energy consumption of the building, reduce the construction cost while ensuring the structural safety, find the optimal balance between multiple goals, and achieve the global optimal design solution. In different construction projects, owners and designers may have different key requirements. The multi-objective optimization method of the present invention can flexibly adjust the weights of each performance indicator according to the specific project goals and requirements, so as to provide personalized design optimization solutions for different projects and meet the diverse architectural design needs.
[0038] The present invention systematically collects and processes architectural design documents and measured data of past projects. Through a series of operations such as Kalman filter denoising, convolutional neural network fitting, and variational autoencoder learning data distribution, the data is deeply mined and analyzed. The scattered data is integrated and valuable information is extracted from it to provide a more comprehensive and scientific basis for architectural design decisions. New architectural design samples are used as data samples and input into the multi-objective comprehensive model, and iterative optimization is continuously performed. With the continuous accumulation of data and continuous training of the model, the design plan will become more and more optimized. This data-driven iterative optimization process makes full use of the value of data, so that architectural design can continuously adapt to new needs and changes.
[0039] The building performance simulation method based on finite element analysis is combined with the optimization method of the present invention. The simulation analysis is performed on the candidate solution set. Since the candidate solution set is obtained through multi-objective optimization selection, it has high quality and rationality. This makes the establishment of the finite element analysis model more accurate and the setting of boundary conditions more in line with the actual situation, thereby improving the accuracy of the simulation results and more realistically reflecting the actual performance of the building. The artificial intelligence model is used to screen out potential design schemes in advance, reducing the number of samples that need to be analyzed by finite element analysis. At the same time, the model's rapid prediction of building performance also provides more targeted input for finite element analysis, avoiding unnecessary calculations. This greatly shortens the time of simulation analysis and improves the efficiency of design optimization.
[0040] This invention applies a variety of artificial intelligence technologies (such as convolutional neural networks, variational autoencoders, long short-term memory networks, etc.) to the field of architectural design optimization, bringing new technical means and methods to the architectural design industry. This intelligent design optimization method not only improves the design quality and efficiency, but also lays the foundation for the future development of architectural design and promotes the architectural design industry to move towards intelligence and digitization. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 This is a working principle diagram of the building design optimization method of the present invention; Figure 2 Diagram of the steps for building a multi-objective comprehensive model; Figure 3 This is a workflow diagram of the building design optimization system described in the present invention. DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] See also Figure 1-3 The present invention provides an artificial intelligence-based architectural design optimization method and system, which aims to improve the quality and efficiency of architectural design and optimize building performance. The overall implementation scheme is as follows: Collect data samples containing building design parameters and building performance. These data samples come from a wide range of sources and can be obtained from building design documents and measured data from past building projects. They cover design parameters such as building geometry, material selection, and spatial layout, as well as building performance data such as energy consumption, lighting, and ventilation. After collection, the data samples are processed by Kalman filtering to denoise and improve data quality.
[0044] The convolutional neural network in deep learning is used to build a mapping model. The model takes the architectural design parameter vector as input and outputs the corresponding architectural performance vector. By continuously learning the characteristics and rules in the data samples, the mapping relationship between the two is established. Its mathematical expression is: ; in are the parameters of the convolutional neural network.
[0045] A distribution expression model is constructed using a variational autoencoder. The encoder maps the input architectural design parameters and building performance data to the latent space, and the decoder reconstructs the latent space vector into the original data, thereby achieving an effective expression of the data distribution.
[0046] The outputs of the distribution expression model and the mapping model are combined into at least one long short-term memory network layer to obtain a multi-objective comprehensive model. The model can comprehensively consider multiple objectives, input multiple design parameters of the building, and output multiple performance indicators.
[0047] After obtaining the multi-objective comprehensive model, solve its non-dominated solution set boundary. This process can be achieved through the multi-objective particle swarm optimization algorithm or the fast non-dominated sorting algorithm to obtain a set of mutually non-dominant building performance index values. Then, based on the hierarchical analysis method, select the candidate solution set that matches the target demand from the non-dominated solution set boundary.
[0048] A new building design sample is obtained on the candidate solution set by using the building performance simulation method based on finite element analysis. If the specified performance of the new building design sample reaches the expected target performance value of the design, the simulation is stopped; if not, the new sample is used as a data sample, and the matching information is input into the multi-objective comprehensive model, and the above steps of solving the boundary of the non-dominated solution set, selecting the candidate solution set and simulating are repeated until the requirements are met. Embodiment 1
[0049] The method of data sample collection and processing is as follows: In the actual application scenario, a commercial building project in a certain city is selected as the data collection object. The building design parameters are extracted from the design documents of 200 commercial buildings built in the city in the past five years, including the total building area, number of floors, standard floor area, area ratio of different functional areas (such as shopping malls, offices, restaurants, etc.), facade material type (glass curtain wall, stone curtain wall, etc.), window area and orientation and other parameters. At the same time, through the actual monitoring of these buildings, the building performance data is obtained, such as the annual energy consumption data (including electricity, gas and other energy consumption), the average daylighting factor of different indoor areas, ventilation volume, etc.
[0050] After the data samples are collected, Kalman filtering is used for denoising because the data may be affected by factors such as environmental noise and measurement errors. Taking building energy consumption data as an example, assume that the power consumption of a building is monitored within a certain period of time and a series of measurement values are obtained. Kalman filtering predicts the energy consumption state at the current moment by establishing a state space model, and then makes corrections based on the measured values. Its basic principle is to use the state estimate at the previous moment and the measured value at the previous moment to calculate the optimal state estimate at the current moment through the Kalman gain. Specifically, suppose For the The actual value of building energy consumption at the moment, For the The measured value at the moment, Based on The estimated value of the moment The predicted value at time, For the The best estimate of time, For the The best estimate of time, is the forecast error covariance, is the estimated error covariance, For the The estimated error covariance at time , is the Kalman gain. Then the prediction process is: (in is the state transfer matrix, is the control matrix, is the control quantity, assuming no external control here, ), ( is the process noise covariance). The update process is: ( is the observation matrix, is the measurement noise covariance), Through multiple iterative calculations, noise can be effectively removed and more accurate energy consumption data can be obtained, providing reliable data support for subsequent model training. Embodiment 2
[0051] Example of building a mapping model: Taking a medium-sized residential building design as an example, a mapping model is constructed. The design parameters of the residential building include the building's apartment structure (such as one bedroom and one living room, two bedrooms and one living room, etc.), the length, width and height of the room, the wall material (such as brick-concrete structure, frame structure, etc.), the type of window (ordinary double-glazed, broken bridge aluminum doors and windows, etc.). These design parameters are digitally encoded to form a building design parameter vector .
[0052] A convolutional neural network (CNN) is used to construct the mapping model. CNN consists of multiple convolutional layers, pooling layers, and fully connected layers. In this embodiment, three convolutional layers are set, and the convolution kernel sizes are , , , the step size is 1, and the padding method is the same to ensure that the size of the feature map after convolution remains unchanged. Each convolution layer is followed by a RelU activation function to increase the nonlinear expression ability of the model. Then set up two pooling layers, using the maximum pooling method, and the pooling kernel size is , with a step size of 2, downsamples the feature map to reduce the amount of data while retaining the main features. Finally, connect two fully connected layers to convert the feature map after convolution and pooling into a building performance vector , such as indoor comfort score (taking into account factors such as temperature, humidity, ventilation, etc.), building cost estimation and other performance indicators.
[0053] During the training process, the collected data samples of 500 residential buildings were divided into training data and test data in a ratio of 7:3. The mean square error loss function was used. and cross entropy loss function As two objectives in the bi-objective optimization problem, is the sample size, is the actual building performance value, is the model prediction value, is the error value obtained by the mean square error loss function, is the error value obtained by the fork entropy loss function. The parameters of the convolutional neural network are conditioned using the stochastic gradient descent algorithm. Updates are made and parameters are continuously adjusted to minimize the loss function. Embodiment 3
[0054] Example of building a multi-objective comprehensive model: Still taking the above residential building as an example, a multi-objective comprehensive model is constructed. After completing the training of the mapping model, a variational autoencoder (VAE) is configured. The encoder of VAE consists of 2 fully connected layers, which maps the building design parameters and building performance data to the latent space. Assuming that the dimension of the latent space is 10, the encoder compresses the high-dimensional data into a 10-dimensional latent space to extract the key features of the data. The decoder also consists of 2 fully connected layers, which reconstructs the latent space vector into building design parameters and building performance data.
[0055] When training VAE, the mean square error loss function is used to measure the difference between the reconstructed data and the original data, and the KL divergence is introduced to constrain the distribution of the latent space to make it close to the standard normal distribution. During the training process, the network parameters of VAE are continuously adjusted. When the loss value change of 5 consecutive rounds of training is less than 0.01, it is considered that the preset training end condition is met. At this time, the encoder in VAE is used as the distribution expression model.
[0056] The outputs of the trained mapping model and distribution expression model are incorporated into a long short-term memory network layer (LSTM). LSTM can effectively process time series data and capture long-term dependencies in the data. In this embodiment, the number of hidden units in the LSTM layer is set to 64. The outputs of the mapping model and distribution expression model are further processed by the LSTM layer, and the relationship between multiple targets is comprehensively considered to obtain a multi-target comprehensive model.
[0057] When training the multi-objective comprehensive model, the mean square error loss function and the cross entropy loss function are used again as the two objectives in the dual-objective optimization problem. The multi-objective particle swarm optimization algorithm is used to obtain the non-dominated solution set boundary of the corresponding mapping model training error. In the multi-objective particle swarm optimization algorithm, the number of particles is set to 50 and the inertia weight is The learning factor decreases linearly from 0.9 to 0.4. .
[0058] After 100 iterations, the boundary of the non-dominated solution set was obtained. Then, the analytic hierarchy process was used to select 10 candidate solution sets from the boundary. According to the final performance of the candidate solution set on the predicted data, such as the predicted building energy consumption, comfort and other indicators, the network parameters of the multi-objective comprehensive model were determined. Finally, a set of network parameters with the best performance on the test data was selected from different solution sets to complete the construction of the multi-objective comprehensive model. Embodiment 4
[0059] Implementation example of non-dominated solution set boundary solution and candidate solution set selection: Taking the design optimization of a large office building as an example, the boundary of the non-dominated solution set of the multi-objective comprehensive model is solved and the candidate solution set is selected. In the design of this office building, three objectives are considered: building energy consumption, indoor lighting effect and building cost. The input of the multi-objective comprehensive model is the design parameters of the building, such as the shape coefficient of the building, the area ratio of the windows, the type of lighting system, etc., and the output is the corresponding building energy consumption value, average daylighting factor and building cost estimation value.
[0060] The fast non-dominated sorting algorithm is used to solve the boundary of the non-dominated solution set. Assume that in a certain iteration, there are 100 architectural design solutions (i.e., the output individuals of the multi-objective comprehensive model). For each individual , calculate its dominated number and dominating set For example, individual The building energy consumption is The average daylight factor is The construction cost is , if there is another individual , its energy consumption , daylight factor And the construction cost , and at least one inequality is strictly true, then the individual Individual dominate, Add 1 and set the individual Add Individual The dominating set .when When Add to the first non-dominated layer . Remove The dominant relationship of individuals in the group to other individuals, recalculate the number of dominated individuals in the group , repeat the above process to get all non-dominated layers , the individuals in these non-dominated layers constitute the boundary of the non-dominated solution set.
[0061] After obtaining the boundary of the non-dominated solution set, the analytic hierarchy process is used to select the candidate solution set that matches the target demand. Assume that the owner of the office building pays more attention to reducing building energy consumption, followed by ensuring good lighting effects, and is relatively less sensitive to building costs. Construct a hierarchical analysis model, the target layer is to select the optimal building design scheme, and the criterion layer is building energy consumption, indoor lighting effects, and building costs. Determine the relative importance of each criterion through expert scoring and construct a judgment matrix. For example, for building energy consumption and indoor lighting effects, experts believe that the importance of building energy consumption is twice that of indoor lighting effects, then the corresponding element in the judgment matrix is 2; for building energy consumption and building costs, experts believe that the importance of building energy consumption is three times that of building costs, then the corresponding element is 3. And so on, construct a complete judgment matrix. By calculating the eigenvector and maximum eigenvalue of the judgment matrix, the weight of each criterion is calculated. It is assumed that the weight of building energy consumption is 0.5, the weight of indoor lighting effects is 0.3, and the weight of building costs is 0.2. Then, for each individual in the non-dominated solution set boundary, the comprehensive score is calculated based on its values in the three indicators of building energy consumption, indoor lighting effect and building cost, combined with the weights. The building energy consumption is The average daylight factor is The construction cost is , then the calculation formula for its comprehensive score is: ; in, is the comprehensive score, , are the maximum and minimum values of building energy consumption in the boundary of the non-dominated solution set, , is the maximum and minimum value of the average daylight factor, , are the maximum and minimum values of the construction cost. Sort the individuals from high to low according to the comprehensive score, and select the first 10 individuals as the candidate solution set. Embodiment 5
[0062] Example of using finite element analysis software to simulate the design: Take the design of a small gymnasium as an example to carry out simulation optimization. After obtaining the candidate solution set, the building performance simulation method based on finite element analysis is used to obtain a new building design sample. Assume that one of the design schemes in the candidate solution set is: the main building adopts a steel structure, the roof is a grid structure, the exterior wall adopts materials with good thermal insulation performance, and the windows adopt double-layer Low-E glass.
[0063] The design scheme was simulated using finite element analysis software. First, a three-dimensional model of the gymnasium was established, the building structure was discretized, and the finite element mesh was divided. For structural analysis, the mechanical performance parameters of the material, such as the elastic modulus and yield strength of the steel, were set, external loads such as gravity load and wind load were applied, the stress conditions of the building under different working conditions were simulated, the stress and strain distribution cloud map of the structure was obtained, and the safety of the structure was judged. For thermal analysis, parameters such as the thermal conductivity coefficient of the building materials were set, the heat transfer process between indoor and outdoor was simulated, the indoor temperature distribution was calculated, and the thermal insulation performance of the building was evaluated. For acoustic analysis, parameters such as the sound absorption coefficient of the indoor surface material were set, the propagation and reflection process of sound in the room was simulated, and the indoor acoustic effects, such as reverberation time and other indicators, were analyzed.
[0064] Through simulation, various performance indicators of the design scheme were obtained, such as the maximum structural stress of 150MPa (less than the yield strength of steel 345MPa, meeting the structural safety requirements), the average indoor temperature under air-conditioning conditions in summer (Meet the design requirements ), but the indoor reverberation time is (Design requirements are , does not meet the acoustic requirements). Since the specified performance (acoustic performance) does not reach the expected target performance value of the design, the new building design sample obtained by simulation is used as a data sample, and its design parameters and performance data are extracted and input into the multi-objective comprehensive model. The steps of solving the boundary of the non-dominated solution set, selecting the candidate solution set and simulating are performed again. After 3 iterations of optimization, a new building design sample is finally obtained, and its various performance indicators all reach the expected target performance value of the design. The simulation optimization process is stopped, and the design scheme is determined to be the final optimized design scheme.
[0065] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0066] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An architectural design optimization method based on artificial intelligence, characterized in that: The method comprises: S1. Collect data samples including building design parameters and building performance; S2. Use the convolutional neural network in deep learning to build a mapping model that can map architectural design parameters to building performance. The input of the mapping model is the architectural design parameter vector , the output is the building performance vector , its mathematical expression is ,in are the parameters of the convolutional neural network; S3. A variational autoencoder is used to construct a distribution expression model that can map the distribution relationship between building design parameters and building performance. The encoder of the variational autoencoder maps the input building design parameters and building performance data to a latent space, and the decoder reconstructs the latent space vector into building design parameters and building performance data. S4, incorporating the outputs of the distribution expression model and the mapping model into at least one long short-term memory network layer to obtain a multi-objective comprehensive model; S5, solving the non-dominated solution set boundary of the multi-objective comprehensive model, and selecting a candidate solution set matching the target requirement in the non-dominated solution set boundary according to the hierarchical analysis method; S6. Obtain a new building design sample on the candidate solution set through the building simulation software. If the specified performance of the new building design sample reaches the expected target performance value of the design, stop the simulation. Otherwise, use the new building design sample as a data sample and input the matching information into the multi-objective comprehensive model, and continue to execute S5 to S6.
2. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: The input of the multi-objective comprehensive model is a plurality of design parameters of the building, and the output is a plurality of performance indicators of the building.
3. The architectural design optimization method based on artificial intelligence according to claim 2 is characterized in that: The non-dominated solution set boundary is composed of a set of non-dominant building performance index values, and is obtained by optimizing and solving the multi-objective comprehensive model through the multi-objective particle swarm optimization algorithm; in the multi-objective particle swarm optimization algorithm, the position of each particle i represents a set of architectural design parameters, the speed of each particle i The update formula is: ; in, Indicates that particle i is The speed of the moment; Indicates that particle i is The speed at the moment, w is the inertia weight, and is the learning factor, and is a random number between [0, 1], is the historical optimal position of the particle, is the global optimal position; position The update formula is: ; in, Indicates that particle i is The location at the moment, Represents particle i in The position at the moment.
4. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: The construction of the multi-objective comprehensive model includes: Use a convolutional neural network to fit the data samples collected by S1, and divide the data samples into training data and test data in a ratio of 7:3; When learning and training the network parameters of the multi-objective comprehensive model based on the training data, the mean square error loss function and the cross entropy loss function are used as the two objectives in the dual-objective optimization problem; and the multi-objective particle swarm optimization algorithm is used to obtain the non-dominated solution set boundary of the corresponding mapping model training error, and then the hierarchical analysis method is used to select a specified number of candidate solution sets from the non-dominated solution set boundary of the mapping model training error, and the network parameters of the mapping model are determined according to the final performance of the candidate solution sets on the predicted data; According to the performance of the multi-objective comprehensive model corresponding to different solution sets on the test data, a set of network parameters with the best performance is finally selected to obtain a trained mapping model; Configure a variational autoencoder that can characterize the distribution of original data samples, and train network parameters based on the training data. When the preset training end conditions are met, use the encoder in the variational autoencoder as a distribution expression model; The output of the distribution expression model and the trained mapping model are then incorporated into at least one long short-term memory network layer to obtain a multi-objective comprehensive model.
5. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: The data samples collected in step S1 and the non-dominated solution set boundary in step S5 have the same physical index.
6. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: In step S5, a fast non-dominated sorting algorithm is used to solve the boundary of the non-dominated solution set of the multi-objective comprehensive model; in the fast non-dominated sorting algorithm, for each individual q, its dominated number is calculated. and dominating set ,like , then add it to the first non-dominated layer , then remove The dominance relationship of individuals in the layer to other individuals is updated, and the number of dominated individuals is updated. This process is repeated to obtain all non-dominated layers.
7. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: In step S6, the simulation method for obtaining the new building design sample is a building performance simulation method based on finite element analysis.
8. The architectural design optimization method based on artificial intelligence according to claim 1 is characterized in that: In the step S1, data samples including building design parameters and building performance are collected based on building design documents and measured data of past building projects.
9. The artificial intelligence-based architectural design optimization method according to claim 8, characterized in that: In the step S1, Kalman filtering denoising is performed on the collected data samples.
10. An artificial intelligence-based architectural design optimization system, characterized in that: The system comprises: A data collection module is used to collect data samples containing building design parameters and building performance based on building design documents and measured data of past building projects, and perform Kalman filtering and denoising on the collected data samples; The mapping model construction module uses the convolutional neural network in deep learning to build a mapping model that can map building design parameters to building performance. The input of the mapping model is the building design parameter vector , the output is the building performance vector , its mathematical expression is ,in are the parameters of the convolutional neural network; The distribution expression model construction module uses a variational autoencoder to construct a distribution expression model that can map the distribution relationship between building design parameters and building performance. The encoder of the variational autoencoder maps the input building design parameters and building performance data to the latent space, and the decoder reconstructs the latent space vector into building design parameters and building performance data. Multi-objective comprehensive model construction module: the outputs of the distribution expression model and the mapping model are merged into at least one long short-term memory network layer to obtain a multi-objective comprehensive model; the specific construction process is to use a convolutional neural network to fit the data samples collected by the data collection module, and divide the data samples into training data and test data according to a certain ratio; when learning and training the network parameters of the multi-objective comprehensive model based on the training data, the mean square error loss function and the cross entropy loss function are used as the two objectives in the dual-objective optimization problem; and the multi-objective particle swarm optimization algorithm is used to obtain the non-dominated solution set boundary of the corresponding mapping model training error, and then the hierarchical analysis method is used to select a specified number of candidate solution sets in the boundary of the mapping model training error, and the network parameters of the mapping model are determined according to the final performance of the candidate solution sets on the predicted data; according to the performance of the multi-objective comprehensive model corresponding to different solution sets on the test data, a set of network parameters with the best performance is finally selected to obtain a trained mapping model; a variational autoencoder that can characterize the distribution of the original data sample is configured, and the network parameters are trained based on the training data. When the preset training end conditions are met, the encoder in the variational autoencoder is used as the distribution expression model; A solution set processing module is used to solve the non-dominated solution set boundary of the multi-objective comprehensive model, and select a candidate solution set matching the target requirement in the non-dominated solution set boundary according to the hierarchical analysis method; The simulation optimization module obtains a new building design sample on the candidate solution set through a building performance simulation method based on finite element analysis. If the specified performance of the new building design sample reaches the expected target performance value of the design, the simulation is stopped. Otherwise, the new building design sample is used as a data sample and the matching information is input into the multi-objective comprehensive model, and the operations of the solution set processing module and the simulation optimization module are continued.
Citation Information
Patent Citations
Systems and methods of automated design and spatial allocation of buildings
CA3136181A1
Optimal allocation method of basin water resources based on multi-objective chaotic genetic algorithm
CN107480813A
SOM-based building multi-objective optimization design decision support method
CN110569616A
Improved multi-objective optimization method based on NSGA-II
CN115511039A
Zero-carbon building multi-objective optimization method based on machine learning hybrid modeling
CN117574785A
Cited By
Green intelligent building multi-factor fusion planning design method
CN120408812A
Pre-training artificial intelligence method for automatically generating BIM, electronic equipment and storage medium
CN120562024A
Building facade optimization method and system based on sustainable analysis
CN121211570A
