Multi-objective optimization method and system for urban block form

Abstract

The invention discloses a multi-objective optimization method for the form of an urban block, and the method comprises the following steps: S100, constructing a parameterized model, carrying out the experimental design, defining a group of geometric parameters used for describing the form of the urban block, and generating a plurality of sample points in the value range of the geometric parameters through employing a sampling method; s200, bidirectional coupling numerical simulation is executed, a data set is generated, at least one thermal comfort evaluation index and one air quality evaluation index are output, and therefore the data set containing input parameters and output indexes is formed; step S300, constructing and verifying a machine learning agent model to obtain an agent model capable of rapidly predicting the evaluation index; step S400, executing multi-objective optimization search, and automatically searching to enable the thermal comfort evaluation index and the air quality evaluation index to simultaneously achieve a better Pareto optimal solution set; and step S500, analyzing and outputting an optimization result, extracting an optimization value range of the geometric parameters, and generating a design guide.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban planning and architectural environment, and particularly relates to a method and system for multi-objective optimization of urban residential block form by integrating computational fluid dynamics (CFD) and machine learning techniques to simultaneously improve outdoor thermal comfort and air quality. BACKGROUND

[0002] Global rapid urbanization and climate change exacerbate urban environmental pressures, among which urban heat island effect and air pollution (especially PM2.5 pollution) are two major challenges. The form of urban blocks, including parameters such as building height, length, width, spacing, and orientation, has a significant and complex impact on local microclimate (thermal environment) and pollutant dispersion (air quality).

[0003] In the prior art, the optimization research on the form of urban blocks has the following main problems and defects:

[0004] Single target limitation: Most studies only focus on a single environmental target, such as optimizing thermal comfort or air quality separately. However, these two targets often conflict with each other, and optimizing one target separately may lead to the deterioration of the other target.

[0005] Ignoring coupling effects: There is a dynamic two-way coupling interaction between thermal environment and air quality. Traditional decoupling analysis methods or one-way models cannot accurately capture this complex interaction, which may lead to the failure of optimization results in practical applications.

[0006] High computational cost: Although CFD simulation based on physics is accurate, it consumes huge computational resources and takes a long time, making it impractical to directly use it in multi-objective optimization algorithms that require thousands of iterations.

[0007] Lack of practical design guidelines: Many studies are based on idealized simplified models and fail to fully consider the constraints in actual planning, making it difficult to directly translate their conclusions into quantitative and operational design guidelines that have clear guiding significance for urban planners and designers.

[0008] Therefore, there is an urgent need in the field for a technical means that can efficiently and accurately consider both thermal comfort and air quality, and output optimization design schemes that meet the constraints of actual planning. SUMMARY

[0009] The present application aims to solve the above-mentioned defects in the prior art, and provides a city block form multi-objective optimization method and system which is computationally efficient, accurate in results, and can output practical design guidelines. Specifically, the first technical problem to be solved by the present application is: how to provide an automated method that can efficiently and accurately optimize both urban block thermal comfort and air quality, while fully considering the bidirectional coupling effect between the two and actual planning constraints, and ultimately generating a quantitative design guideline.

[0010] To solve the above technical problems, the present application adopts the following technical solutions:

[0011] A city block form multi-objective optimization method, comprising the following steps: step S100, constructing a parameterized model and performing experimental design: including defining a set of geometric parameters for describing the city block form, the geometric parameters including building length, building width, building height, building orientation angle, building lateral spacing, and building longitudinal spacing, and using a sampling method to generate a plurality of sample points within the value range of the geometric parameters;

[0012] Step S200, performing bidirectional coupling numerical simulation and generating a data set: including performing computational fluid dynamics simulation coupled with the interaction of thermal environment and particulate matter concentration on each block form represented by the sample points generated in step S100, and outputting at least one thermal comfort evaluation index and one air quality evaluation index, thereby forming a data set containing input parameters and output indicators;

[0013] Step S300, constructing and verifying a machine learning agent model: including using the data set generated in step S200 to train a machine learning algorithm to establish a mapping relationship from the geometric parameters to the thermal comfort evaluation index and the air quality evaluation index, thereby obtaining an agent model that can quickly predict the evaluation index;

[0014] Step S400, performing multi-objective optimization search: including using the agent model obtained in step S300 as a target function calculator, integrated into a multi-objective optimization algorithm, to automatically search for a Pareto optimal solution set that makes the thermal comfort evaluation index and the air quality evaluation index simultaneously better under the condition of meeting the preset planning constraints;

[0015] Step S500, analyzing and outputting the optimization results: including analyzing the Pareto optimal solution set obtained in step S400, extracting the optimization value range of the geometric parameters, and generating a design guideline.

[0016] The "computational fluid dynamics simulation coupled with the interaction between thermal environment and particulate matter concentration" in step S200 includes: a first-direction coupling process, which includes introducing the Businesk approximation into the flow field control equation to simulate the thermal buoyancy effect and considering the thermophoretic force effect in the particulate matter force balance equation to simulate the influence of the thermal environment on the particulate matter diffusion and transport process; and a second-direction coupling process, which includes using a user-defined function to dynamically adjust the atmospheric solar radiation absorption coefficient and scattering coefficient according to the local particulate matter concentration calculated in real time, to simulate the feedback effect of particulate matter concentration on solar radiation transmission and thermal environment.

[0017] The thermal comfort evaluation index is the Universal Thermal Climate Index (UTCI), and the air quality evaluation index is the Air Quality Index (AQIPM2.5) based on PM2.5 concentration.

[0018] The machine learning algorithm described in step S300 is a nonparametric regression algorithm, specifically the locally weighted scatter smoothing method LEOSS.

[0019] The multi-objective optimization algorithm mentioned in step S400 is a multi-objective genetic algorithm.

[0020] The sampling method described in step S100 is Latin hypercube sampling.

[0021] The preset planning constraints mentioned in step S400 include one or more of the following: floor area ratio, building density, building setback distance, and minimum building spacing.

[0022] This invention also proposes a multi-objective optimization system for urban block morphology to perform any of the methods described above, comprising: a parameterization and sampling unit for receiving geometric parameters and their ranges input by a user and performing sampling to generate sample points; a bidirectional coupled simulation unit for simulating and calculating the sample points and outputting a dataset; a surrogate model construction unit for receiving the dataset and training a machine learning model, and outputting a trained surrogate model; a multi-objective optimization unit for integrating the surrogate model and the optimization algorithm, performing optimization search and outputting a Pareto optimal solution set; and a result analysis and output unit for receiving the Pareto optimal solution set, performing analysis, and generating a visual design guide.

[0023] The bidirectional coupling simulation unit is a computational fluid dynamics software module that integrates user-defined functions (UDFs) to realize the coupling process in the second direction.

[0024] The results analysis output section also includes a sensitivity analysis unit, used to calculate the Spearman rank correlation coefficient to quantify the degree of influence of each of the geometric parameters on the evaluation index.

[0025] Compared with existing technologies, the multi-objective optimization method and system for urban block morphology provided by this invention have the following beneficial effects:

[0026] This invention constructs a computational fluid dynamics simulation coupling the interaction between the thermal environment and particulate matter concentration, uses the generated dataset to train a machine learning surrogate model, and finally integrates this surrogate model into a multi-objective optimization algorithm for search. This framework of "high-precision CFD simulation generating data + efficient machine learning model replacing evaluation" firstly solves the feasibility problem of directly using computationally expensive CFD simulations for optimization algorithms requiring numerous iterations, significantly improving optimization efficiency. Secondly, the bidirectional coupling mechanism introduced in the CFD simulation (thermal buoyancy, thermophoretic force effects, and dynamic feedback of particulate matter concentration on radiation) ensures the physical realism and accuracy of the simulation, providing high-quality training data for the machine learning model, thereby guaranteeing the prediction accuracy of the surrogate model. Finally, through the rapid invocation of the surrogate model by the multi-objective optimization algorithm, the complex solution space can be efficiently explored to find the optimal trade-off between simultaneously improving thermal comfort and air quality (Pareto optimal solution set), and specific, quantifiable morphological parameter design guidelines can be extracted from it. Therefore, this invention successfully achieves an excellent balance between computational efficiency and result accuracy, and provides a practical and automated optimization tool that can guide actual urban planning and design.

[0027] Furthermore, some embodiments of the present invention also have the following beneficial effects:

[0028] The detailed bidirectional coupled CFD simulation ensures that the physical simulation process can more realistically reflect the complex interaction between the thermal environment and air quality, thus providing a key guarantee for the accuracy of the basic data and the effectiveness of the final optimization results.

[0029] The specific evaluation indicators UTCI and AQIPM2.5 enable the optimization objectives to comprehensively and systematically reflect the complex human experience of thermal comfort and air quality, enhancing the practicality and comparability of the optimization results.

[0030] Nonparametric regression algorithms such as LEOSS, as machine learning surrogate models, can better fit the complex nonlinear relationship that may exist between input variables and output targets, thereby improving the predictive and generalization abilities of surrogate models.

[0031] Multi-objective genetic algorithms can effectively handle multi-objective optimization problems and search in parallel to obtain a well-distributed Pareto optimal solution set, providing decision-makers with a wealth of choices.

[0032] The Latin hypercube sampling method can efficiently and uniformly generate sample points within the design variable space, maximizing the coverage of the parameter space with fewer simulations, thereby improving the representativeness of the dataset and the efficiency of surrogate model construction.

[0033] The actual planning constraints make the optimization search process and its results more in line with the actual laws and requirements of urban construction, thus enhancing the practical value and feasibility of the output results of this invention.

[0034] The entire optimization process is modularized, enabling automated processing from parameter input to result output, thus improving the ease of use of the method and the efficiency of engineering applications.

[0035] Integrating UDFs into the bidirectional coupling simulation unit provides a flexible and powerful technical means to achieve complex physical process coupling (such as second-direction coupling) beyond the default software functions.

[0036] Adding a sensitivity analysis unit can quantify the impact of various design parameters on environmental performance indicators, providing priority guidance for design decisions and making the generated design guidelines more insightful. Attached Figure Description

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] Figure 1-1 This is a flowchart summarizing a multi-objective optimization method for urban street morphology according to the present invention.

[0039] Figure 1-2 This is a flowchart illustrating the specific calculation process of multi-objective optimization based on numerical simulation and response surface methodology provided in this embodiment of the invention.

[0040] Figure 2 This is a schematic diagram of the ideal residential block layout and parameterized control variables provided in the embodiments of the present invention.

[0041] Figure 3 This is a flowchart of a two-way coupling numerical simulation between the outdoor thermal environment and PM2.5 provided in an embodiment of the present invention.

[0042] Figure 4 This is a land use division and geometric model diagram of the case study area provided in the embodiments of the present invention.

[0043] Figure 5 This is a response surface model (RSM) accuracy verification diagram (predicted value vs. actual value) provided in the embodiments of the present invention.

[0044] Figure 6 These are the morphological parameters and the response surface model (RSM) of UTCI provided in the embodiments of the present invention ((a) L and S, (b) W and S', (c) A and H).

[0045] Figure 7 The morphological parameters and response surface model (RSM) of AQIPM2.5 provided in this embodiment of the invention are shown in the figures ((a) L and S, (b) W and S', (c) A and H).

[0046] Figure 8 This is a diagram of the multi-objective optimization iterative convergence process provided in an embodiment of the present invention.

[0047] Figure 9 This is a distribution diagram of dominated and non-dominated solutions (Pareto fronts) of UTCI and AQIPM2.5 provided in the embodiments of the present invention.

[0048] Figure 10 This is a parameter interval distribution diagram corresponding to the Pareto front solution provided in an embodiment of the present invention.

[0049] Figure 11 This is a graph showing the sensitivity index (Spearman coefficient) of the physical environment (UTCI / AQIPM2.5) to morphological parameters provided in this embodiment of the invention.

[0050] Figure 12 This is a comparison chart of simulated air temperature contour lines before and after optimization in the case study area provided in this embodiment of the invention.

[0051] Figure 13 This is a comparison chart of simulated wind speed contour lines before and after optimization in the case study area provided in this embodiment of the invention.

[0052] Figure 14 This is a comparison of the simulated PM2.5 concentration contour lines before and after optimization in the case study area provided in this embodiment of the invention.

[0053] Figure 15 This is a graph showing the hourly anthropogenic heat emissions from buildings, provided in an embodiment of the present invention.

[0054] Figure 16 This is a comparison chart of simulated air temperature and measured values ​​provided in an embodiment of the present invention.

[0055] Figure 17 This is a comparison chart of simulated wind speed and measured value provided in an embodiment of the present invention.

[0056] Figure 18 This is a comparison between simulated PM2.5 concentration and measured values ​​provided in the embodiments of the present invention.

[0057] Figure 19 This is a comparison chart of simulated air temperature and measured results provided in an embodiment of the present invention.

[0058] Figure 20 This is a comparison chart of radiation temperature at measuring points under different models provided in the embodiments of the present invention.

[0059] Figure 21 This is a comparison chart of wind speed simulation and actual measurement results provided in the embodiments of the present invention.

[0060] Figure 22This is a comparison chart of turbulence intensity at measurement points under different models provided in the embodiments of the present invention.

[0061] Figure 23 This is a comparison chart of simulated and measured PM2.5 concentrations provided in the embodiments of the present invention.

[0062] Figure 24 This is a schematic diagram of the surface temperature at each of the nine measuring points in the TWC model provided in this embodiment of the invention.

[0063] Figure 25 This is a schematic diagram of the surface temperature at each measuring point at time 13 of the TWC model provided in this embodiment of the invention.

[0064] Figure 26 This is a schematic diagram of the streamlines and PM2.5 concentrations of different models at 9 different time points provided in the embodiments of the present invention.

[0065] Figure 27 This is a schematic diagram of the streamlines and PM2.5 concentrations of different models at 13 time points provided in the embodiments of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0067] Implementation Concept

[0068] The core of this invention lies in providing an automated framework that integrates bidirectional coupled CFD simulation and machine learning proxy models for efficiently and accurately optimizing urban block morphology to simultaneously improve thermal comfort (UTCI) and air quality (AQIPM2.5), and outputting quantitative design parameter guidelines. The following embodiments will illustrate the key implementation steps of this framework in detail.

[0069] Example 1: A multi-objective optimization method for urban street block morphology

[0070] The core steps of this embodiment include: parameterization and experimental design, bidirectional coupled CFD simulation and dataset generation, machine learning surrogate model construction and verification, multi-objective optimization search, result analysis and design guide generation.

[0071] Specific defects in the identification and resolution of technical problems: This embodiment aims to solve the problems mentioned in the background technology, such as low computational efficiency, neglect of bidirectional coupling effect, and lack of practical guidelines, and achieves efficient and accurate multi-objective optimization through an automated process.

[0072] Detailed Technical Solution: Step S100, Constructing a Parametric Model and Designing Experiments: This step corresponds to "Constructing a parametric model and designing experiments: including defining a set of geometric parameters to describe the morphology of urban blocks, wherein the geometric parameters include building length, building width, building height, building orientation angle, building lateral spacing, and building longitudinal spacing, and using a sampling method to generate multiple sample points within the value range of the geometric parameters." The parametric control variables for the block morphology described in this step can be found in [reference needed]. Figure 2 The diagram shown is shown in the image.

[0073] Components and Relationships: First, define the geometric parameters describing the morphology of a standard urban block (e.g., a 6x6 building array). These parameters are building length (L), building width (W), building height (H), building orientation angle (A), building lateral spacing (S), and building longitudinal spacing (S'). Each parameter needs to be set with a reasonable range of values ​​according to the planning conventions of the target city (e.g., Shenzhen), for example: L: [30, 60] meters, W: [20, 40] meters, H: [30, 100] meters, A: [0, 180] degrees, S: [20, 80] meters, S': [20, 80] meters.

[0074] Implementation steps: Latin hypercube sampling (LHS) is used to generate N sample points (N is, for example, 200) within the aforementioned six-dimensional parameter space. This sampling method ensures that the samples are uniformly distributed within the parameter space. This step can be implemented using Python's pyDOE library or MATLAB's lhsdesign function. After generating the sample points, each sample point represents a specific street block morphological configuration.

[0075] Step S200, Perform a two-way coupled numerical simulation and generate a dataset: This step corresponds to "Perform a two-way coupled numerical simulation and generate a dataset: including performing a computational fluid dynamics simulation coupled with the interaction between thermal environment and particulate matter concentration on the street morphology represented by each sample point generated in step S100, and outputting at least one thermal comfort evaluation index and one air quality evaluation index, thereby forming a dataset containing input parameters and output indicators." The two-way coupled numerical simulation process implemented in this step can be found in [link to relevant documentation]. Figure 3 An example of the geometric model of the case study area can be found in [reference needed]. Figure 4 .

[0076] Components and Relationships: The core of this step is computational fluid dynamics (CFD) simulation software (such as ANSYS Fluent). For each block morphology geometry model generated in step S100, a corresponding CFD calculation model needs to be established, including the computational domain, mesh, physical model settings, and boundary conditions.

[0077] Working principle: CFD simulation predicts the velocity field, temperature field, and pollutant concentration field within a block by solving a set of partial differential equations controlling fluid flow, heat transfer, and mass transfer. The key to this embodiment lies in achieving bidirectional coupling:

[0078] First directional coupling (thermal environment → particulate matter): Enable the Businesk approximation in the flow field control equations to account for the thermal buoyancy effect; in the discrete phase model (DPM) settings, enable the thermophoretic force option for PM2.5 particles to simulate the effect of temperature gradient on particle motion.

[0079] The second directional coupling (particulate matter → thermal environment): By writing a user-defined function (UDF), the atmospheric absorption and scattering coefficients of solar radiation are correlated with the locally calculated real-time PM2.5 concentration (e.g., coefficient = baseline value + proportionality coefficient × PM2.5 concentration), thereby achieving dynamic feedback of particulate matter concentration on radiative transfer and the thermal environment. This UDF needs to be loaded into the radiation model.

[0080] Implementation steps:

[0081] 1. Geometric Modeling and Meshing: A 3D geometric model is generated using ANSYS DesignModeler or SCDM based on sample point parameters. The computational domain size should follow standards (e.g., 5H from the entrance and sides of the building, 15H from the exit, 6H from the top, where H is the highest building height). An unstructured mesh is used, with refinement applied to the building surface, ground, and near-ground areas, and boundary layer meshes are set. The total mesh size is approximately 30 million elements, and mesh independence verification is performed.

[0082] 2. Physics Model Setup:

[0083] Solver: Pressure-based steady-state solver.

[0084] Turbulence model: RNG k-ε model with buoyancy enabled.

[0085] Radiation model: Discrete coordinate (DO) model, with solar ray tracing enabled, and the above UDF loaded.

[0086] Discrete Phase Model (DPM): Activated, PM2.5 particles (diameter 2.5e-6m, density 2650kg / m³) are injected. 3 Set up particulate sources (such as simulated road emissions) and consider thermophoretic forces.

[0087] Material properties: Set the density, specific heat capacity, thermal conductivity, emissivity, etc. of air, concrete, asphalt, etc.

[0088] 3. Boundary conditions:

[0089] Boundary condition settings: This example selects meteorological data from two typical summer pollution days as the boundary conditions for the CFD simulation. The dates were chosen based on the characteristics of high PM2.5 concentrations and complete local meteorological data. Specific boundary condition parameters are shown in the table below:

[0090]

[0091] Turbulent boundary conditions are calculated based on atmospheric boundary layer theory, and the specific formula is as follows:

[0092] Turbulent kinetic energy: k = u* 2 / √Cμ

[0093] Turbulent dissipation rate: ε=u* 3 / (κ*(z+z0)).

[0094] Where u* is the friction velocity (value 0.16 m / s), κ is the Karman constant (0.4), z is the height, z0 is the ground roughness length (value 0.1 m), and Cμ is the empirical constant (0.09).

[0095] No constant heat flow was set; convective heat transfer was used, with its coefficient calculated using the formula hc = 5.7 + 3.8u (where u is the wind speed). Simultaneously, the building's exterior walls incorporated hourly-varying anthropogenic heat emissions as an internal heat source, with a peak value of 140 W / m². 2 Its 24-hour variation curve is as follows Figure 15 As shown.

[0096] DPM boundaries: Ground and building walls can be set to "trap" or "escape", and exits are set to "escape".

[0097] 4. Solution and Monitoring: The Coupled algorithm is used for solution, with a convergence residual of 1e-6. Multiple monitoring points are set up at pedestrian height (1.5m) to monitor speed, temperature, and PM2.5 concentration until they stabilize.

[0098] 5. Data Output: After simulation convergence, data from each monitoring point is extracted, and the regional average Universal Thermal Climate Index (UTCI) and the Air Quality Index (AQIPM2.5) based on the average PM2.5 concentration are calculated. Finally, a dataset is generated containing N rows (number of samples), 6 input columns (L, W, H, A, S, S'), and 2 output columns (UTCI, AQIPM2.5).

[0099] Step S300, Construct and validate the machine learning surrogate model: This step corresponds to "Construct and validate the machine learning surrogate model: including using the dataset generated in step S200 to train a machine learning algorithm to establish a mapping relationship from the geometric parameters to the thermal comfort evaluation index and the air quality evaluation index, thereby obtaining a surrogate model that can quickly predict the evaluation index." The accuracy validation results of the surrogate model (response surface model) can be found in [reference needed]. Figure 5 The established morphological parameters and evaluation indices UTCI and AQIPM2.5 response surface models (RSM) are respectively referred to [reference needed]. Figure 6 and Figure 7 .

[0100] Components and Relationships: Machine learning libraries are used (such as Python's statsmodels, scikit-learn, or R). The input is the dataset generated in step S200, and the output is two trained surrogate models (one predicting UTCI, and one predicting AQIPM2.5).

[0101] How it works: The machine learning algorithm learns the complex nonlinear mapping relationship between the input parameters (L, W, H, A, S, S') and the output target (UTCI, AQIPM2.5). After training, the surrogate model can complete a prediction in milliseconds, replacing time-consuming CFD simulations.

[0102] Implementation steps:

[0103] 1. Data preprocessing: Randomly divide the dataset into a training set (80%) and a test set (20%). Standardize the input features.

[0104] 2. Model training: Non-parametric regression (NON) method is used.

[0105] 3. Model Validation: Evaluate model performance using a test set. Calculate the coefficient of determination (R²) between predicted values ​​and actual CFD simulation values. 2 Metrics such as root mean square error (RMSE) are used. The surrogate model needs to be tested on a validation set, requiring a coefficient of determination (R²) of 100%. 2 The root mean square error (RMSE) should be less than 5% of the measured range of UTCI and AQIPM2.5, and the scatter plot of the predicted and true values ​​should be evenly distributed near the y=x line.

[0106] Step S400, Perform multi-objective optimization search: This step corresponds to "Perform multi-objective optimization search: including using the surrogate model obtained in step S300 as the objective function calculator, integrating it into the multi-objective optimization algorithm, and automatically searching for a Pareto optimal solution set that simultaneously achieves the thermal comfort evaluation index and the air quality evaluation index under preset planning constraints." The multi-objective optimization iterative convergence process can be found in [link to documentation]. Figure 8 The final Pareto front solution set distribution can be found in [reference needed]. Figure 9 .

[0107] Components and Relationships: Use a multi-objective optimization algorithm library (such as Python's pymoo, DEAP, or MATLAB's GlobalOptimization Toolbox). The optimization algorithm calls the surrogate model trained in step S300 to compute the objective function value.

[0108] Working principle: The multi-objective optimization problem is formalized as: Minimize F(x)=[UTCI(x),AQIPM2.5(x)], where x=[L,W,H,A,S,S']. Optimization algorithms (such as multi-objective genetic algorithms (MOGA), e.g. NSGA-II) evolve multiple solutions in parallel under variable constraints and planning constraints [e.g., floor area ratio FAR=(NLWH) / (plot area) must be between [3.0,6.0], building density ≤25%, S≥24m, S'≥24m, setback distance ≥9m], to find the Pareto optimal solution set.

[0109] Implementation steps:

[0110] 1. Problem Setup: Define design variables and their boundaries (same as step S100). Define the objective function (calculate UTCI and AQIPM2.5 by calling the surrogate model).

[0111] 2. Algorithm Execution: Select the NSGA-II algorithm, set the population size (e.g., 2000) and maximum number of generations (e.g., 500). Run the optimization algorithm.

[0112] 3. Output Results: After the algorithm finishes running, it outputs a Pareto optimal solution set, which is a set of optimal solutions that cannot improve one objective without harming the other.

[0113] Step S500, analyze and output optimization results: This step corresponds to "analyze and output optimization results: including analyzing the Pareto optimal solution set obtained in step S400, extracting the optimized value range of the geometric parameters, and generating design guidelines." The parameter interval distribution corresponding to the Pareto front solution can be found in [reference needed]. Figure 10 The sensitivity analysis results (Spearman coefficient) of morphological parameters to physical environment evaluation indicators can be found in [reference needed].Figure 11 A comparison of simulation results (air temperature, wind speed, PM2.5 concentration) for the case area before and after optimization can be found in the respective sections. Figure 12 , Figure 13 and Figure 14 .

[0114] Components and Relationships: Use data analysis tools or programming languages ​​(such as Python, MATLAB). The input is a Pareto optimal solution set.

[0115] Working principle: Statistical analysis is performed on the Pareto solution set to observe the distribution range of each design parameter in the optimal solution set, thereby extracting the optimal parameter value range. Sensitivity analysis can be performed, such as calculating the Spearman rank correlation coefficient to quantify the degree and direction of influence of each morphological parameter on UTCI and AQIPM2.5.

[0116] Implementation steps:

[0117] 1. Parameter range extraction: Analyze the value distribution of each parameter (L,W,H,A,S,S') in the Pareto solution set and give its optimal value range (e.g., mean ± standard deviation, or 5%-95% quantile).

[0118] 2. Sensitivity analysis: Calculate the Spearman correlation coefficient between each morphological parameter and UTCI and AQIPM2.5 in all Pareto solutions, and sort them.

[0119] 3. Generate Design Guidelines: Integrate the above analysis results into charts and text descriptions to form design recommendations for different planning constraints (such as different floor area ratio ranges). For example: "In high-density development (FAR~6.0), it is recommended that the building height-to-width ratio H / W be maintained at 1.3-1.6, the building orientation be as close as possible to the prevailing wind direction (angle <15°), and priority should be given to ensuring a larger longitudinal spacing S'."

[0120] Effectiveness verification framework: To verify the effectiveness of the method in this embodiment, the following comparisons can be made:

[0121] Comparison objects: 1. Unoptimized original street block design scheme; 2. Scheme considering only a single objective (such as optimizing only thermal comfort); 3. Scheme optimized using uncoupled CFD simulation.

[0122] Verification metrics: UTCI, AQIPM2.5, CFD simulation computation time, total optimization time, and the rationality of the parameter design guidelines.

[0123] Expected Results: The optimized scheme obtained by the method in this embodiment should significantly outperform the original scheme in both UTCI and AQIPM2.5, and is superior to the single-objective scheme in both objectives. Its total optimization time should be much lower than that of the scheme optimized directly using CFD. Comparison with high-precision CFD simulation results verifies that the prediction error of the surrogate model is within acceptable limits, with relative root mean square error (rRMSE) below 8.13% (for AQIPM2.5) and 5.37% (for UTCI), respectively.

[0124] Building materials: Absorption rate 0.6, emissivity 0.7, specific heat capacity 750 J / (kg·K), density 2400 kg / m³ 3 Thermal conductivity is 1.5 W / (m·K).

[0125] Ground / Road: Absorbance 0.6, Emissivity 0.95, Specific Heat Capacity 880 J / (kg·K), Density 2600 kg / m³ 3 Thermal conductivity is 2 W / (m·K).

[0126] The solar radiation absorption coefficient for building surfaces is set at 0.48, and for the ground and roads at 0.8.

[0127] Specific parameter values ​​for boundary conditions:

[0128] Meteorological parameters (for ideal street block optimization): Air temperature 29.97℃, composite wind speed at 10m height 1.59m / s, direct radiation 545.40W / m². 2 Scattered radiation 123.34 W / m 2 The PM2.5 concentration at the entrance was 35.78 ug / m³. 3 .

[0129] Particulate matter properties: diameter 2.5e-6m, density 2650kg / m³ 3 .

[0130] The optimization algorithm parameter settings are based on the following: a multi-objective genetic algorithm (MOGA) is used, with a population size of 2000 and a maximum number of generations of 500. These parameters are standard settings for genetic algorithms, designed to ensure a sufficiently large search space and a convergent generation.

[0131] Example 2: Detailed implementation of bidirectional coupled CFD simulation. The bidirectional coupling mechanism process involved in this example can also be found in [reference needed]. Figure 3 .

[0132] This embodiment is a further refinement of step S200, focusing on the specific settings of the bidirectional coupled CFD simulation.

[0133] Difference from Example 1: This example describes in detail the specific CFD configuration for implementing the bidirectional coupling mechanism, especially the writing and application of UDFs.

[0134] Detailed technical solution: This embodiment is based on step S200 of embodiment 1.

[0135] Implementation of the first-direction coupling process:

[0136] In the Viscous Model settings of ANSYS Fluent, select the “RNG k-epsilon” model and check the “Full Buoyancy Effects” option to enable the Buscous approximation for thermal buoyancy.

[0137] In the Discrete Phase Model settings, activate "Interaction with Continuous Phase". When creating the injection, under the "Physical Models" tab, check "Thermophoretic Force" to account for thermophoretic force effects. Use the default values ​​for the thermophoretic force coefficients.

[0138] Implementation of the second-direction coupling process:

[0139] This process requires the use of user-defined functions (UDFs). UDFs must be written in C and use the DEFINE macro provided by Fluent.

[0140] The main function of UDF is to read the PM2.5 concentration in each calculation cell in real time (obtained through the DPM concentration field), and then dynamically calculate the radiation absorption coefficient and scattering coefficient of that cell according to preset formulas. For example: absorption_coeff=C_ABS0+C_ABS_PM*C_PM(cell); scattering_coeff=C_SCA0+C_SCA_PM*C_PM(cell); where C_ABS0 and C_SCA0 are the absorption and scattering coefficients under clean air background, respectively, with values ​​of 0.01km. -1 and 0.05km -1 C_ABS_PM and C_SCA_PM are the contribution coefficients of PM2.5, where C_PM(cell) is the PM2.5 concentration in the cell. C_ABS_PM and C_SCA_PM are not fixed constants, but are calculated using the formula (3*q / (2*dp*ρp))*fraction, where the dimensionless extinction efficiency q is 2.29772E-06, dp is the particle diameter, and ρp is the particle density. The scattering component (fraction) accounts for 0.9, and the absorption component accounts for 0.1.

[0141] This UDF needs to be compiled and hooked into Fluent. The hook point is usually chosen in "Define->User-Defined->Function Hooks->Adjust (every iteration)" or a hook point specifically defined for the radiation model to ensure that the radiation properties are updated in each iteration step.

[0142] Implementation steps: After completing the conventional CFD settings in step S200 of Example 1, strictly follow the above instructions to configure the buoyancy effect and thermophoretic force effect, and write, compile, and mount the UDF to implement the second-direction coupling. The remaining mesh generation, boundary conditions, and solution settings are the same as in Example 1.

[0143] Effect verification framework: To verify the necessity of bidirectional coupling, a comparative simulation can be set up:

[0144] Comparative examples: 1. Completely uncoupled simulation (buoyancy and thermophoretic force off, radiation coefficient constant); 2. Simulation with coupling only in the first direction; 3. Simulation with coupling only in the second direction.

[0145] Validation metrics: Compare the simulation results of temperature and PM2.5 concentration at key monitoring points under different coupling modes. Among them:

[0146] UDF Relationship Basis: The relationships in UDFs are based on Ogle's fitting formula:

[0147]

[0148] This formula establishes a quantitative relationship between PM2.5 concentration Cp and extinction coefficient δe,λ. The reference is (ZHAO Y, AMBROSE RP KA real-time method for sensing suspended dust concentration from the light extinction coefficient[J]. Journal of Loss Prevention in the Process Industries, 2020, 67: 104-242.). The extinction coefficient is decomposed into the absorption coefficient and the scattering coefficient, the ratio of which (0.1 and 0.9, respectively) is determined by the single-scatter albedo (SSA = 0.9), which is selected based on the references (HUANG X, YUN H, GONG Z, et al. Source apportionment and secondary organic aerosolestimation of PM2.5 in an urban atmosphere in China[J]. Science China Earth Sciences, 2014, 57(6): 1352-1362. HE LY, HUANG XF, XUE L, et al. Submicron aerosol analysis and organic source apportionment in an urban atmosphere in Pearl River Delta of China using high-resolution aerosol mass spectrometry[J]. Journal of Geophysical Research: Atmospheres, 2011, 116(D12): 1-15. LIU T, ABBATTJ P D. Oxidation of sulfur dioxide by nitrogen dioxide accelerated at the interface of deliquesced aerosol particles[J]. Nature Chemistry, 2021, 13(12): 1173-1177.DENG T, ZOU Y, HU S, et al. Study on the characteristics of actinicradiation and direct aerosol radiation effects in the Pearl River Deltaregion[J]. Atmospheric Environment, 2023, 309: 119937. LI Z, ECK T, ZHANG Y, et al. Observations of residual submicron fine aerosol particles related to cloud and fog processing during a major pollution event in Beijing [J]. Atmospheric Environment, 2014, 86: 187-192. LI J,CARLSON B,LACIS A.Using Single ScatteringAlbedo Spectral Curvature to Characterize East Asian Aerosol Mixtures[J].Journal of Geophysical Research:Atmospheres,2015,120:2037-2052. ).

[0149] Thermophoretic coefficient: Thermophoretic mobility is usually not a fixed constant, but rather a function of multiple physical properties. Its determination is relatively complex. This example uses the built-in thermophoretic model in ANSYS Fluent.

[0150] Validation Data: This validation was based on long-term field measurement data obtained in a street in Shenzhen, and the constructed two-way coupled model was rigorously verified and validated. Hourly measurement data for a specific day was used to quantitatively validate the model's predicted temperature, wind speed, and PM2.5 concentration. The results show that the model has high prediction accuracy, and the coefficients of determination (R²) for temperature, wind speed, and PM2.5 concentration are [value missing]. 2 The values ​​reached 0.83, 0.73, and 0.79 respectively, with root mean square errors (RMSE) of 1.14℃, 0.10m / s, and 4.79μg / m. 3 (See details) Figure 16 , Figure 17 , Figure 18 (and related statistical analysis).

[0151] Further, using measured data, a systematic comparative analysis was conducted between the two-way coupled model and several simplified models (independent model, unidirectional coupled model). The comparison results show that the two-way coupled model can more accurately simulate the spatiotemporal distribution patterns of street thermal environment parameters (air temperature, radiation temperature, turbulence intensity) and PM2.5 concentration under different weather conditions (see [reference]). Figure 19 , Figure 20 , Figure 21 , Figure 22 , Figure 23 , Figure 24 , Figure 25 , Figure 26 , Figure 27 Especially during periods of strong solar radiation, high temperature, and high PM2.5 concentration, the advantages of considering the two-way coupling effect are more significant, fully verifying the necessity and accuracy of the thermal buoyancy effect and PM2.5 radiation feedback mechanism in the model.

[0152] Example 3: Optimization implementation based on multi-objective genetic algorithm. The optimization iteration process and results of this example can be found in [reference needed]. Figure 8 and Figure 9 .

[0153] This embodiment is a further refinement of step S400, focusing on the application of the multi-objective genetic algorithm.

[0154] Difference from Example 1: This example describes in detail the specific process and parameter settings for optimization search using the multi-objective genetic algorithm (NSGA-II).

[0155] Detailed technical solution: This embodiment corresponds to step S400 of embodiment 1.

[0156] Platform and tools optimization: The NSGA-II algorithm was implemented using the Python optimization library pymoo.

[0157] Problem definition:

[0158] Design variables: x = [x1, x2, x3, x4, x5, x6] corresponds to [L, W, H, A, S, S'], and each variable has its boundaries (low and high).

[0159] Objective functions: Define two objective functions obj1(x) and obj2(x), which are respectively used to make predictions by calling the UTCI surrogate model and the AQIPM2.5 surrogate model trained in step S300 of Example 1. The objective is to minimize these two functions.

[0160] Constraints: Define the constraint function constr(x). Constraints include:

[0161] Variable boundary constraints (already handled by algorithm variable definition).

[0162] Planning constraints: For example: constr1 = 3.0 - FAR(x) # FAR >= 3.0 -> constr1 <= 0 constr2 = FAR(x) - 6.0 # FAR <= 6.0 -> constr2 <= 0 constr3 = 0.25 - BD(x) # Building Density <= 25% -> constr3 <= 0 constr4 = 24 - x5 # S >= 24m -> constr4 <= 0 constr5 = 24 - x6 # S' >= 24m -> constr5 <= 0 (where FAR(x) and BD(x) need to be calculated based on the variable x)

[0163] The final constraint vector g = [constr1, constr2, constr3, constr4, constr5, ...] requires all components to be less than or equal to 0.

[0164] Algorithm execution:

[0165] Initialize the NSGA-II algorithm and set the parameters:

[0166] Population size (pop_size): 2000

[0167] Maximum algebra (n_gen): 500

[0168] Crossover probability: 0.9

[0169] Crossover index (eta_crossover): using the single-point crossover method

[0170] The variability index (eta_mutatin): using uniform mutation, the variability rate is 0.01.

[0171] Convergence stability percentage: 1%

[0172] Planning constraint calculation formula:

[0173] Floor Area Ratio (R): (L*W*36*H / 3) / (L*6+S*5+18) / (W*6+S'*5+18)

[0174] Interval rate (J1): (L*6) / (L*6+S*5+18)

[0175] Interval rate (J2): (W*6) / (W*6+S'*5+18)

[0176] Coverage (F): (L*W*36) / (L*6+S*5+18) / (W*6+S'*5+18)

[0177] Run the algorithm. At each generation, the algorithm evaluates the objective function and constraints for all individuals in the population.

[0178] Evaluation process: For each individual (i.e., a set of morphological parameters x), the preloaded surrogate model (.predict() method) is called to quickly calculate the predicted values ​​of UTCI and AQIPM2.5 as the objective function value; at the same time, the degree of constraint violation is calculated.

[0179] Result extraction: After the algorithm terminates, extract the Pareto optimal solution set (opt.X, i.e., the design variables of all non-dominated solutions), the corresponding objective function values ​​(opt.F, i.e., UTCI and AQIPM2.5 values) and the constraint satisfaction status from the result object.

[0180] Effect verification framework:

[0181] Validation metrics: 1. Shape of the Pareto front (whether it is uniformly distributed and broad); 2. Convergence of the optimization process (the curve of the objective function value changing with the algebra); 3. Randomly select several Pareto solutions and verify them using high-precision CFD simulation, calculating the prediction error (R²) of the surrogate model. 2 ,RMSE).

[0182] Expected Results: The algorithm should converge stably and yield a well-distributed Pareto front. The prediction error of the surrogate model should be controlled within an acceptable range (e.g., RMSE < 5%). The final Pareto solution set should satisfy all planning constraints.

[0183] Example 4: Multi-objective optimization system for urban street block morphology

[0184] This embodiment also proposes a multi-objective optimization system for urban street morphology. A summary of the overall process of this system can be found in [link to relevant documentation]. Figure 1-1 For detailed calculation procedures, please refer to [link / reference]. Figure 1-2 .include:

[0185] Parameterization and Sampling Unit: This unit can be a graphical user interface (GUI) or a data interface, receiving user-input geometric parameters and their value ranges. It integrates a Latin hypercube sampling algorithm to generate sample points. The output is a list of sample parameters.

[0186] Two-way Coupled Simulation Unit: This unit is a module that encapsulates the functionality of CFD simulation software (such as ANSYS Fluent). This module can automatically read the sample list generated by the parameterization and sampling unit, and sequentially generate geometry, mesh, configure the physical model (including integrating user-defined functions (UDFs) to achieve the second-direction coupling process) for each sample, set boundary conditions, submit the calculation and monitor the solution process, and finally extract the result data (UTCI, AQIPM2.5). This module automates and batches the simulation process.

[0187] Proxy Model Construction Unit: This unit receives the dataset output from the bidirectional coupled simulation unit and calls machine learning libraries (such as libraries containing the LOESS algorithm) to train and validate the model. The output is a trained, serializable proxy model file (such as a .pkl or .joblib file).

[0188] Multi-objective Optimization Unit: This unit integrates a multi-objective genetic algorithm library (such as PyMoo) and loads the surrogate model file generated by the surrogate model construction unit as the objective function calculator. This module allows users to set constraints for the optimization problem (such as the aforementioned planning constraints), then execute the optimization search, and finally output the Pareto optimal solution set.

[0189] Results Analysis and Output Section: This section performs statistical analysis on the Pareto optimal solution set output by the multi-objective optimization section, extracts the parameter optimization range, and generates visualization charts (such as parallel coordinate plots and scatter plot matrices) and design guide reports. This section also includes a sensitivity analysis unit for calculating indicators such as the Spearman rank correlation coefficient to quantify the influence of each input parameter on the output objective.

[0190] How it works: Users initiate the optimization process through parameterization and sampling. The system automatically executes steps such as sample generation, CFD batch simulation, surrogate model training, optimization search, and result analysis in sequence, ultimately presenting the design guidelines to the user, achieving full automation of the process.

[0191] Implementation: This system can be deployed on a high-performance computing (HPC) cluster. Each component can be a script or program written in languages ​​such as Python or Java, exchanging data via files or message queues. A web interface can be provided at the front end for convenient user interaction.

[0192] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective optimization method for urban block form, characterized by, The method comprises the following steps: Step S100, constructing a parameterized model and performing an experimental design: comprising defining a set of geometric parameters for describing the morphology of urban blocks, the geometric parameters including building length, building width, building height, building orientation angle, building lateral spacing, and building longitudinal spacing, and generating a plurality of sample points in the value range of the geometric parameters using a sampling method; Step S200, performing a two-way coupled numerical simulation and generating a data set: comprising performing a computational fluid dynamics simulation coupled with the interaction of thermal environment and particulate matter concentration on each block morphology represented by the sample points generated in step S100, and outputting at least one thermal comfort evaluation index and one air quality evaluation index, thereby forming a data set containing input parameters and output indicators; Step S300, constructing and verifying a machine learning proxy model: comprising training a machine learning algorithm using the data set generated in step S200 to establish a mapping relationship from the geometric parameters to the thermal comfort evaluation index and the air quality evaluation index, thereby obtaining a proxy model capable of quickly predicting the evaluation index; Step S400, performing a multi-objective optimization search: comprising integrating the proxy model obtained in step S300 as a target function calculator into a multi-objective optimization algorithm, and automatically searching for a more optimal Pareto optimal solution set that simultaneously satisfies the thermal comfort evaluation index and the air quality evaluation index under the condition of meeting the preset planning constraints; Step S500, analyzing and outputting the optimization results: comprising analyzing the Pareto optimal solution set obtained in step S400, extracting the optimized value range of the geometric parameters, and generating a design guide.

2. The method of claim 1, wherein, The "computational fluid dynamics simulation coupled with the interaction of thermal environment and particulate matter concentration" in step S200 comprises: a first direction coupling process: comprising introducing the Boussinesq approximation in the flow field control equation to simulate the thermal buoyancy effect, and considering the thermal migration force effect in the particle force balance equation to simulate the influence of the thermal environment on the diffusion and transport process of the particulate matter; a second direction coupling process: comprising dynamically adjusting the solar radiation absorption coefficient and scattering coefficient of the atmosphere according to the real-time calculated local particulate matter concentration through a user-defined function, to simulate the feedback effect of the particulate matter concentration on the solar radiation transmission and the thermal environment.

3. The method of claim 2, wherein, The thermal comfort evaluation index is the universal thermal climate index UTCI, and the air quality evaluation index is the air quality sub-index AQIPM2.5 based on PM2.5 concentration.

4. The method of claim 1, wherein, The machine learning algorithm in step S300 is a non-parametric regression algorithm, specifically a locally weighted scatterplot smoothing method LOESS.

5. The method of claim 1, wherein, The multi-objective optimization algorithm in step S400 is a multi-objective genetic algorithm.

6. The method of claim 1, wherein, The sampling method in step S100 is Latin hypercube sampling.

7. The method of claim 1, wherein, The preset planning constraints in step S400 include one or more of the following: plot ratio, building density, building setback distance, and minimum building spacing.

8. A system for performing the multi-objective optimization of urban street patterns of any of claims 1-7, characterized in that, The method comprises the following steps: The parameterization and sampling unit is configured to receive user inputted geometric parameters and their ranges, and perform sampling to generate sample points; the two-way coupling simulation unit is configured to perform simulation calculation on the sample points and output a data set; the surrogate model construction unit is configured to receive the data set and train a machine learning model, and output a trained surrogate model; the multi-objective optimization unit is configured to integrate the surrogate model and an optimization algorithm, perform optimization search, and output a set of Pareto optimal solutions; and the result analysis and output unit is configured to receive the set of Pareto optimal solutions, perform analysis, and generate a visualized design guideline.

9. The system of claim 8, wherein, The two-way coupling simulation unit is a computational fluid dynamics software module integrated with a user-defined function (UDF) to realize the second direction coupling process.

10. The system of claim 8, wherein, The result analysis and output unit further comprises a sensitivity analysis unit configured to calculate a Spearman rank correlation coefficient to quantify the influence of each geometric parameter on the evaluation index.

Images