Community public space intelligent optimization design method and system oriented to resident sharing

Abstract

This invention proposes a resident-shared, oriented intelligent optimization design method and system for community public spaces. The intelligent optimization design method acquires real trajectory sets and environmental data from different groups, and uses a conditional generative adversarial network (GAN) to fill data sparsity gaps, generating spatiotemporal heatmaps of different group behaviors. Subsequently, a spatiotemporal graph convolutional network is used to fuse grid topology, heatmaps, and social association matrices to parse a demand priority map. A dual-channel deep learning architecture is then employed to couple cross-modal features, generating constrained physical design parameters through a bidirectional attention mechanism. This invention uses GAN to enhance data, improving the physical fidelity of the spatiotemporal heatmap, and integrates different group behaviors with spatiotemporal data to overcome priority biases in public space design caused by subjective weight allocation. Finally, cross-modal feature fusion integrates spatial topological gradients with the evolutionary patterns of group behavior, improving the demand response accuracy and dynamic adaptability of public space optimization.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and in particular to a method and system for intelligent optimization design of community public spaces oriented towards resident sharing. Background Technology

[0002] The current mainstream approach to public space renovation primarily employs static environmental behavior analysis. This approach involves recording residents' activity patterns through manual questionnaires and fixed-point observations, marking facility distribution using 2D CAD site maps, and dividing functional areas (such as rest areas and passageways) using simple clustering algorithms. Environmental analysis relies on historical meteorological data and simplified thermal comfort indices, with renovation parameters manually set based on standardized specifications. The optimization process uses a linear programming model to allocate facility quantities with cost minimization as the objective, outputting static layout drawings and a materials list. While this method can achieve basic functional configuration, data processing and design decisions have an open-loop structure. The manually acquired behavioral data has low spatiotemporal resolution and a limited sample size, making it difficult to capture the dynamic interaction patterns of different groups, leading to distorted heat maps. Secondly, demand analysis relies on subjective questionnaire weights and fails to establish a correlation model between spatiotemporal behavioral characteristics and physical space, causing priority maps to ignore the shared needs of different groups. Finally, parameter generation lacks cross-modal coupling mechanisms, and spatial topological gradients are not included in accessibility calculations, resulting in a misalignment between facility layout and actual behavioral flow, making it difficult to meet the needs of most residents in the renovation of community public spaces. Summary of the Invention

[0003] In view of the above, this invention proposes a resident-shared-oriented intelligent optimization design method and system for community public spaces, which is used to solve at least one of the problems in the above-mentioned technologies.

[0004] This invention is achieved through the following technical solution. On the one hand, this invention provides a community public space intelligent optimization design method oriented towards resident sharing, which includes: a data processing stage, a demand feature solving stage, a space-behavior coupling stage, and a design parameter decoding stage.

[0005] The following steps are performed during the data processing phase:

[0006] S11: Obtain a set of real trajectories of various user groups within a preset continuous time window, and perform conditional adversarial fitting based on the set of real trajectories to obtain an enhanced trajectory set;

[0007] S12: Obtain environmental data of the space to be optimized to generate raw point cloud data; and generate a grid space and a corresponding spatial topology matrix based on the raw point cloud data and the boundary of the space to be optimized. The grid space includes multiple grid cells, and each grid cell includes center coordinates and grid size.

[0008] S13: Obtain environmental element data within a preset continuous time window and construct an environmental feature tensor;

[0009] The following steps are performed during the demand characteristics solving phase:

[0010] S21A: Calculate the social association matrix based on the spatiotemporal distribution characteristics of the social situations of different groups in the enhanced trajectory set in the grid space;

[0011] S21B: Based on the center coordinates of the enhanced trajectory set and the grid cells, the thermal value of each grid cell is generated using the kernel density estimation method, thus obtaining the spatiotemporal thermal map tensor of the enhanced trajectory set;

[0012] S22: Construct the spatiotemporal graph nodes and corresponding spatiotemporal feature vectors for each grid cell based on the spatial topology matrix and the spatiotemporal heatmap tensor, thereby obtaining the spatiotemporal graph structure and the temporal causality matrix;

[0013] S23: Perform multi-level graph convolution and mapping on the spatiotemporal graph structure based on the social association matrix, spatial topology matrix and temporal causality matrix to obtain the demand priority graph.

[0014] The following steps are performed during the space-behavior coupling phase:

[0015] S31A: Calculate the temporal characteristics of each group based on the demand priority map and the spatiotemporal heat map tensor to obtain the group behavior feature vector;

[0016] S31B: After enhancing the environmental feature tensor according to the demand priority map, perform spatial convolution operation to obtain environmental spatial topological features;

[0017] S32: Perform cross-modal feature fusion on the group behavior feature vector and the environmental spatial topological features to obtain fused features;

[0018] The following steps are performed during the design parameter decoding phase:

[0019] S40: The fused features are decoded using a fully connected network to obtain optimized design parameters.

[0020] Further, step S21A includes:

[0021] S21A.1: Based on the enhanced trajectory set and the grid space, count the number of times different groups appear in different time windows of different grids to obtain multiple spatiotemporal distribution frequency vectors of different groups;

[0022] S21A.2: Construct a social association matrix based on the spatiotemporal distribution frequency vectors between different groups.

[0023] Furthermore, the formula for constructing the social association matrix is ​​as follows:

[0024]

[0025]

[0026] in, For group indexing, Representing a group In the grid time Frequency of occurrence Representing a group In the grid time Frequency of occurrence As the normalization factor, For indicator functions, The total number of time windows. For grid space.

[0027] Further, step S23 includes:

[0028] S231: The spatiotemporal graph structure is processed by multiple convolutional layers of spatiotemporal convolutional kernels constructed from the social association matrix, spatial topology matrix and temporal causality matrix to obtain group features;

[0029] S232: Perform global flat pooling on the group features to obtain the group representation vector;

[0030] S233: Perform fully connected mapping and normalization on the group representation vector to obtain the priority map.

[0031] Further, step S31A includes:

[0032] S31A.1: Perform group weighting processing on the heat map tensor according to the demand priority map to obtain weighted heat maps of different groups in different time windows of different grid cells;

[0033] S31A.2: Spatially flatten the weighted heatmap of each group to obtain a one-dimensional time series for each group;

[0034] S31A.3: Sequentially extract time series features from the one-dimensional time series of each group and perform time series max pooling to obtain a group behavior feature vector that includes the behavior feature vector of each group.

[0035] Further, step S31B includes:

[0036] S31B.1: Extract the gradient field of the geometric model corresponding to the mesh space;

[0037] S31B.2: The environmental feature tensor and gradient field are modified and enhanced according to the demand priority map to obtain the enhanced environmental feature tensor for each group;

[0038] S31B.3: Sequentially perform convolution processing and global spatiotemporal average pooling on the enhanced environmental feature tensor of each group to obtain environmental spatial topology features including the spatial topology features of each group.

[0039] Furthermore, the resident-shared-oriented intelligent optimization design method for community public spaces also includes:

[0040] During the iterative optimization phase, the following steps are performed:

[0041] S51; Calculate the group matching degree of the high-frequency activity area of ​​each group based on the optimized design parameters, spatiotemporal heat map tensor, and demand priority map, and obtain the global matching degree of the space to be optimized by weighted summation;

[0042] S52: Iteratively optimize the design parameters based on the global matching degree until the evaluation function converges, and output the final design parameters.

[0043] Furthermore, the formula for calculating the global matching degree is as follows:

[0044]

[0045]

[0046] Among them, the For the group The set of high-frequency activity regions, that is, the set of grids whose thermal values ​​exceed a preset threshold; For the group Demand weight, The total number of group types The total number of time windows. Let the vitality intensity function be... This is a function for environmental comfort.

[0047] On the other hand, the present invention also provides a community public space intelligent optimization design system oriented towards resident sharing, which includes: a data processing module, a demand feature solving module, a space-behavior coupling module, and a design parameter decoding module;

[0048] The data processing module includes: a user trajectory processing unit, which is used to acquire the real trajectory set of various different groups of users within a preset continuous time window, and perform conditional adversarial generation fitting based on the real trajectory set to obtain an enhanced trajectory set;

[0049] Spatial data processing unit: used to acquire environmental data of the space to be optimized to generate raw point cloud data; and based on the raw point cloud data and the boundary of the space to be optimized, to generate a grid space and a corresponding spatial topology matrix, wherein the grid space includes multiple grid cells, and each grid cell includes center coordinates and grid size;

[0050] Environmental data processing unit: used to acquire environmental element data within a preset continuous time window and construct an environmental feature tensor;

[0051] The demand feature solving module includes: a social association matrix solving unit, which is used to calculate the social association matrix based on the spatiotemporal distribution characteristics of the social situations of different groups in the enhanced trajectory set in the grid space;

[0052] Spatiotemporal heatmap generation unit: used to generate the thermal value of each grid cell based on the center coordinates of the enhanced trajectory set and the grid cell using the kernel density estimation method, thus obtaining the spatiotemporal heatmap tensor of the enhanced trajectory set;

[0053] Spatiotemporal graph construction unit: used to construct the spatiotemporal graph nodes and corresponding spatiotemporal feature vectors of each grid unit based on the spatial topology matrix and the spatiotemporal heatmap tensor, so as to obtain the spatiotemporal graph structure and the temporal causality matrix;

[0054] Demand priority graph generation unit: used to perform multi-level graph convolution and mapping on the spatiotemporal graph structure based on the social association matrix, spatial topology matrix and temporal causality matrix to obtain the demand priority graph;

[0055] The spatial-behavior coupling module includes: a temporal feature extraction unit, used to calculate the temporal features of each group based on the demand priority map and the spatiotemporal heat map tensor, and obtain the group behavior feature vector;

[0056] Spatial feature extraction unit: used to enhance the environmental feature tensor according to the demand priority map and then perform spatial convolution operation to obtain environmental spatial topological features;

[0057] Modal fusion feature extraction unit: used to perform cross-modal feature fusion of the group behavior feature vector and environmental spatial topological features to obtain fused features;

[0058] Design parameter decoding module: used to decode the parameters of the fused features through a fully connected network to obtain optimized design parameters.

[0059] Furthermore, it also includes: an iterative optimization module, which includes:

[0060] A global matching degree calculation unit is used to calculate the group matching degree of the high-frequency activity area of ​​each group based on the optimization design parameters, spatiotemporal heat map tensor, and demand priority map, and obtain the global matching degree of the space to be optimized by weighted summation.

[0061] Matching degree iterative optimization unit: used to iteratively optimize the optimization design parameters based on the global matching degree until the evaluation function converges and outputs the final design parameters.

[0062] This invention presents a resident-shared, community-oriented intelligent optimization design method for public spaces. It integrates a sensor network to collect resident behavior trajectories and environmental data, and utilizes conditional generative adversarial networks to address the sparsity of behavioral data in older communities, generating a high-precision spatiotemporal heatmap. Based on a spatiotemporal graph convolutional network, it fuses grid topology, heatmaps, and social association matrices related to the social behaviors of different groups to deduce a normalized demand priority map. A dual-channel deep learning architecture couples cross-modal features, generating constrained physical design parameters through a bidirectional attention mechanism. Finally, a quantification model for vitality intensity and comfort is established to calculate the entropy-weighted global matching degree, driving gradient optimization iteration and outputting executable instructions and environmental configurations. After deployment, real-time data from the Internet of Things provides closed-loop verification. When the measured deviation exceeds a threshold, the optimization process is automatically restarted, achieving continuous adaptive optimization of the space transformation.

[0063] Therefore, compared with the prior art, the present invention has the following advantages:

[0064] 1. Conditional generative adversarial networks are used to fill the data sparsity blind spots, significantly improving the physical fidelity and sample coverage of spatiotemporal heatmaps.

[0065] 2. Introduce spatiotemporal graph convolutional networks to fuse gridded topology and social association matrices, so as to realize interpretable quantification of the interaction needs of different resident groups in the community, and overcome the priority bias in public space design caused by subjective weight allocation.

[0066] 3. By coupling cross-modal features with a bidirectional attention mechanism, the features are integrated with the spatial topological gradient and the evolution law of group behavior, ensuring that the generated optimized design parameters can meet the spatial distribution requirements and practical scenario requirements of most groups.

[0067] 4. The closed-loop verification mechanism is based on dynamic optimization and iteration of monitoring feedback to ensure that the transformation plan significantly improves the actual effect in the optimization application of public spaces, and automatically restarts the optimization process when the deviation exceeds the threshold, so as to continuously strengthen the vitality and sustainability of public spaces.

[0068] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0069] Figure 1This is a structural block diagram of the resident-shared-oriented intelligent optimization design system for community public spaces according to the present invention;

[0070] Figure 2 For execution Figure 1 The flowchart illustrates a resident-shared-oriented intelligent optimization design method for community public spaces in the system shown.

[0071] Figure 3 for Figure 1 The diagram shown is a structural block diagram of the resident-shared community public space intelligent optimization design system after adding an iterative optimization module.

[0072] Figure 4 To execute Figure 3 The execution flowchart of the iterative optimization module of the system shown is illustrated. Detailed Implementation

[0073] To address the shortcomings of existing public space optimization methods that rely on static questionnaires and manual analysis, this invention proposes a resident-shared-oriented intelligent optimization design method and system for community public spaces. This method and system can provide scientific and quantitative intelligent decision support for public space renovation, guided by the shared needs of user groups.

[0074] This invention uses the community renovation process as an example for illustration; please refer to [link / reference]. Figure 1 The present invention relates to a resident-shared, community-oriented intelligent optimization design system for public spaces, which is connected to an integrated sensor network. Based on user trajectory data, spatial data of the space to be optimized, and environmental data collected by the integrated sensor network, the system performs analysis and decision-making. Specifically, the intelligent optimization design system includes: a data processing module, a demand feature solving module, a space-behavior coupling module, and a design parameter decoding module. The data processing module first preprocesses the acquired data and uses a conditional generative adversarial network to address the sparsity problem of behavioral data in older communities. The demand feature solving module generates a high-precision spatiotemporal heatmap from the data output of the data processing module. Then, based on a spatiotemporal graph convolutional network, it fuses the grid topology, heatmap, and social association matrix to parse a normalized demand priority map. The space-behavior coupling module employs a dual-channel deep learning architecture to couple cross-modal features. Finally, the design parameter decoding module decodes the fused features and outputs constrained physical design parameters for the design optimization of public spaces.

[0075] Please see Figure 2 The working process of each component of the community public space intelligent optimization design system based on resident sharing includes:

[0076] In the data processing phase, each unit of the data processing module performs the following steps:

[0077] The user trajectory processing unit 11 is used to perform step S11: obtain a set of real trajectories of various groups of users within a preset continuous time window, and perform conditional adversarial generation fitting based on the set of real trajectories to obtain an enhanced trajectory set.

[0078] The system tracks the daily movement trajectories of different residents within the community in real time using GPS positioning. The trajectory set is defined as follows: ,in This indicates the total number of individual residents surveyed, and a single trajectory. Includes time series data, where It's a timestamp. Indicates the index of the number of trajectory points. The number of trajectory points for each individual. Represents an individual At any moment The spatial coordinates are obtained using local coordinate system values. However, since the number of residents participating in the survey differs significantly from the number of people at different times, behavioral data augmentation is used to compensate for the sparsity of the original trajectories. Specifically, a Conditional Generative Adversarial Network (cGAN) is employed for high-density mapping. The cGAN consists of a generator... and discriminator Composition, generator input noise vector (in (representing standard Gaussian distribution) and conditional data (i.e., the original set of trajectories), output the generated trajectory. The generator employs a U-Net architecture, where the encoder extracts trajectory features and the decoder generates a new sequence of trajectory points. The discriminator, based on a convolutional network structure, distinguishes between real and generated input trajectories. The loss function is designed to ensure that the distribution of the generated trajectories approximates the real distribution while maintaining spatial consistency; the adversarial loss function is defined as:

[0079]

[0080] in, Represents the expectation operator. It is the actual data distribution. This is the noise distribution.

[0081] The reconstruction loss function is defined as:

[0082]

[0083] in It is the L1 norm, which constrains the generation position. and measured position The deviation.

[0084] Total loss weighted combination is ,in, and Using typical weight values, priority is given to ensuring reconstruction accuracy to avoid spatial distortion. This yields the enhanced trajectory set. .

[0085] The spatial data processing unit 12 is used to perform step S12: acquire environmental data of the space to be optimized to generate raw point cloud data; and generate a grid space and a corresponding spatial topology matrix based on the raw point cloud data and the boundary of the space to be optimized. The grid space includes multiple grid cells, and each grid cell includes center coordinates and grid size.

[0086] Point cloud geometry is obtained by scanning raw point cloud data using LiDAR. .

[0087]

[0088] in, For point index, The total number of point clouds describes the three-dimensional geometric features of the space.

[0089] The acquired point cloud data is then preprocessed, including denoising and registration operations, to generate an accurate 3D spatial surface model. The model is then projected onto a two-dimensional ground plane and discretized into a mesh structure to accommodate behavioral analysis. Specifically, the space is divided into... There are 1 grid cells, each denoted as _____. ,in and These represent the row and column indices of the grid. The formula for calculating the center coordinates of a grid cell is as follows:

[0090]

[0091] in, and This represents the total length and width of the space, with a typical mesh resolution of 0.5m × 0.5m to balance accuracy and computational efficiency. Topology connectivity is constructed by defining a spatial topology matrix. Implementation. Matrix elements satisfy the condition: when the mesh element... and Spatially adjacent (i.e.) and When there are 8 adjacent relationships, Otherwise, it is 0. This matrix is ​​subsequently used for spatial correlation modeling in spatiotemporal graph convolutional networks.

[0092] The environmental data processing unit 13 is used to perform step S13: acquire environmental element data of a preset continuous time window and construct an environmental feature tensor;

[0093] Microclimate parameters are obtained by collecting time-series data using temperature and humidity sensors and anemometers. These represent temperature, humidity, and wind speed, respectively; environmental factors such as light intensity, noise, and vegetation cover data are acquired through dedicated sensors. The environmental feature tensor constructed from this foundational data is used to ensure the coupling of behavior with spatial input.

[0094] In the requirement feature solving phase, the execution flow of each component of the requirement feature solving module includes:

[0095] The social association matrix solving unit 21A is used to perform step S21A: calculate the social association matrix based on the spatiotemporal distribution characteristics of the social situations of different groups in the enhanced trajectory set in the grid space.

[0096] Step S21A includes: S21A.1: Based on the enhanced trajectory set and the grid space, count the number of times different groups appear in different time windows of different grids to obtain the spatiotemporal distribution frequency vectors of multiple groups.

[0097] S21A.2: Construct a social association matrix based on the spatiotemporal distribution frequency vectors between different groups.

[0098] To quantify social connections between groups, define A group of residents. In this application, ( , , ), and construct the social connection matrix using the following formula :

[0099]

[0100]

[0101] in, For group indexing, Representing a group In the grid time Frequency of occurrence Representing a group In the grid time Frequency of occurrence As the normalization factor, For indicator functions, The total number of time windows. This represents a grid space. This matrix reveals the co-occurrence patterns of the population, which will be subsequently mapped to a node-level graph structure.

[0102] The spatiotemporal heatmap generation unit 21B is used to perform step S21B: generate the thermal value of each grid cell based on the center coordinates of the enhanced trajectory set and the grid cell using the kernel density estimation method, and obtain the spatiotemporal heatmap tensor of the enhanced trajectory set.

[0103] High-density behavioral heatmap generation based on enhanced trajectories The spatiotemporal thermodynamic values ​​are calculated to quantify the density of resident behavior. Specifically, the kernel density estimation method is used for the calculation of grid cells. within a specific time window The internal thermal value is defined as:

[0104]

[0105] in, It is the heat value. For grid cells, For the target time point, and These are individual resident indexes and trajectory point indexes, It is individual weight. Represents Euclidean distance. These are the coordinates of the grid center. Kernel function. Using Gaussian kernel ,in It is the normalized distance; bandwidth parameter To control the attenuation of spatial influence, a typical value is 1.5 times the grid side length, ensuring a smooth distribution of thermal values ​​in local areas. The time dimension is discretized into a full day. The final output is a spatiotemporal heatmap tensor for each time period. The complete encoding behavior is distributed spatiotemporally.

[0106] The spatiotemporal graph construction unit 22 is used to perform step S22: construct the spatiotemporal graph nodes and corresponding spatiotemporal feature vectors of each grid unit according to the spatial topology matrix and the spatiotemporal heatmap tensor, so as to obtain the spatiotemporal graph structure and the time causality matrix.

[0107] High-density behavioral heatmap generated in the preceding sequence With spatial topological matrix By combining these elements, a spatiotemporal graph structure is constructed as the computational foundation. Spatiotemporal graph nodes are defined as follows: Each node Corresponding grid cell At time step ( The entity's feature vector integrates multidimensional information about its behavior and environment:

[0108]

[0109] in, Statistical time window Entering the grid The number of residents, Calculate the average length of stay for individuals within the same time period. The calculation is performed by counting the frequency of instances where the minimum Euclidean distance between individuals is less than the interaction threshold. This is the thermal value output by the previous module. This is a microclimate vector.

[0110] The edge structure of the spatiotemporal graph contains dual associations. Spatial adjacency edges inherit the mesh topology matrix. At each time step Inside, if the grid cell and satisfy and (8 adjacency relationships) then establish an undirected edge and assign a basic weight. .

[0111] For time-related characteristics, the weights of the temporal causal edges are calculated. This is done in adjacent time steps. and Between, for the same grid cell The formula for calculating the weight of a directed edge with temporal causality is as follows:

[0112]

[0113] in As the attenuation factor, t This represents the time window number. The temporal causal edge weight characterizes the continuous evolution of behavioral patterns. When two nodes are in the same grid cell and in adjacent time steps, the temporal causal edge weight is... Otherwise By traversing the temporal relationships of all nodes in the spatiotemporal graph, the weights of all temporal causal edges are obtained, and a temporal causal matrix is ​​generated. .

[0114] The demand priority graph generation unit 23 is used to perform step S23: perform multi-level graph convolution and mapping on the spatiotemporal graph structure according to the social association matrix, spatial topology matrix and temporal causality matrix to obtain the demand priority graph.

[0115] Step S23 includes: S231: The spatiotemporal graph structure is processed by a multi-layer convolutional process using a spatiotemporal convolutional kernel constructed from the social association matrix, spatial topology matrix and temporal causality matrix to obtain group features.

[0116] Spatiotemporal graph convolution operations extract desired features by hierarchically aggregating neighborhood information. The layer feature update formula is:

[0117]

[0118] in, For the first Layer node feature matrix, convolution kernel set These correspond to three types of adjacency matrices, including spatial adjacency matrices. ( (For degree matrix, achieve symmetric normalization) Time causality matrix Preserving directionality, social connection matrix By Give to all who belong to the group and Node pairs, parameter matrix Trainable The ReLU activation function is used. After multiple convolutional layers, the group features are obtained. .

[0119] S232: Perform global flat pooling on the group features to obtain the group representation vector;

[0120] group characteristics The vectors are aggregated into a group-level representation vector using global average pooling:

[0121]

[0122]

[0123] S233: Perform fully connected mapping and normalization on the group representation vector to obtain the priority map.

[0124] Mapping to requirement priority scores via a fully connected layer:

[0125]

[0126] in, 、 、 These are trainable parameters. The obtained priority scores are then normalized using Softmax to satisfy the following condition:

[0127]

[0128]

[0129] Finally, a demand priority map was obtained. This heatmap quantitatively characterizes the weight allocation of each group in the optimization of public space. Its output strictly adheres to spatiotemporal topological constraints and social relevance, providing interpretable decision input for the space-behavior coupled optimization module. In this step, the heatmap... Provides behavioral density and spatial grid topology. Define spatial adjacency and temporal causal edge weights. To ensure the continuity of behavioral evolution, the social matrix Softmax normalization forces the elimination of population cardinal bias by normalizing frequency. Avoid optimizing for divergent goals.

[0130] In the spatial-behavior coupling phase, the execution flow of each component of the spatial-behavior coupling module includes:

[0131] The temporal feature extraction unit 31A is used to perform step S31A: calculate the temporal features of each group based on the demand priority map and the spatiotemporal heat map tensor to obtain the group behavior feature vector.

[0132] Step S31A includes: S31A.1: Performing group weighting processing on the heat map tensor according to the demand priority map to obtain weighted heat maps of different groups in different time windows of different grid cells;

[0133] Based on demand priority map The spatiotemporal heatmap is subjected to group weighting. The weighted heatmap value of each point in the weighted heatmap of any group is calculated using the following formula:

[0134]

[0135]

[0136] Among them, the For the group spatiotemporal coverage Here, t represents the spatial grid coordinates, and t represents the time window number. For the group Priority map.

[0137] S31A.2: Spatially flatten the weighted heatmap of each group to obtain a one-dimensional time series for each group.

[0138] Flatten the 3D weighted heatmap using the following formula:

[0139]

[0140] in This represents the overall spatial behavior density, where T is the total number of time windows.

[0141] S31A.3: Sequentially extract time series features from the one-dimensional time series of each group and perform time series max pooling to obtain a group behavior feature vector that includes the behavior feature vector of each group.

[0142] Temporal features are extracted using an LSTM encoder, and the cell state update mechanism is as follows:

[0143]

[0144]

[0145] in, For input gate, For the Gate of Oblivion For output gate, This involves element-wise multiplication. The final group behavior feature vector is obtained through temporal max pooling. .

[0146] The spatial feature extraction unit 31B is used to perform step S31B: after enhancing the environmental feature tensor according to the demand priority map, perform spatial convolution operation to obtain environmental spatial topological features.

[0147] Specifically, step S31B includes:

[0148] S31B.1: Extract the gradient field of the geometric model corresponding to the grid space.

[0149] S31B.2: The environmental feature tensor and gradient field are modified and enhanced according to the demand priority map to obtain the enhanced environmental feature tensor for each group.

[0150] The demand weights are combined with the spatial geometric gradient. The environmental feature tensor is enhanced as follows:

[0151]

[0152] in ， Geometric model corresponding to the grid space The gradient field.

[0153] S31B.3: Sequentially perform convolution processing and global spatiotemporal average pooling on the enhanced environmental feature tensor of each group to obtain environmental spatial topology features including the spatial topology features of each group.

[0154] Feature extraction using a three-layer 3D-CNN encoder:

[0155]

[0156]

[0157] in, Represents a three-dimensional convolution operation. The kernel is used for convolution. Global spatiotemporal average pooling is then applied to obtain the environmental spatial topological features.

[0158] The modal fusion feature extraction unit 32 is used to perform step S32: cross-modal feature fusion of the group behavior feature vector and the environmental spatial topology features to obtain fused features.

[0159] Cross-modal feature fusion employs a bidirectional cross-channel attention mechanism. The behavior-to-space attention module uses behavioral features as queries and spatial features as keys.

[0160]

[0161] Space → Behavioral Attention operates in reverse:

[0162]

[0163] in, ， ， This is the projection matrix. The final fused feature is... ( (Indicates splicing).

[0164] The design parameter decoding module 40 is used to perform step S40: decode the fused features through a fully connected network to obtain optimized design parameters.

[0165] Parametric design decoding is implemented through a fully connected network:

[0166]

[0167] in, This includes physical design parameters such as rest facility density, main passage width, shading coverage, and vegetation bandwidth. To meet engineering constraints, a differentiable projection operator is embedded in the output layer, including boundary constraints. and normalization constraints ( (For mutually exclusive parameter groups).

[0168] The closed-loop optimization mechanism is implemented using a differentiable surrogate evaluator. This applies to the weighted aggregation scheme. Calculate the overall score:

[0169]

[0170] Among them, the spatial vitality function Predicting future usage density using an LSTM-based behavioral simulator, thermal comfort function The PMV index is calculated by integrating microclimate models. Iterative optimization is performed using gradient ascent.

[0171]

[0172] The final output is standardized design instructions, geometric parameters are converted into BIM entities in IFC format, and material parameters are generated into XML configuration files (defining vegetation reflectivity and the thermal conductivity of shading materials). Requirements Map The weighting mechanism ensures fairness within the group, heatmap The time step and grid size are consistent with the spatiotemporal graph convolution module. Spatial gradient. The mesh topology, derived from the point cloud geometry processing module, ensures geometric continuity.

[0173] Please see Figure 3 In a preferred embodiment, the resident-shared community public space intelligent optimization design system further includes an iterative optimization module, which iteratively optimizes the optimization design parameters output by the design parameter decoding module. The execution flow of each component of the iterative optimization module is as follows: Figure 4 As shown, the details are as follows:

[0174] The global matching degree calculation unit 51 executes step S51; it is used to calculate the group matching degree of the high-frequency activity area of ​​each group according to the optimization design parameters, spatiotemporal heat map tensor, and demand priority map, and obtain the global matching degree of the space to be optimized by weighted summation.

[0175] This step involves generating parameterized design vectors using preorder space-behavior coupling optimization. Using the demand priority graph as the core input. Spatiotemporal behavior heatmap and environmental feature tensors This involves establishing a mapping relationship between design parameters and spatial performance. Its core lies in quantifying the matching degree of design schemes to the needs of different groups and generating executable engineering instructions.

[0176] First, perform matching metric optimization and design parameters. Spatial performance is evaluated using two key metrics: vitality intensity and environmental comfort. Vitality intensity... Comprehensive behavioral density facility density （ The number of rest facilities per unit area, and accessibility of pathways (path width) With spatial topological gradient The product of (the product of the two elements) is standardized to a value in the interval [0,1] by the Sigmoid function. Environmental comfort Integrated thermodynamic indicators PMV and shading coverage With solar radiation The interaction effect and vegetation bandwidth noise The attenuation effect.

[0177] Group-level matching degree Focus Group High-frequency activity area (defined as the heat value continuously exceeding a threshold) (grid set), by demand weight The weighted performance of vitality and comfort is used to calculate the overall matching degree, as follows:

[0178]

[0179]

[0180] Among them, the For the group The set of high-frequency activity regions, that is, the set of grids whose thermal values ​​exceed a preset threshold; For the group Demand weight, The total number of group types The total number of time windows. Let the vitality intensity function be... The environmental comfort function is used. The global matching degree adopts an entropy-weighted form, and the sensitivity to the needs of small groups is enhanced through logarithmic transformation, avoiding design bias towards the mainstream group.

[0181] The matching degree iterative optimization unit 52 executes step S52: iteratively optimizing the optimization design parameters based on the global matching degree until the evaluation function converges and outputs the final design parameters.

[0182] Finally, based on the gradient ascent strategy and matching degree feedback, the parameters are continuously adjusted, and the update formula is as follows:

[0183]

[0184] in, For learning rate, To suppress overfitting, an L2 regularization term is used. The optimization process must enforce physical boundary constraints and geometric conflict constraints. Physical boundary constraints are ensured through projection operators that physical parameters remain within the feasible region (e.g., the lower limit of channel width). Meet the standards, maximum facility density (To prevent overcrowding). Geometric conflict constraints are based on Voronoi diagrams and repulsion field algorithms to iteratively adjust facility locations, ensuring minimum spacing.

[0185] Corrected parameters Convert geometric instructions to IFC format and environment instructions to XML format. The geometric instructions will include the site elevation gradient. Mapped to the `IfcSite` entity in Building Information Modeling (BIM), aisle width The topology properties of the `IfcPath` object are written, and the coordinates of the rest facilities are generated using the Poisson disk sampling algorithm and output as an `IfcBuildingElementProxy` instance. The environment directive defines the reflectivity of the shading structure. And based on vegetation bandwidth Plant communities are configured in a hierarchical manner.

[0186] After deployment, real-world data is collected via IoT sensors. Calculate the prediction bias of the matching degree .like If this is not the case, closed-loop optimization will be initiated. The optimization process first integrates the measured heatmap. Secondly, spatiotemporal graph convolutional networks are used to correct the group weights. and with Restart the coupling optimization process with initial values. The iterative process is determined by the evaluation function. Driven by the Adam optimizer, the solution is solved until convergence.

[0187] In summary, the resident-shared-oriented intelligent optimization design method for community public spaces of this invention collects resident behavior trajectories and environmental data through an integrated sensor network. It utilizes conditional generative adversarial networks to address the sparsity of behavioral data in older communities, generating a high-precision spatiotemporal heatmap. Based on a spatiotemporal graph convolutional network, it fuses grid topology, heatmaps, and social association matrices related to the social behaviors of different groups to parse a normalized demand priority map. A dual-channel deep learning architecture couples cross-modal features, generating constrained physical design parameters through a bidirectional attention mechanism. Finally, a quantification model for vitality intensity and comfort is established to calculate the entropy-weighted global matching degree, driving gradient optimization iteration and outputting executable instructions and environmental configurations. After deployment, real-time data from the Internet of Things provides closed-loop verification. When the measured deviation exceeds a threshold, the optimization process is automatically restarted, achieving continuous adaptive optimization of space transformation.

[0188] Therefore, the present invention has the following advantages compared to the prior art:

[0189] 1. Conditional generative adversarial networks are used to fill the data sparsity blind spots, significantly improving the physical fidelity and sample coverage of spatiotemporal heatmaps.

[0190] 2. Introduce spatiotemporal graph convolutional networks to fuse gridded topology and social association matrices, so as to realize interpretable quantification of the interaction needs of different resident groups and overcome the priority bias in public space design caused by subjective weight allocation.

[0191] 3. By coupling cross-modal features with a bidirectional attention mechanism, the features are integrated with the spatial topological gradient and the evolution law of group behavior, ensuring that the generated optimized design parameters can meet the spatial distribution requirements and practical scenario requirements of most groups.

[0192] 4. The closed-loop verification mechanism is based on dynamic optimization and iteration of monitoring feedback to ensure that the transformation plan significantly improves the actual effect in the optimization application of public spaces, and automatically restarts the optimization process when the deviation exceeds the threshold, so as to continuously strengthen the vitality and sustainability of public spaces.

[0193] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.

Claims

1. A resident-shared, oriented intelligent optimization design method for community public spaces, characterized in that, include: The stages include data processing, requirement feature solving, space-behavior coupling, and design parameter decoding. The following steps are performed during the data processing phase: S11: Obtain a set of real trajectories of various user groups within a preset continuous time window, and perform conditional adversarial fitting based on the set of real trajectories to obtain an enhanced trajectory set; S12: Obtain environmental data of the space to be optimized and generate raw point cloud data; Based on the original point cloud data and the boundary of the space to be optimized, a grid space and a corresponding spatial topology matrix are generated. The grid space includes multiple grid cells, and each grid cell includes center coordinates and grid size. S13: Obtain environmental element data within a preset continuous time window and construct an environmental feature tensor; The following steps are performed during the demand characteristics solving phase: S21A.1: Based on the enhanced trajectory set and the grid space, count the number of times different groups appear in different time windows of different grids to obtain multiple spatiotemporal distribution frequency vectors of different groups; S21A.2: Construct a social association matrix based on the spatiotemporal distribution frequency vectors between pairs of different groups; S21B: Based on the center coordinates of the enhanced trajectory set and the grid cells, the thermal value of each grid cell is generated using the kernel density estimation method, thus obtaining the spatiotemporal thermal map tensor of the enhanced trajectory set; S22: Construct the spatiotemporal graph nodes and corresponding spatiotemporal feature vectors for each grid cell based on the spatial topology matrix and the spatiotemporal heatmap tensor, thereby obtaining the spatiotemporal graph structure and the temporal causality matrix; S23: Perform multi-level graph convolution and mapping on the spatiotemporal graph structure based on the social association matrix, spatial topology matrix and temporal causality matrix to obtain the demand priority graph. The following steps are performed during the space-behavior coupling phase: S31A: Calculate the temporal characteristics of each group based on the demand priority map and the spatiotemporal heat map tensor to obtain the group behavior feature vector; S31B: After enhancing the environmental feature tensor according to the demand priority map, perform spatial convolution operation to obtain environmental spatial topological features; S32: Perform cross-modal feature fusion on the group behavior feature vector and the environmental spatial topological features to obtain fused features; The following steps are performed during the design parameter decoding phase: S40: The fused features are decoded using a fully connected network to obtain optimized design parameters.

2. The intelligent optimization design method for resident-shared community public spaces according to claim 1, characterized in that, The formula for constructing the social association matrix is ​​as follows: in, For group indexing, Representing a group In the grid time Frequency of occurrence Representing a group In the grid time Frequency of occurrence As the normalization factor, For indicator functions, The total number of time windows. For grid space.

3. The intelligent optimization design method for resident-shared community public spaces according to claim 2, characterized in that, Step S23 includes: S231: The spatiotemporal graph structure is processed by multiple convolutional layers of spatiotemporal convolutional kernels constructed from the social association matrix, spatial topology matrix and temporal causality matrix to obtain group features; S232: Perform global average pooling on the group features to obtain the group representation vector; S233: Perform fully connected mapping and normalization on the group representation vector to obtain the priority map.

4. The intelligent optimization design method for resident-shared community public spaces according to claim 3, characterized in that, Step S31A includes: S31A.1: Perform group weighting processing on the heat map tensor according to the demand priority map to obtain weighted heat maps of different groups in different time windows of different grid cells; S31A.2: Spatially flatten the weighted heatmap of each group to obtain a one-dimensional time series for each group; S31A.3: Sequentially extract time series features from the one-dimensional time series of each group and perform time series max pooling to obtain a group behavior feature vector that includes the behavior feature vector of each group.

5. The resident-shared-oriented intelligent optimization design method for community public spaces according to claim 4, characterized in that, Step S31B includes: S31B.1: Extract the gradient field of the geometric model corresponding to the mesh space; S31B.2: The environmental feature tensor and gradient field are modified and enhanced according to the demand priority map to obtain the enhanced environmental feature tensor for each group; S31B.3: Sequentially perform convolution processing and global spatiotemporal average pooling on the enhanced environmental feature tensor of each group to obtain environmental spatial topology features including the spatial topology features of each group.

6. The resident-shared-oriented intelligent optimization design method for community public spaces according to any one of claims 1-5, characterized in that, Also includes: During the iterative optimization phase, the following steps are performed: S51: Calculate the group matching degree of the high-frequency activity area of ​​each group based on the optimized design parameters, spatiotemporal heat map tensor, and demand priority map, and obtain the global matching degree of the space to be optimized by weighted summation; S52: Iteratively optimize the design parameters based on the global matching degree until the evaluation function converges, and output the final design parameters.

7. The intelligent optimization design method for resident-shared community public spaces according to claim 6, characterized in that, The formula for calculating the global matching degree is as follows: Among them, the For the group The set of high-frequency activity areas, that is, the set of grids whose thermal values ​​exceed a preset threshold; For the group Demand weight, The total number of group types The total number of time windows. Let the vitality intensity function be... This is a function for environmental comfort.

8. A resident-shared, intelligent optimization design system for community public spaces, characterized in that: include: The module includes a data processing module, a requirement feature solving module, a space-behavior coupling module, and a design parameter decoding module. The data processing module includes: a user trajectory processing unit, which is used to acquire the real trajectory set of various different groups of users within a preset continuous time window, and perform conditional adversarial generation fitting based on the real trajectory set to obtain an enhanced trajectory set; Spatial data processing unit: used to acquire environmental data of the space to be optimized to generate raw point cloud data; and based on the raw point cloud data and the boundary of the space to be optimized, to generate a grid space and a corresponding spatial topology matrix, wherein the grid space includes multiple grid cells, and each grid cell includes center coordinates and grid size; Environmental data processing unit: used to acquire environmental element data within a preset continuous time window and construct an environmental feature tensor; The demand feature solving module includes: a social association matrix solving unit, which is used to calculate the social association matrix based on the spatiotemporal distribution characteristics of the social situations of different groups in the enhanced trajectory set in the grid space; Spatiotemporal heatmap generation unit: used to generate the thermal value of each grid cell based on the center coordinates of the enhanced trajectory set and the grid cell using the kernel density estimation method, thus obtaining the spatiotemporal heatmap tensor of the enhanced trajectory set; Spatiotemporal graph construction unit: used to construct the spatiotemporal graph nodes and corresponding spatiotemporal feature vectors of each grid unit based on the spatial topology matrix and the spatiotemporal heatmap tensor, so as to obtain the spatiotemporal graph structure and the temporal causality matrix; Demand priority graph generation unit: used to perform multi-level graph convolution and mapping on the spatiotemporal graph structure based on the social association matrix, spatial topology matrix and temporal causality matrix to obtain the demand priority graph; The space-behavior coupling module includes: a temporal feature extraction unit, used to calculate the temporal features of each group based on the demand priority map and the spatiotemporal heat map tensor, and obtain the group behavior feature vector; Spatial feature extraction unit: used to enhance the environmental feature tensor according to the demand priority map and then perform spatial convolution operation to obtain environmental spatial topological features; Modal fusion feature extraction unit: used to perform cross-modal feature fusion of the group behavior feature vector and environmental spatial topological features to obtain fused features; Design parameter decoding module: used to decode the parameters of the fused features through a fully connected network to obtain optimized design parameters.

9. The resident-shared, community public space intelligent optimization design system according to claim 8, characterized in that, Also includes: The iterative optimization module includes: A global matching degree calculation unit is used to calculate the group matching degree of the high-frequency activity area of ​​each group based on the optimization design parameters, spatiotemporal heat map tensor, and demand priority map, and obtain the global matching degree of the space to be optimized by weighted summation. Matching degree iterative optimization unit: used to iteratively optimize the optimization design parameters based on the global matching degree until the evaluation function converges and outputs the final design parameters.

Images