Urban renewal full life cycle design method and system based on multi-source heterogeneous data

Abstract

The application provides a city renewal full life cycle design method and system based on multi-source heterogeneous data, relates to the technical field of city renewal, and comprises the following steps: dividing a building into functional sub-regions through a space-time dynamic sensing network and implanting virtual sensing nodes to form a regional dynamic feature map, capture performance change trend characteristics, and generate a dynamic performance tensor; deducing a performance response curve under a transformation parameter combination, and establishing a mapping relationship table; calculating constraint violation degrees and comprehensive constraint scores through interpolation processing; and determining an optimal transformation scheme based on population evolution and local search. The application can realize accurate decision-making for city renewal and improve the level of sustainable development of cities.

Description

Technical Field

[0001] This invention relates to the field of urban renewal technology, and in particular to a method and system for designing the entire lifecycle of urban renewal based on multi-source heterogeneous data. Background Technology

[0002] With the acceleration of urbanization, urban renewal, as an important means to enhance urban functions and improve the living environment, has become an inevitable choice for current urban development. Urban renewal involves many aspects, including building renovation, environmental optimization, and functional enhancement, placing higher demands on the full life-cycle management of buildings. Traditional urban renewal design methods mainly rely on experience-based judgment and simple data analysis, which are insufficient to meet the refined and intelligent renewal needs of today's complex urban environment. In recent years, with the development of technologies such as the Internet of Things, big data, and artificial intelligence, the application of multi-source heterogeneous data in the field of urban renewal has gradually increased, providing new possibilities for accurate analysis and optimization decision-making throughout the entire building life cycle.

[0003] Existing urban renewal design methods have the following defects and shortcomings: Existing technologies struggle to effectively integrate and utilize multi-source heterogeneous data for building functional zoning analysis. Traditional methods typically employ a single data source or simple data overlay, failing to capture the complex relationships and dynamic characteristics between different functional areas within a building. This results in irrational functional zoning and a lack of targeted renovation solutions.

[0004] There is a lack of dynamic prediction capabilities for changes in building performance over time. Existing technologies mostly use static evaluation models, which cannot accurately simulate and predict the dynamic changes in building performance under different combinations of renovation parameters. This makes it difficult to support renovation decisions from a life-cycle perspective, and can easily lead to problems where short-term renovations show significant effects but long-term performance degradation is rapid. Summary of the Invention

[0005] This invention provides a method and system for the full lifecycle design of urban renewal based on multi-source heterogeneous data, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for designing the entire lifecycle of urban renewal based on multi-source heterogeneous data, comprising: Multi-source heterogeneous data corresponding to the target building is input into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions and implants virtual sensing nodes in each functional sub-region. The information interaction between the virtual sensing nodes forms a regional dynamic feature map. At the same time, the recursive attention mechanism is used to capture the performance change trend features of each functional sub-region at different time scales. The spatiotemporal dynamic perception network enhances the regional dynamic feature map and the performance change trend feature through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building. Based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping relationship table is generated based on the performance response curves. The transformation parameter-performance mapping table is subjected to Hermit interpolation. Based on the interpolation results and the constraint threshold of the target building, the constraint violation degree of the performance index is calculated. Based on the constraint violation degree, the comprehensive constraint score of the transformation parameter combination is calculated. Based on the comprehensive constraint score, the modification parameter combination and corresponding fitness are set. A new modification parameter combination is generated through population evolution operation. A local search is performed based on the new modification parameter combination. The optimal parameters obtained from the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the modification parameter combination with the highest fitness is determined as the final modification scheme.

[0007] The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, embeds virtual sensing nodes in each sub-region, and forms a region-level dynamic feature map through information interaction between the virtual sensing nodes. Simultaneously, it utilizes a recursive attention mechanism to capture the performance change trend characteristics of each functional sub-region at different time scales, including: Based on the spatiotemporal dynamic perception network, spatial correlation parameters and functional correlation parameters are calculated respectively, and the target building is divided into multiple functional sub-regions according to the spatial correlation parameters and the functional correlation parameters. For each functional sub-region, the sensing radius of the virtual sensing node is determined based on its area, and the virtual sensing node is implanted within the functional sub-region based on the sensing radius. Calculate the Euclidean distance between any two virtual sensing nodes and combine them to obtain a distance matrix. Generate node information transmission weights based on the distance matrix. Generate initial node features based on the performance monitoring data of each virtual sensing node. Determine the neighborhood range of each virtual sensing node based on the node information transmission weights. Then, weight and aggregate the initial features of the virtual sensing node itself with the initial features of other nodes within its neighborhood range to obtain interaction features. Integrate the interaction features of each virtual sensing node to form a regional dynamic feature map. The temporal state features of each functional sub-region at different time scales are extracted using a recursive gating unit. Attention weight coefficients are calculated based on the temporal state features. The temporal state features at different time scales are weighted and combined according to the attention weight coefficients to obtain the performance change trend features.

[0008] The spatiotemporal dynamic perception network enhances the regional dynamic feature map and the performance change trend features through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building, including: Obtain interaction channels at different scales, calculate the correlation value between the row vector of the regional dynamic feature map and the column vector of the performance change trend feature in each interaction channel, divide the correlation value by the dimension parameter and normalize it to obtain the region-trend correlation matrix; The region-trend correlation matrix is ​​multiplied with the performance change trend feature and then added to the region-level dynamic feature map to obtain a region complementary result. The transpose of the region-trend correlation matrix is ​​multiplied with the region-level dynamic feature map and then added to the performance change trend feature to obtain a trend complementary result. The region complementation results and the trend complementation results are upsampled to obtain aligned region data and aligned trend data, respectively. The aligned region data and the aligned trend data are concatenated and then input into a fully connected network to obtain the interaction weight coefficients. The region-level dynamic feature map and the performance change trend feature are used as residual terms; the alignment region data and the alignment trend data in each interaction channel are multiplied by their corresponding interaction weight coefficients to obtain weighted data. The weighted data of all interaction channels are accumulated and then superimposed with the residual terms to obtain the information fusion tensor. The information fusion tensor is subjected to layer normalization to generate the dynamic performance tensor of the target building.

[0009] Based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping table is generated based on the performance response curves, including: Obtain the combination of modification parameters, and map and transform the combination of modification parameters to the performance feature space to obtain the parameter representation vector; The initial response value is obtained by performing a dot product operation between the parameter representation vector and the feature slice of the dynamic performance tensor at each time step. The initial response value is then weighted and fused with the performance response information of the previous time step to obtain the performance response prediction value. The performance response prediction values ​​of all time steps are calculated sequentially to form the initial response curve. The sum of squared differences between adjacent time step data of the initial response curve is calculated to obtain a smoothing loss term. The smoothing loss term is used as the optimization objective, and the performance response prediction value is iteratively updated through gradient descent until the optimization objective converges, thus obtaining the corrected performance response curve. The maximum value of the corrected performance response curve is extracted as the peak index, the mean value of the curve is calculated as the average index, and the standard deviation of the curve is calculated as the fluctuation index. The change rate of each combination of modification parameters to the peak index, average index and fluctuation index is calculated to obtain the parameter sensitivity. Based on the parameter sensitivity, the sampling interval of each combination of modification parameters is determined, and modification parameter grid points are generated. Based on the modification parameter grid points and the pre-acquired performance indicators, a modification parameter-performance mapping relationship table is generated through support vector regression.

[0010] The aforementioned modification parameter-performance mapping table is subjected to Hermitian interpolation. Based on the interpolation results and the constraint thresholds of the target building, the constraint violation degree of the performance indicators is calculated. Based on the constraint violation degree, the comprehensive constraint score of the modification parameter combination is calculated, including: Interpolation nodes are determined in the modified parameter-performance mapping table. The first derivative is calculated at each interpolation node as the tangent slope, and the second derivative between adjacent interpolation nodes is calculated as the curvature constraint. After determining the interpolation coefficients based on the tangent slope and curvature constraint, interpolation processing is performed to obtain the performance response characteristic quantity. Based on the performance response characteristics, the peak energy consumption threshold, average energy consumption threshold, and energy consumption fluctuation threshold are determined. The peak deviation is obtained by subtracting the peak energy consumption threshold from the peak index, the mean deviation is obtained by subtracting the average energy consumption threshold from the average index, and the fluctuation deviation is obtained by subtracting the fluctuation threshold from the fluctuation index. The squares of the peak deviation, mean deviation, and fluctuation deviation are calculated respectively and then weighted and summed to obtain the constraint violation degree. The disturbance response value of constraint violation caused by the change of the combination of modification parameters is statistically analyzed. Based on the disturbance response value, Monte Carlo simulation is used to generate parameter disturbance samples. The change rate of constraint violation caused by each parameter disturbance sample is calculated. The change rate of constraint violation is standardized to obtain the parameter action coefficient. The constraint violation is multiplied by the parameter action coefficient to obtain the comprehensive constraint score of the combination of modification parameters.

[0011] Based on the comprehensive constraint score, a combination of modification parameters and corresponding fitness are set. A new combination of modification parameters is generated through population evolutionary computation. A local search is then performed based on the new combination of modification parameters, and the optimal parameters obtained from the local search are reintroduced into the population, including: The modified parameter combination is used as the initial population individuals; the mean of the parameter standard deviation of the initial population individuals is calculated and mapped to a preset interval to obtain the fitness adjustment factor; the reciprocal of the comprehensive constraint score of the initial population individuals is multiplied by the fitness adjustment factor to obtain the fitness. The ratio of fitness to the current number of individuals in the population is used as the individual selection probability for roulette wheel selection, while retaining the individual with the highest fitness as the elite individual. Selected individuals are randomly paired, and the parameters of the paired individuals are linearly combined according to a preset crossover coefficient to generate offspring individuals; the ratio of the current generation to the maximum generation is calculated, and the ratio is multiplied by a preset parameter perturbation coefficient to obtain the parameter adjustment range; the parameters of the offspring individuals are adjusted based on the parameter adjustment range to obtain a new combination of modified parameters. When the fitness of the elite individual remains unchanged for a consecutive preset number of generations, a search radius is determined based on the new combination of modification parameters, with the elite individual as the center, and a local search is performed along the coordinate axis within the search radius. If the fitness of the new parameter obtained by the local search is higher than that of the original parameter, the search step size is increased by a preset factor; otherwise, the search step size is decreased by a preset factor until the search step size is less than a preset step size threshold. The parameter with the highest fitness obtained during the local search is taken as the optimal parameter, and a new individual is constructed based on the optimal parameter to replace the individual with the lowest fitness in the population.

[0012] A second aspect of this invention provides an urban renewal full lifecycle design system based on multi-source heterogeneous data, comprising: The first unit is used to input multi-source heterogeneous data corresponding to the target building into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, implants virtual sensing nodes in each functional sub-region, and forms a regional dynamic feature map through information interaction between the virtual sensing nodes. At the same time, it uses a recursive attention mechanism to capture the performance change trend features of each functional sub-region at different time scales. The second unit is used by the spatiotemporal dynamic perception network to perform multi-scale interactive enhancement of the regional dynamic feature map and the performance change trend feature and introduce residual connection correction to generate the dynamic performance tensor of the target building. The third unit is used to deduce the performance response curve of the target building under different combinations of modification parameters based on the dynamic performance tensor, and to generate a modification parameter-performance mapping relationship table based on the performance response curve. The fourth unit is used to perform Hermit interpolation on the modified parameter-performance mapping table, calculate the constraint violation degree of the performance index based on the interpolation result and the constraint threshold of the target building, and calculate the comprehensive constraint score of the modified parameter combination based on the constraint violation degree. The fifth unit is used to set the combination of modification parameters and the corresponding fitness according to the comprehensive constraint score, generate a new combination of modification parameters through population evolution operation, and perform a local search based on the new combination of modification parameters. The optimal parameters obtained by the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the combination of modification parameters with the highest fitness is determined as the final modification scheme.

[0013] A third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] Fourth aspect of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] The beneficial effects of this application are as follows: This invention achieves precise division of functional sub-regions of a target building through a spatiotemporal dynamic sensing network, uses virtual sensing nodes to form regional dynamic feature maps, and effectively captures performance change trends at different time scales through a recursive attention mechanism, thereby realizing comprehensive dynamic monitoring and evaluation of building performance.

[0016] This invention extrapolates performance response curves under a combination of transformation parameters based on dynamic performance tensors, and establishes an accurate transformation parameter-performance mapping relationship through Hermitian interpolation. Combined with constraint violation calculation, a comprehensive constraint score is obtained, providing a scientific basis for the optimization of transformation schemes and significantly improving the accuracy and reliability of urban renewal decisions.

[0017] This invention employs a combination of population evolutionary computation and local search, which can effectively avoid local optima and find the optimal transformation scheme under various constraints. This significantly improves the intelligence level and resource utilization efficiency of the entire life cycle design of urban renewal, and provides strong support for the sustainable development of urban buildings. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the urban renewal full lifecycle design method based on multi-source heterogeneous data, as described in an embodiment of the present invention. Figure 2 A diagram showing the performance comparison of different methods in identifying building energy consumption anomalies; Figure 3 A schematic diagram comparing the convergence performance of different methods in urban building renovation and optimization. Figure 4 A schematic diagram for selecting the three-dimensional model and floor structure of urban buildings; Figure 5 A schematic diagram for managing parameters related to urban building renovation. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] refer to Figures 1 to 5 The present invention provides a method for designing the entire lifecycle of urban renewal based on multi-source heterogeneous data, comprising: Multi-source heterogeneous data corresponding to the target building is input into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions and implants virtual sensing nodes in each functional sub-region. The information interaction between the virtual sensing nodes forms a regional dynamic feature map. At the same time, the recursive attention mechanism is used to capture the performance change trend features of each functional sub-region at different time scales. The spatiotemporal dynamic perception network enhances the regional dynamic feature map and the performance change trend feature through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building. Based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping relationship table is generated based on the performance response curves. The transformation parameter-performance mapping table is subjected to Hermit interpolation. Based on the interpolation results and the constraint threshold of the target building, the constraint violation degree of the performance index is calculated. Based on the constraint violation degree, the comprehensive constraint score of the transformation parameter combination is calculated. Based on the comprehensive constraint score, the modification parameter combination and corresponding fitness are set. A new modification parameter combination is generated through population evolution operation. A local search is performed based on the new modification parameter combination. The optimal parameters obtained from the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the modification parameter combination with the highest fitness is determined as the final modification scheme.

[0022] Multi-source heterogeneous data mainly includes Building Information Modeling (BIM) data, Geographic Information System (GIS) data, Computer-Aided Design (CAD) data, 3D modeling data, IoT sensor data, and historical archive data. BIM data originates from software such as Revit and ArchiCAD, containing information on the building's geometry, materials, and components. GIS data is acquired through platforms such as ArcGIS and QGIS, providing macro-spatial information on urban roads, topography, and infrastructure. CAD data primarily consists of 2D drawings generated by software such as AutoCAD and MicroStation, including traditional architectural design documents and existing building survey data. 3D modeling data, such as models created using SketchUp and 3ds Max, provides information on the building's appearance and spatial form. IoT sensor data comes from various monitoring devices installed inside the building, recording the building's actual operating status. Historical archive data includes unstructured information such as original drawings and maintenance records.

[0023] In a historical district renovation project, the system simultaneously processed historical building survey maps in AutoCAD format, renovation scheme models created in Revit, urban infrastructure data provided by ArcGIS, and point cloud data of building structure conditions collected on-site. The integration and processing of multi-source data consisted of three steps: data format standardization and conversion, spatial reference consistency processing, and semantic-level association establishment.

[0024] The system extracts relevant information from the integrated data for decision support based on different scales and precision requirements. At the macro-planning decision-making level, GIS data and urban models are used to analyze regional functional layout and transportation accessibility, assisting in determining updated regional boundaries. At the architectural design level, BIM models and historical building data are combined to analyze space utilization efficiency and structural safety, providing quantitative basis for renovation plans.

[0025] At the level of construction details, the system integrates BIM component libraries and material databases to generate detailed construction drawings for different modification parameters. The system not only assists in the design of door and window structures but also automatically generates key construction details such as exterior wall nodes and roof details, and performs performance verification to ensure compliance with technical specifications.

[0026] At the construction and operation management level, the system combines BIM models and on-site data to support refined management of the renovation process and intelligent operation of the renovated building. By integrating IoT sensor data, the system can predict equipment failures and recommend maintenance strategies, extending the service life of the renovated building.

[0027] In one optional implementation, the spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, embeds virtual sensing nodes in each functional sub-region, and forms a regional-level dynamic feature map through information interaction between the virtual sensing nodes. Simultaneously, it utilizes a recursive attention mechanism to capture the performance change trend characteristics of each functional sub-region at different time scales, including: Based on the spatiotemporal dynamic perception network, spatial correlation parameters and functional correlation parameters are calculated respectively, and the target building is divided into multiple functional sub-regions according to the spatial correlation parameters and the functional correlation parameters. For each functional sub-region, the sensing radius of the virtual sensing node is determined based on its area, and the virtual sensing node is implanted within the functional sub-region based on the sensing radius. Calculate the Euclidean distance between any two virtual sensing nodes and combine them to obtain a distance matrix. Generate node information transmission weights based on the distance matrix. Generate initial node features based on the performance monitoring data of each virtual sensing node. Determine the neighborhood range of each virtual sensing node based on the node information transmission weights. Then, weight and aggregate the initial features of the virtual sensing node itself with the initial features of other nodes within its neighborhood range to obtain interaction features. Integrate the interaction features of each virtual sensing node to form a regional dynamic feature map. The temporal state features of each functional sub-region at different time scales are extracted using a recursive gating unit. Attention weight coefficients are calculated based on the temporal state features. The temporal state features at different time scales are weighted and combined according to the attention weight coefficients to obtain the performance change trend features.

[0028] Spatial connectivity parameters are determined by calculating the physical connectivity between adjacent spatial units within a building. For example, in an office building, adjacent offices with direct access are assigned a high spatial connectivity value of 0.8; those connected via a public area are assigned a medium value of 0.5; and those located on different floors with poor connectivity are assigned a low value of 0.2. Functional connectivity parameters are calculated based on the usage of the spatial units. For example, spatial units that are both meeting areas are assigned a functional connectivity value of 0.9; meeting areas and office areas are assigned a functional connectivity value of 0.6; and office areas and equipment rooms are assigned a functional connectivity value of 0.3. By combining these two parameters, a three-story office building can be divided into five functional sub-areas: Area A (first-floor office area, 200 square meters), Area B (first-floor meeting area, 100 square meters), Area C (second-floor office area, 250 square meters), Area D (second-floor meeting area, 80 square meters), and Area E (third-floor equipment area, 150 square meters).

[0029] For each functional sub-region, the sensing radius is calculated by multiplying the square root of the area by a coefficient, based on its size. For example, with a coefficient of 0.15, the sensing radius for region A is 2.12 meters, for region B it is 1.5 meters, for region C it is 2.37 meters, for region D it is 1.34 meters, and for region E it is 1.84 meters. Based on the sensing radius, virtual sensing nodes are implanted within each functional sub-region. The number of nodes is proportional to the area, specifically one node for every 50 square meters. Therefore, 4 nodes are implanted in region A, 2 in region B, 5 in region C, 2 in region D, and 3 in region E, for a total of 16 virtual sensing nodes.

[0030] Calculate the Euclidean distance between any two virtual sensing nodes, forming a 16×16 distance matrix. Taking two nodes A1 and A2 in area A as an example, assuming their spatial coordinates are (10,15,0) and (15,20,0) respectively, the Euclidean distance between them is 7.07 meters. Based on the distance matrix, generate node information transmission weights using a Gaussian kernel function transformation; the closer the distance, the greater the weight. For example, for nodes A1 and A2, which are 7.07 meters apart, the information transmission weight is 0.72; for nodes A1 and C1, which are 25 meters apart (located on different floors), the information transmission weight is 0.21.

[0031] The performance monitoring data of each virtual sensor node includes multi-dimensional information such as temperature, humidity, CO2 concentration, and energy consumption. For example, the monitoring data of node A1 at a certain moment are: temperature 24.5℃, humidity 45%, CO2 concentration 650ppm, and lighting energy consumption 120W / h. These data are standardized and combined to form the initial feature vector of the node. The neighborhood range of each virtual sensor node is determined based on the node information transmission weight, with a threshold of 0.4. Nodes with a weight greater than this value are considered neighborhood nodes. For example, the neighborhood of node A1 includes nodes A2, A3, and A4, with weights of 0.72, 0.65, and 0.43, respectively.

[0032] Each node performs a weighted aggregation of its own initial features with the initial features of its neighboring nodes. Taking node A1 as an example, its interaction features are calculated as a weighted combination of its own features and the features of nodes A2, A3, and A4, with weights of 1.0, 0.72, 0.65, and 0.43, respectively. After normalization, these weights are 0.36, 0.26, 0.23, and 0.15. The interaction features of virtual sensing nodes within the same functional sub-region are integrated to form a regional dynamic feature map. For example, the regional dynamic feature map of region A consists of the interaction features of four nodes: A1, A2, A3, and A4, representing the environmental and energy consumption status of the entire region A.

[0033] Taking area A as an example, regional dynamic feature maps of the past 72 hours are collected. Using recursive gating units, the temporal state features of each functional sub-region at different time scales are extracted according to three time scales: short-term (1 hour), medium-term (6 hours), and long-term (24 hours). Through recursive processing, the temporal state features at different time scales are obtained. Short-term features reflect the current regional state, medium-term features capture the change patterns within the working period, and long-term features reflect the daily cycle change patterns.

[0034] Attention weight coefficients are calculated based on the matching degree between features at each time scale and historical patterns. For example, at 9:00 AM on a weekday, the attention weight coefficients for short-term, medium-term, and long-term features in area A are 0.5, 0.3, and 0.2, respectively, indicating that the current state is more critical. However, at night or on weekends, the weight of long-term features rises to 0.5 because historical patterns are more valuable for reference. By weighting and combining the temporal state features at different time scales according to the attention weight coefficients, the final performance change trend feature is obtained. This feature comprehensively reflects the historical, current, and future state change trends of the functional sub-region.

[0035] Through the above implementation methods, the spatiotemporal dynamic sensing network of the present invention can accurately capture the performance change characteristics of different functional areas within a building, providing strong support for building energy consumption optimization and intelligent control.

[0036] In one optional implementation, the spatiotemporal dynamic sensing network enhances the regional dynamic feature map and the performance change trend features through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building, including: Obtain interaction channels at different scales, calculate the correlation value between the row vector of the regional dynamic feature map and the column vector of the performance change trend feature in each interaction channel, divide the correlation value by the dimension parameter and normalize it to obtain the region-trend correlation matrix; The region-trend correlation matrix is ​​multiplied with the performance change trend feature and then added to the region-level dynamic feature map to obtain a region complementary result. The transpose of the region-trend correlation matrix is ​​multiplied with the region-level dynamic feature map and then added to the performance change trend feature to obtain a trend complementary result. The region complementation results and the trend complementation results are upsampled to obtain aligned region data and aligned trend data, respectively. The aligned region data and the aligned trend data are concatenated and then input into a fully connected network to obtain the interaction weight coefficients. The region-level dynamic feature map and the performance change trend feature are used as residual terms; the alignment region data and the alignment trend data in each interaction channel are multiplied by their corresponding interaction weight coefficients to obtain weighted data. The weighted data of all interaction channels are accumulated and then superimposed with the residual terms to obtain the information fusion tensor. The information fusion tensor is subjected to layer normalization to generate the dynamic performance tensor of the target building.

[0037] Three interactive channels at different scales are set up: a small-scale channel, a medium-scale channel, and a large-scale channel, with corresponding dimensional parameters of 8, 16, and 32, respectively. Each interactive channel independently processes regional dynamic feature maps and performance change trend features, performing information exchange at a specific scale. For example, for the small-scale channel, the dimension of the regional dynamic feature map is 128×64, and the dimension of the performance change trend feature is 24×64.

[0038] In each interaction channel, each row vector (1×64 dimension) of the regional dynamic feature map is taken and its dot product is calculated with each column vector (24×1 dimension) of the performance change trend feature, resulting in a 128×24-dimensional relevance matrix. Subsequently, each element in the relevance matrix is ​​divided by the corresponding dimension parameter (e.g., divided by 8 in the small-scale channel), and then normalized using the Softmax function to obtain the region-trend association matrix. This step ensures the stability of feature interactions at different scales. For example, the region-trend association matrix obtained in the small-scale channel has a dimension of 128×24.

[0039] A matrix multiplication operation is performed between the region-trend correlation matrix (128×24 dimension) and the performance change trend feature (24×64 dimension) to obtain an enhanced regional feature with a dimension of 128×64. This enhanced feature is then added to the original region-level dynamic feature map (128×64 dimension) to obtain a complementary region result. Similarly, a matrix multiplication operation is performed between the transpose of the region-trend correlation matrix (24×128 dimension) and the region-level dynamic feature map (128×64 dimension) to obtain an enhanced trend feature with a dimension of 24×64. This enhanced trend feature is then added to the original performance change trend feature (24×64 dimension) to obtain a complementary trend result. This bidirectional enhancement mechanism ensures full interaction between regional and trend information.

[0040] Linear interpolation is used to upsample the region complement results (128×64 dimension) and trend complement results (24×64 dimension) to a common dimension of 256×64, obtaining aligned region data and aligned trend data. The aligned region data and aligned trend data are then concatenated along the feature dimension to obtain a concatenated feature of dimension 256×128. This concatenated feature is processed by a fully connected network, outputting interaction weight coefficients of dimension 1 to adjust the contribution of different scale channels to the final feature. The fully connected network consists of two layers: the first layer has an input dimension of 256×128 and an output dimension of 64, and the second layer has an input dimension of 64 and an output dimension of 1. The ReLU activation function is used.

[0041] When introducing the residual connection correction mechanism, the original regional-level dynamic feature map (128×64 dimension) and performance change trend feature (24×64 dimension) are used as residual terms. For each interaction channel, the aligned region data (256×64 dimension) and aligned trend data (256×64 dimension) are multiplied by their corresponding interaction weight coefficients. For example, if the interaction weight coefficient of the small-scale channel is 0.7, then the aligned region data and aligned trend data of that channel are multiplied by 0.7 respectively. The weighted aligned region data of all interaction channels (small-scale, medium-scale, and large-scale) are summed to obtain the comprehensive region feature; the weighted aligned trend data of all interaction channels are summed to obtain the comprehensive trend feature. The comprehensive region feature is added to the original regional-level dynamic feature map, and the comprehensive trend feature is added to the original performance change trend feature to obtain the fused region feature and trend feature. The fused region feature and trend feature are concatenated along the feature dimensions to obtain the information fusion tensor with a dimension of (128+24)×64.

[0042] The mean and variance of the information fusion tensor along the feature dimensions are calculated, and then each element is standardized to obtain a standardized tensor with a mean of 0 and a variance of 1. The distribution of the standardized tensor is then adjusted using learnable scaling and translation parameters to generate the dynamic performance tensor of the target building. The initial value of the scaling parameter for layer normalization is set to 1, and the initial value of the translation parameter is set to 0.

[0043] Using the methods described above, the spatiotemporal dynamic sensing network can effectively capture the complex correlation between regional dynamic features and performance change trends, generating a dynamic performance tensor containing rich spatiotemporal information. In practical applications, for a 10-story commercial building, this method was used to process energy consumption data collected over 24 hours, successfully identifying abnormal energy consumption patterns on each floor at different times with an accuracy rate of 92.5%, which is 15% higher than traditional methods, providing a reliable basis for building energy efficiency optimization.

[0044] In one optional implementation, based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping table is generated based on the performance response curves, including: Obtain the combination of modification parameters, and map and transform the combination of modification parameters to the performance feature space to obtain the parameter representation vector; The initial response value is obtained by performing a dot product operation between the parameter representation vector and the feature slice of the dynamic performance tensor at each time step. The initial response value is then weighted and fused with the performance response information of the previous time step to obtain the performance response prediction value. The performance response prediction values ​​of all time steps are calculated sequentially to form the initial response curve. The sum of squared differences between adjacent time step data of the initial response curve is calculated to obtain a smoothing loss term. The smoothing loss term is used as the optimization objective, and the performance response prediction value is iteratively updated through gradient descent until the optimization objective converges, thus obtaining the corrected performance response curve. The maximum value of the corrected performance response curve is extracted as the peak index, the mean value of the curve is calculated as the average index, and the standard deviation of the curve is calculated as the fluctuation index. The change rate of each combination of modification parameters to the peak index, average index and fluctuation index is calculated to obtain the parameter sensitivity. Based on the parameter sensitivity, the sampling interval of each combination of modification parameters is determined, and modification parameter grid points are generated. Based on the modification parameter grid points and the pre-acquired performance indicators, a modification parameter-performance mapping relationship table is generated through support vector regression.

[0045] The system obtains combinations of modification parameters, such as a window-to-wall ratio ranging from 0.2 to 0.6, an insulation layer thickness ranging from 50 mm to 200 mm, and a shading angle ranging from 0° to 60°. These parameters are transformed into parameter representation vectors in the performance feature space through mapping transformation. For example, the combination of a window-to-wall ratio of 0.3, an insulation layer thickness of 100 mm, and a shading angle of 30° is mapped into a 128-dimensional representation vector [0.15, 0.23, ..., 0.42] through a pre-trained encoder network.

[0046] The dynamic performance tensor contains 24-hour feature slices, each with a dimension of 128. During computation, the parameter representation vector is multiplied by the feature slice at each time step; for example, at time 0, the initial response value is 26.8℃. Considering the building's thermal inertia, the current initial response value is weighted and fused with the predicted value from the previous time step, with a weight ratio of 0.7:0.3, resulting in a predicted performance response value of 26.5℃ at time 0. This process is repeated to calculate the temperature prediction values ​​for all 24 time steps of the day, forming the initial response curve [26.5, 26.8, ..., 29.7, 28.9]℃.

[0047] Taking the initial curve mentioned above as an example, the sum of squares of the differences between adjacent time steps yielded a smoothing loss of 12.6. Using this as the optimization objective, the predicted value was iteratively updated using the gradient descent algorithm. Specifically, the learning rate was set to 0.01, the maximum number of iterations was 500, and convergence was determined when the smoothing loss changed by less than 0.001 after 5 consecutive iterations. After 187 iterations, the smoothing loss decreased to 2.3, resulting in the corrected performance response curve [26.4, 26.7, ..., 29.5, 29.0]℃.

[0048] Key indicators were extracted from the corrected performance response curve: the maximum temperature of 29.5℃ was taken as the peak value, the average temperature of 27.9℃ was calculated as the average value, and the temperature standard deviation of 0.8℃ was calculated as the fluctuation value. By performing the above calculations on different combinations of modification parameters, the correspondence between parameters and performance indicators was obtained.

[0049] The sensitivity of the parameters was analyzed to determine the sampling strategy. In practice, taking the window-to-wall ratio as an example, calculations showed that when the window-to-wall ratio changed from 0.2 to 0.3, the peak index changed by 5.2%, the average index by 3.8%, and the fluctuation index by 8.6%. Considering the change rates of these three indicators, the sensitivity coefficient for the window-to-wall ratio was determined to be 5.9%. Similarly, the sensitivity coefficient for insulation layer thickness was calculated to be 3.2%, and the sensitivity coefficient for shading angle was calculated to be 4.5%.

[0050] Based on the sensitivity analysis results, sampling intervals were set for each parameter: window-to-wall ratio sampling interval was 0.05, insulation layer thickness sampling interval was 25mm, and shading angle sampling interval was 10°. Based on these intervals, a grid of modification parameter points was generated, totaling 343 combination points (9 window-to-wall ratio values ​​× 7 insulation layer thickness values ​​× 5 shading angle values).

[0051] For each grid point, the aforementioned method is applied to calculate performance metrics, and a training dataset is constructed. An example dataset is: {Window-to-wall ratio: 0.25, Insulation thickness: 75mm, Shading angle: 20°} → {Peak temperature: 28.9℃, Average temperature: 27.2℃, Fluctuation index: 0.7℃}.

[0052] Support vector regression was used to establish the mapping relationship between parameters and performance. A radial basis function kernel was selected, with the penalty coefficient C set to 100 and the epsilon value set to 0.1. After training, the model can achieve rapid mapping from parameters to performance. For example, with the input parameter combination {window-to-wall ratio: 0.32, insulation layer thickness: 115mm, shading angle: 35°}, the predicted performance indicators are {peak temperature: 28.4℃, average temperature: 26.9℃, fluctuation index: 0.75℃}.

[0053] The final generated modification parameter-performance mapping table contains all grid point parameter combinations and corresponding performance index values, achieving a comprehensive mapping from parameter space to performance space. Using this table, architects can quickly select suitable modification parameter combinations based on target performance requirements. For example, when the required indoor temperature peak is no more than 28.5℃ and the average temperature is no higher than 27.0℃, they can select parameter combinations that meet the conditions from the table, significantly improving the efficiency of modification scheme decision-making.

[0054] In one optional implementation, the modification parameter-performance mapping table is subjected to Hermitian interpolation. Based on the interpolation result and the constraint threshold of the target building, the constraint violation degree of the performance index is calculated. Based on the constraint violation degree, the comprehensive constraint score of the modification parameter combination is calculated, including: Interpolation nodes are determined in the modified parameter-performance mapping table. The first derivative is calculated at each interpolation node as the tangent slope, and the second derivative between adjacent interpolation nodes is calculated as the curvature constraint. After determining the interpolation coefficients based on the tangent slope and curvature constraint, interpolation processing is performed to obtain the performance response characteristic quantity. Based on the performance response characteristics, the peak energy consumption threshold, average energy consumption threshold, and energy consumption fluctuation threshold are determined. The peak deviation is obtained by subtracting the peak energy consumption threshold from the peak index, the mean deviation is obtained by subtracting the average energy consumption threshold from the average index, and the fluctuation deviation is obtained by subtracting the fluctuation threshold from the fluctuation index. The squares of the peak deviation, mean deviation, and fluctuation deviation are calculated respectively and then weighted and summed to obtain the constraint violation degree. The disturbance response value of constraint violation caused by the change of the combination of modification parameters is statistically analyzed. Based on the disturbance response value, Monte Carlo simulation is used to generate parameter disturbance samples. The change rate of constraint violation caused by each parameter disturbance sample is calculated. The change rate of constraint violation is standardized to obtain the parameter action coefficient. The constraint violation is multiplied by the parameter action coefficient to obtain the comprehensive constraint score of the combination of modification parameters.

[0055] The modification parameters can include three dimensions: window-to-wall ratio, insulation thickness, and shading angle. Five points are selected as interpolation nodes for the window-to-wall ratio, ranging from 0.2 to 0.6: 0.2, 0.3, 0.4, 0.5, and 0.6. Five points are selected for the insulation thickness, ranging from 50mm to 150mm: 50mm, 75mm, 100mm, 125mm, and 150mm. Five points are selected for the shading angle, ranging from 0° to 60°: 0°, 15°, 30°, 45°, and 60°. The first derivative value is calculated at each interpolation node as the tangent slope. For example, at the node with a window-to-wall ratio of 0.3, the first derivative of energy consumption performance with respect to the window-to-wall ratio is 2.5 kWh / (m²). 2 •a) indicates that for every 0.1 increase in the window-to-wall ratio, building energy consumption increases by 2.5 kWh / (m²). 2•a). For the calculation of the second derivative between adjacent interpolation nodes, taking the window-to-wall ratio range of 0.2 to 0.3 as an example, the second derivative is 0.8 kWh / (m²). 2 •a) is used as a curvature constraint to control the shape change of the interpolation curve.

[0056] After determining the interpolation coefficients based on the aforementioned tangent slope and curvature constraints, Hermitian interpolation is performed on any point in the modification parameter space. For the parameter combination of a window-to-wall ratio of 0.35, insulation thickness of 85mm, and shading angle of 25°, the building's average annual energy consumption is calculated to be 120kWh / (m²). 2 •a) Peak summer energy consumption is 25 kWh / (m³). 2 (month), the standard deviation of energy consumption fluctuation is 5 kWh / (m²). 2 (Month). The performance metrics obtained from these interpolations are the performance response characteristic quantities.

[0057] Based on building codes and owner requirements, the peak energy consumption threshold is set at 22 kWh / (m³). 2 (month), with an average energy consumption threshold of 115 kWh / (m³). 2 •a) The energy consumption fluctuation threshold is 4 kWh / (m³). 2 (Month). The actual performance indicators are compared with the thresholds to calculate the peak deviation, mean deviation, and fluctuation deviation. Taking the above parameter combination as an example, the peak deviation is 25-22=3kWh / (m²). 2 (Month), the mean deviation is 120-115=5kWh / (m³). 2 •a) The fluctuation deviation is 5-4=1kWh / (m 2 ·moon).

[0058] If the weights for peak deviation, mean deviation, and fluctuation deviation are set to 0.4, 0.4, and 0.2 respectively, then the constraint violation degree is calculated as: 0.4 × 3 2 +0.4×5 2 +0.2×1 2 =0.4×9+0.4×25+0.2×1=3.6+10+0.2=13.8. The larger the constraint violation value, the less the parameter combination meets the building performance constraint requirements.

[0059] To assess the impact of changes in modification parameters on constraint violation, the response values ​​of parameter disturbances need to be statistically analyzed. Increasing the window-to-wall ratio by 0.01 from a window-to-wall ratio of 0.35, while keeping the insulation thickness and shading angle unchanged, results in a constraint violation of 13.8 changing to 14.5, a change rate of (14.5-13.8) / 13.8×100%=5.07%. Increasing the insulation thickness by 5mm results in a constraint violation of 13.2, a change rate of (13.2-13.8) / 13.8×100%=-4.35%. Increasing the shading angle by 5° results in a constraint violation of 13.0, a change rate of (13.0-13.8) / 13.8×100%=-5.80%.

[0060] Monte Carlo simulation was used to generate 1000 sets of parameter perturbation samples. In each set, the window-to-wall ratio, insulation thickness, and shading angle were randomly selected within their allowable ranges. The rate of change of constraint violation caused by each set of perturbation samples was calculated, and the statistical distribution was obtained. The rate of change was standardized to obtain the action coefficients of each parameter. After standardization, the action coefficients for the window-to-wall ratio, insulation thickness, and shading angle were 0.4, 0.3, and 0.3, respectively.

[0061] For the above parameter combination, the overall constraint score is 13.8 × (0.4 + 0.3 + 0.3) = 13.8. Furthermore, the contribution of each parameter to the constraint score can be calculated. The contribution of the window-to-wall ratio is 13.8 × 0.4 = 5.52, the contribution of the insulation thickness is 13.8 × 0.3 = 4.14, and the contribution of the shading angle is 13.8 × 0.3 = 4.14.

[0062] By adjusting the combination of modification parameters, such as reducing the window-to-wall ratio to 0.3, increasing the insulation thickness to 100mm, and increasing the shading angle to 30°, the constraint violation rate was reduced to 7.2, and the overall constraint score was also reduced to 7.2 accordingly, indicating that the adjusted parameter combination is more in line with the building performance constraint requirements.

[0063] In summary, this implementation method establishes a precise mapping relationship between renovation parameters and building performance through Hermit interpolation, calculates a comprehensive constraint score based on constraint violation degree and parameter action coefficient, provides a quantitative decision-making basis for building renovation, and effectively optimizes building performance.

[0064] In one optional implementation, the modification parameter combination and corresponding fitness are set according to the comprehensive constraint score, a new modification parameter combination is generated through population evolution operation, and a local search is performed based on the new modification parameter combination. The optimal parameters obtained from the local search are then added back to the population, including: The modified parameter combination is used as the initial population individuals; the mean of the parameter standard deviation of the initial population individuals is calculated and mapped to a preset interval to obtain the fitness adjustment factor; the reciprocal of the comprehensive constraint score of the initial population individuals is multiplied by the fitness adjustment factor to obtain the fitness. The ratio of fitness to the current number of individuals in the population is used as the individual selection probability for roulette wheel selection, while retaining the individual with the highest fitness as the elite individual. Selected individuals are randomly paired, and the parameters of the paired individuals are linearly combined according to a preset crossover coefficient to generate offspring individuals; the ratio of the current generation to the maximum generation is calculated, and the ratio is multiplied by a preset parameter perturbation coefficient to obtain the parameter adjustment range; the parameters of the offspring individuals are adjusted based on the parameter adjustment range to obtain a new combination of modified parameters. When the fitness of the elite individual remains unchanged for a consecutive preset number of generations, a search radius is determined based on the new combination of modification parameters, with the elite individual as the center, and a local search is performed along the coordinate axis within the search radius. If the fitness of the new parameter obtained by the local search is higher than that of the original parameter, the search step size is increased by a preset factor; otherwise, the search step size is decreased by a preset factor until the search step size is less than a preset step size threshold. The parameter with the highest fitness obtained during the local search is taken as the optimal parameter, and a new individual is constructed based on the optimal parameter to replace the individual with the lowest fitness in the population.

[0065] For a certain production line renovation project, the initial population contains 20 individuals, each with 5 renovation parameters: equipment upgrade rate, automation level, process optimization coefficient, layout adjustment range, and personnel allocation ratio. These parameters are randomly generated within the ranges of [0.1, 0.9], [0.2, 0.8], [0.3, 0.7], [0.2, 0.6], and [0.4, 0.8], respectively.

[0066] Calculate the standard deviation of 20 individuals across 5 parameters to obtain 5 standard deviation values. Take the mean of these deviations and denote it as σ. Map σ to the interval [0.8, 1.2] to obtain the fitness adjustment factor k. The mapping formula can be expressed as follows: when σ is less than 0.1, k is 0.8; when σ is greater than 0.3, k is 1.2; when σ is between [0.1, 0.3], k is mapped according to a linear relationship. For example, when σ = 0.2, k = 1.0.

[0067] The fitness is obtained by multiplying the reciprocal of the overall constraint score of an individual in the initial population by the fitness adjustment factor. The overall constraint score considers multiple dimensions such as cost, time, and quality. For example, if an individual has an overall constraint score of 0.5 and a fitness adjustment factor of 1.1, then the fitness of that individual is 1 / 0.5 × 1.1 = 2.2.

[0068] In a certain generation of evolution, there are 20 individuals with fitness values ​​of [2.2, 1.8, 1.5, ..., 0.9], and a total fitness of 28.6. The probabilities of each individual being selected are [2.2 / 28.6, 1.8 / 28.6, 1.5 / 28.6, ..., 0.9 / 28.6]. A roulette wheel method is used to select 19 individuals, and the individual with the highest fitness (2.2) is directly retained as the elite individual for the next generation.

[0069] Nineteen selected individuals are randomly paired, and the remaining one is paired with an elite individual, forming 10 pairs of parent individuals. Assuming a crossover coefficient of 0.7, the parameters of the offspring individual S for each pair of parent individuals P1 and P2 are calculated as follows: S = 0.7 × P1 + 0.3 × P2.

[0070] Calculate the ratio of the current generation to the maximum generation. For example, if the current generation is 30 and the maximum generation is 100, the ratio is 0.3. If the preset parameter perturbation coefficient is 0.2, then the parameter adjustment range is 0.3 × 0.2 = 0.06. Randomly perturb each parameter of the offspring individual within the range of [-0.06, 0.06] to obtain a new combination of modification parameters.

[0071] When the fitness of an elite individual remains unchanged for 5 consecutive generations, a local search is performed within the range of 10% of the parameter range, centered on the elite individual, along the 5 parameter axes.

[0072] If the fitness of the new parameters obtained from the local search is higher than that of the original parameters, the search step size is increased by 2 times; otherwise, the search step size is decreased by 0.5 times, until the search step size is less than a preset step size threshold (e.g., 0.001). The parameter with the highest fitness obtained during the local search is taken as the optimal parameter, and a new individual is constructed based on this optimal parameter to replace the individual with the lowest fitness in the population.

[0073] In a specific application case, a manufacturing company's production line renovation project had an initial population size of 20, a maximum evolutionary generation of 100, a crossover coefficient of 0.7, a parameter perturbation coefficient of 0.2, and a local search trigger condition of 5 consecutive generations of elite individuals with unchanged fitness. The initial search step size was 0.05, with an expansion factor of 2 and a shrinking factor of 0.5, and a step size threshold of 0.001. After 37 generations of evolution, the algorithm triggered a local search in the 30th generation. After the local search, the optimal parameter combination was [0.63, 0.54, 0.42, 0.35, 0.67], corresponding to a fitness value of 2.85 and a comprehensive constraint score of 0.35. Compared to the initial optimal fitness value of 2.2, this represents a 29.5% improvement. This parameter combination was applied to the actual production line renovation, resulting in a 15% reduction in cost, a 23% increase in production efficiency, and a 10% improvement in product quality, validating the effectiveness of the proposed method.

[0074] By combining population evolution algorithms with local search using the above method, the advantages of global search and local fine-grained search are fully utilized. While ensuring population diversity, the search efficiency is improved, and the problem of getting trapped in local optima is effectively avoided. This provides an effective solution for optimizing modification parameters under complex constraints.

[0075] For example, this application also provides a data fusion platform system based on BIM technology. This system achieves seamless collaboration between CAD clients and Revit clients by constructing a unified core functional module for data fusion, and provides comprehensive data management and security through electronic document modules and system management modules.

[0076] The core functional module of data fusion constitutes the technical core of this system. At its top level, this module forms a complete project coordination system through project management, BQ management, and project Gantt charts, enabling all activities throughout the project lifecycle to be planned, monitored, and adjusted on a unified platform. Building upon this, the data fusion layer deeply integrates BIM model management functions into the platform, achieving three-dimensional hierarchical organization of architectural space through spatial management, and providing designers with precise positioning and annotation capabilities in conjunction with auxiliary points and drawing tools. The system's supported two-dimensional linked sectioning function breaks down the barriers between traditional two-dimensional drawings and three-dimensional models. When a user performs a sectioning operation in the two-dimensional view, the three-dimensional model responds in real time and displays the corresponding sectioning result. This two-way linkage mechanism significantly improves design efficiency.

[0077] In terms of model management, the system provides a clear hierarchical relationship of components through a model tree, enabling the orderly organization of thousands of components in complex architectural projects. The attribute setting function allows designers to batch edit the attributes of selected components, while the attribute viewing function supports quick retrieval of detailed information for any component. For common concrete components in building engineering, the system provides specialized management tools, enabling professional management of key parameters such as concrete strength grade and reinforcement information. The system also has multi-format loading capabilities, being compatible with the file formats of mainstream BIM software, ensuring that model data from different sources can be effectively integrated by the platform.

[0078] Data analysis capabilities are a key feature of this system. Collision analysis automatically detects spatial conflicts between different professional components in the model, enabling the identification and resolution of potential construction problems during the design phase. The floor slab drop analysis function assesses the rationality of changes in building floor slab elevation, ensuring the design meets functional requirements and regulatory specifications. The economic analysis module automatically extracts quantities from component information in the BIM model and dynamically calculates costs using a cost database, providing an economic basis for project decisions. Planning analysis and simulation analysis functions evaluate the project's planning layout and building performance at a macro level, using visualized analysis results to assist in scheme optimization.

[0079] At the construction management level, the system organically integrates resources, schedule, and cost. Resource and schedule analysis monitors the matching relationship between the usage of various resources and the construction progress, promptly identifying resource supply bottlenecks or idle waste issues. Scheduling analysis dynamically optimizes the construction plan based on the actual progress of the project, ensuring that processes on the critical path are prioritized. The construction simulation analysis function uses 3D animation to preview the construction process, enabling managers to identify construction difficulties in a virtual environment and develop targeted technical solutions.

[0080] The collaborative design module integrates multiple aspects of work, including model collaboration, time management, cost control, and quality inspection. Through its project archive function, this module unifies the archiving and management of Revit modeling results, quantity survey data, and on-site inspection records, forming a complete project data asset. It integrates BIM functionality, supporting the merging of models from different disciplines such as architecture, structure, and MEP into a unified model, resolving interdisciplinary conflicts through clash detection and coordination meetings. The system innovatively introduces a GIS earth module, combining the BIM model with a geographic information system, supporting the import and visualization of laser measurement data, enabling project analysis and presentation within a real geographical environment.

[0081] The system's data input and maintenance mechanisms ensure data accuracy and timeliness. Topographic data, imported through a platform provided by a professional surveying unit, serves as a benchmark for site design. Quantity survey data is linked to the BIM model, enabling automatic extraction and dynamic updating of quantities. On-site measurement data is collected in real-time via mobile terminals and synchronized to the platform, ensuring consistency between the design and the site. BIM data from different disciplines are organized and stored on the platform according to unified standards, ensuring data traceability and reusability.

[0082] This system was successfully applied to a large industrial park renovation project. The project first used oblique photogrammetry to obtain a 3D model of the park, realistically reproducing existing buildings, roads, green spaces, and other site elements. Based on this, the design team created a BIM model of the renovation plan using the Revit client. By overlaying the new building model with the existing model, designers could intuitively assess the impact of the new building on the surrounding environment. The system's sectioning function was used to display the internal structure of key components, assisting in structural optimization design. The scheme comparison function allowed the client to simultaneously view the effects of multiple design schemes in the same scene. Using auxiliary points and composition tools, designers precisely annotated key dimensions and technical requirements on the 3D model; this annotation information was automatically linked to the corresponding components, forming a complete design document system.

[0083] The system management module provides a fundamental guarantee for the secure and stable operation of the platform. Through personnel and role management, the system supports setting differentiated operation permissions for different users, ensuring that sensitive data is only accessible to authorized personnel. Password management and the permission engine employ multiple encryption mechanisms to protect user credentials and prevent unauthorized access. The optimized graphics engine can smoothly render large models containing millions of components, meeting the display needs of complex projects. The server management module monitors system load in real time and automatically allocates computing resources during peak access periods to ensure a stable user experience. A secure data connection mechanism ensures encrypted transmission when exchanging data with external systems, and project transfer and storage functions support version management and off-site backup of project data, effectively preventing the risk of data loss.

[0084] This invention provides an urban renewal full lifecycle design system based on multi-source heterogeneous data, the system comprising: The first unit is used to input multi-source heterogeneous data corresponding to the target building into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, implants virtual sensing nodes in each functional sub-region, and forms a regional dynamic feature map through information interaction between the virtual sensing nodes. At the same time, it uses a recursive attention mechanism to capture the performance change trend features of each functional sub-region at different time scales. The second unit is used by the spatiotemporal dynamic perception network to perform multi-scale interactive enhancement of the regional dynamic feature map and the performance change trend feature and introduce residual connection correction to generate the dynamic performance tensor of the target building. The third unit is used to deduce the performance response curve of the target building under different combinations of modification parameters based on the dynamic performance tensor, and to generate a modification parameter-performance mapping relationship table based on the performance response curve. The fourth unit is used to perform Hermit interpolation on the modified parameter-performance mapping table, calculate the constraint violation degree of the performance index based on the interpolation result and the constraint threshold of the target building, and calculate the comprehensive constraint score of the modified parameter combination based on the constraint violation degree. The fifth unit is used to set the combination of modification parameters and the corresponding fitness according to the comprehensive constraint score, generate a new combination of modification parameters through population evolution operation, and perform a local search based on the new combination of modification parameters. The optimal parameters obtained by the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the combination of modification parameters with the highest fitness is determined as the final modification scheme.

[0085] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0086] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0087] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for designing the entire lifecycle of urban renewal based on multi-source heterogeneous data, characterized in that: include: Multi-source heterogeneous data corresponding to the target building is input into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions and implants virtual sensing nodes in each functional sub-region. The information interaction between the virtual sensing nodes forms a regional dynamic feature map. At the same time, the recursive attention mechanism is used to capture the performance change trend features of each functional sub-region at different time scales. The spatiotemporal dynamic perception network enhances the regional dynamic feature map and the performance change trend feature through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building. Based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping relationship table is generated based on the performance response curves. The transformation parameter-performance mapping table is subjected to Hermit interpolation. Based on the interpolation results and the constraint threshold of the target building, the constraint violation degree of the performance index is calculated. Based on the constraint violation degree, the comprehensive constraint score of the transformation parameter combination is calculated. Based on the comprehensive constraint score, the modification parameter combination and corresponding fitness are set. A new modification parameter combination is generated through population evolution operation. A local search is performed based on the new modification parameter combination. The optimal parameters obtained from the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the modification parameter combination with the highest fitness is determined as the final modification scheme.

2. The method according to claim 1, characterized in that, The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, embeds virtual sensing nodes in each sub-region, and forms a region-level dynamic feature map through information interaction between the virtual sensing nodes. Simultaneously, it utilizes a recursive attention mechanism to capture the performance change trend characteristics of each functional sub-region at different time scales, including: Based on the spatiotemporal dynamic perception network, spatial correlation parameters and functional correlation parameters are calculated respectively, and the target building is divided into multiple functional sub-regions according to the spatial correlation parameters and the functional correlation parameters. For each functional sub-region, the sensing radius of the virtual sensing node is determined based on its area, and the virtual sensing node is implanted within the functional sub-region based on the sensing radius. Calculate the Euclidean distance between any two virtual sensing nodes and combine them to obtain a distance matrix. Generate node information transmission weights based on the distance matrix. Generate initial node features based on the performance monitoring data of each virtual sensing node. Determine the neighborhood range of each virtual sensing node based on the node information transmission weights. Then, weight and aggregate the initial features of the virtual sensing node itself with the initial features of other nodes within its neighborhood range to obtain interaction features. Integrate the interaction features of each virtual sensing node to form a regional dynamic feature map. The temporal state features of each functional sub-region at different time scales are extracted using a recursive gating unit. The attention weight coefficient is calculated based on the temporal state features. The temporal state features at different time scales are weighted and combined according to the attention weight coefficient to obtain the performance change trend features.

3. The method according to claim 1, characterized in that, The spatiotemporal dynamic perception network enhances the regional dynamic feature map and the performance change trend features through multi-scale interaction and introduces residual connection correction to generate the dynamic performance tensor of the target building, including: Obtain interaction channels at different scales, calculate the correlation value between the row vector of the regional dynamic feature map and the column vector of the performance change trend feature in each interaction channel, divide the correlation value by the dimension parameter and normalize it to obtain the region-trend correlation matrix; The region-trend correlation matrix is ​​multiplied with the performance change trend feature and then added to the region-level dynamic feature map to obtain a region complementary result. The transpose of the region-trend correlation matrix is ​​multiplied with the region-level dynamic feature map and then added to the performance change trend feature to obtain a trend complementary result. The region complementation results and the trend complementation results are upsampled to obtain aligned region data and aligned trend data, respectively. The aligned region data and the aligned trend data are concatenated and then input into a fully connected network to obtain the interaction weight coefficients. The region-level dynamic feature map and the performance change trend feature are used as residual terms; the alignment region data and the alignment trend data in each interaction channel are multiplied by their corresponding interaction weight coefficients to obtain weighted data. The weighted data of all interaction channels are accumulated and then superimposed with the residual terms to obtain the information fusion tensor. The information fusion tensor is subjected to layer normalization to generate the dynamic performance tensor of the target building.

4. The method according to claim 1, characterized in that, Based on the dynamic performance tensor, the performance response curves of the target building under different combinations of modification parameters are derived, and a modification parameter-performance mapping table is generated based on the performance response curves, including: Obtain the combination of modification parameters, and map and transform the combination of modification parameters to the performance feature space to obtain the parameter representation vector; The initial response value is obtained by performing a dot product operation between the parameter representation vector and the feature slice of the dynamic performance tensor at each time step. The initial response value is then weighted and fused with the performance response information of the previous time step to obtain the performance response prediction value. The performance response prediction values ​​of all time steps are calculated sequentially to form the initial response curve. The sum of squared differences between adjacent time step data of the initial response curve is calculated to obtain a smoothing loss term. The smoothing loss term is used as the optimization objective, and the performance response prediction value is iteratively updated through gradient descent until the optimization objective converges, thus obtaining the corrected performance response curve. The maximum value of the corrected performance response curve is extracted as the peak index, the mean value of the curve is calculated as the average index, and the standard deviation of the curve is calculated as the fluctuation index. The change rate of each combination of modification parameters to the peak index, average index and fluctuation index is calculated to obtain the parameter sensitivity. Based on the parameter sensitivity, the sampling interval of each combination of modification parameters is determined, and modification parameter grid points are generated. Based on the modification parameter grid points and the pre-acquired performance indicators, a modification parameter-performance mapping relationship table is generated through support vector regression.

5. The method according to claim 1, characterized in that, The aforementioned modification parameter-performance mapping table is subjected to Hermitian interpolation. Based on the interpolation results and the constraint thresholds of the target building, the constraint violation degree of the performance indicators is calculated. Based on the constraint violation degree, the comprehensive constraint score of the modification parameter combination is calculated, including: Interpolation nodes are determined in the modified parameter-performance mapping table. The first derivative is calculated at each interpolation node as the tangent slope, and the second derivative between adjacent interpolation nodes is calculated as the curvature constraint. After determining the interpolation coefficients based on the tangent slope and curvature constraint, interpolation processing is performed to obtain the performance response characteristic quantity. Based on the performance response characteristics, the peak energy consumption threshold, average energy consumption threshold, and energy consumption fluctuation threshold are determined. The peak deviation is obtained by subtracting the peak energy consumption threshold from the peak index, the mean deviation is obtained by subtracting the average energy consumption threshold from the average index, and the fluctuation deviation is obtained by subtracting the fluctuation threshold from the fluctuation index. The squares of the peak deviation, mean deviation, and fluctuation deviation are calculated respectively and then weighted and summed to obtain the constraint violation degree. The disturbance response value of constraint violation caused by the change of the combination of modification parameters is statistically analyzed. Based on the disturbance response value, Monte Carlo simulation is used to generate parameter disturbance samples. The change rate of constraint violation caused by each parameter disturbance sample is calculated. The change rate of constraint violation is standardized to obtain the parameter action coefficient. The constraint violation is multiplied by the parameter action coefficient to obtain the comprehensive constraint score of the combination of modification parameters.

6. The method according to claim 1, characterized in that, Based on the comprehensive constraint score, a combination of modification parameters and corresponding fitness are set. A new combination of modification parameters is generated through population evolutionary computation. A local search is then performed based on the new combination of modification parameters, and the optimal parameters obtained from the local search are reintroduced into the population, including: The modified parameter combination is used as the initial population individuals; the mean of the parameter standard deviation of the initial population individuals is calculated and mapped to a preset interval to obtain the fitness adjustment factor; the reciprocal of the comprehensive constraint score of the initial population individuals is multiplied by the fitness adjustment factor to obtain the fitness. The ratio of fitness to the current number of individuals in the population is used as the individual selection probability for roulette wheel selection, while retaining the individual with the highest fitness as the elite individual. Selected individuals are randomly paired, and the parameters of the paired individuals are linearly combined according to a preset crossover coefficient to generate offspring individuals; the ratio of the current generation to the maximum generation is calculated, and the ratio is multiplied by a preset parameter perturbation coefficient to obtain the parameter adjustment range; the parameters of the offspring individuals are adjusted based on the parameter adjustment range to obtain a new combination of modified parameters. When the fitness of the elite individual remains unchanged for a consecutive preset number of generations, a search radius is determined based on the new combination of modification parameters, with the elite individual as the center, and a local search is performed along the coordinate axis within the search radius. If the fitness of the new parameter obtained by the local search is higher than that of the original parameter, the search step size is increased by a preset factor; otherwise, the search step size is decreased by a preset factor until the search step size is less than a preset step size threshold. The parameter with the highest fitness obtained during the local search is taken as the optimal parameter, and a new individual is constructed based on the optimal parameter to replace the individual with the lowest fitness in the population.

7. A city renewal full lifecycle design system based on multi-source heterogeneous data, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to input multi-source heterogeneous data corresponding to the target building into the spatiotemporal dynamic perception network. The spatiotemporal dynamic perception network divides the target building into multiple functional sub-regions, implants virtual sensing nodes in each functional sub-region, and forms a regional dynamic feature map through information interaction between the virtual sensing nodes. At the same time, it uses a recursive attention mechanism to capture the performance change trend features of each functional sub-region at different time scales. The second unit is used by the spatiotemporal dynamic perception network to perform multi-scale interactive enhancement of the regional dynamic feature map and the performance change trend feature and introduce residual connection correction to generate the dynamic performance tensor of the target building. The third unit is used to deduce the performance response curve of the target building under different combinations of modification parameters based on the dynamic performance tensor, and to generate a modification parameter-performance mapping relationship table based on the performance response curve. The fourth unit is used to perform Hermit interpolation on the modified parameter-performance mapping table, calculate the constraint violation degree of the performance index based on the interpolation result and the constraint threshold of the target building, and calculate the comprehensive constraint score of the modified parameter combination based on the constraint violation degree. The fifth unit is used to set the combination of modification parameters and the corresponding fitness according to the comprehensive constraint score, generate a new combination of modification parameters through population evolution operation, and perform a local search based on the new combination of modification parameters. The optimal parameters obtained by the local search are added back to the population. When the fitness change rate of the optimal solution in the population is less than the preset convergence coefficient, the combination of modification parameters with the highest fitness is determined as the final modification scheme.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images