A digital-twin-based building energy consumption dynamic optimization system

Abstract

The present application belongs to the technical field of energy consumption optimization, and particularly relates to a digital-twin building energy consumption dynamic optimization system, comprising: a modeling module that acquires building physical, structural and equipment layout parameters, combines three-dimensional modeling and grid segmentation algorithm, and constructs a target building three-dimensional region segmentation model containing geometric transmission constraints; a control mapping module that collects data such as regional passenger flow density and equipment energy consumption, constructs a multi-level correlation control mapping through correlation analysis algorithm, including a first correlation mapping of passenger flow-rate, a second correlation mapping of rate-energy consumption, and a second correlation feedback mapping based on comfort score; a simulation module that constructs a generator and a discriminator with the correlation mapping, and does synchronous adversarial simulation combined with the three-dimensional region segmentation model to solve the optimal energy consumption rate; and a feedback adjustment module that compares actual operation and feedback data, corrects mapping parameters and optimization algorithm weights, realizes building energy consumption dynamic optimization, and ensures that overall energy consumption and comfort score meet the preset threshold.

Description

Technical Field

[0001] This invention belongs to the field of energy consumption optimization technology, and in particular relates to a digital twin-based dynamic optimization system for building energy consumption. Background Technology

[0002] With the popularization of green and intelligent building concepts, building energy management has become a key aspect of building operation, especially since the energy consumption of air conditioning systems in public areas accounts for a high proportion of the total building energy consumption. Currently, existing technologies for controlling air conditioning energy consumption in public areas mostly adopt methods based on fixed schedules, simple sensor feedback, or independent subsystem control. For example, air conditioning systems often start and stop according to preset temperature ranges, while lighting systems rely on timed switching or human body sensing control. Therefore, how to achieve coordinated and adaptive adjustment of air conditioning start and stop temperature thresholds based on real-time pedestrian density, real-time temperature and humidity inside the building, and indoor air conditioning equipment operating status data through dynamic modeling, in order to meet comfort requirements while reducing energy waste at different times, remains an unsolved problem. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a digital twin-based dynamic optimization system for building energy consumption. The system comprises: a modeling module that acquires building physics, structural, and equipment layout parameters, and constructs a 3D regional segmentation model of the target building with geometrical transmission constraints by combining 3D modeling and mesh segmentation algorithms; a control mapping module that collects data such as regional pedestrian density and equipment energy consumption, and constructs multi-level correlation control mappings through correlation analysis algorithms, including a first correlation mapping of pedestrian flow-rate, a second correlation mapping of rate-energy consumption, and a second correlation feedback mapping based on comfort scores; a simulation module that constructs a generator and discriminator using the correlation mappings, performs synchronous adversarial simulations in conjunction with the 3D regional segmentation model, and solves for the optimal energy consumption rate using a dynamic multi-objective evolutionary algorithm; and a feedback adjustment module that compares actual operation with feedback data, corrects mapping parameters and optimizes algorithm weights to achieve dynamic optimization of building energy consumption, ensuring that overall energy consumption and comfort scores meet preset thresholds.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A digital twin-based dynamic optimization system for building energy consumption includes: a modeling module, a control mapping module, and a simulation module;

[0006] The modeling module is used to acquire the physical parameters, structural parameters, and layout parameters of the target building and the target equipment. It then constructs a 3D model of the target building using a 3D modeling algorithm. Based on the distribution of the target equipment in the 3D model, the functional coverage of a single target equipment, and the geometric transmission constraints of the target building, it performs region segmentation using a mesh segmentation algorithm to obtain a 3D region segmentation model of the target building. The geometric transmission constraints include the thermal conductivity coefficient of the building envelope, airflow paths, and regional barrier characteristics.

[0007] The control mapping module is used to collect real-time pedestrian density, energy consumption status of the target equipment, temperature change rate, target temperature range and comfort score within the control area of ​​the target equipment, and construct a multi-level correlation control mapping for each target equipment control area through correlation analysis algorithm;

[0008] The simulation module constructs a generator based on the multi-level associated control mapping of each target device's control area, and constructs a discriminator based on the prediction results of the multi-level associated control mapping of all target device control areas combined with a multi-objective optimization algorithm. Through the generator, discriminator, and simulation algorithm, combined with the target building's three-dimensional area segmentation model, the module performs simulation and outputs the cooperative control strategy of the target devices in real time.

[0009] Specifically, the building energy consumption dynamic optimization system also includes a feedback adjustment module; the feedback adjustment module is used to collect the actual operating data and comfort score feedback data of the target equipment under the collaborative control strategy, compare the actual operating data and comfort score feedback data with preset thresholds, and adjust the parameters of the multi-level associated control mapping and the constraint weights of the multi-objective optimization algorithm according to the comparison results, so that the overall energy consumption and comfort score of the target equipment meet the preset thresholds.

[0010] Specifically, the control mapping module includes a people flow-rate correlation unit and a rate-energy consumption conversion unit;

[0011] The population flow-rate correlation unit is used to obtain the first correlation mapping corresponding to the initial population flow density and cooling rate by combining the different population flow densities and corresponding actual cooling rates collected in each target equipment control area under the preset target comfort temperature range with the population density time distribution function, and by combining the multiple regression algorithm. The population density time distribution function is used to construct the population density time distribution function based on the number of people per unit area in each target equipment control area in each preset time period and the geometric probability distribution.

[0012] The rate-energy conversion unit is used to obtain a second correlation mapping between cooling rate and target energy consumption by combining the covariates constructed based on the target equipment energy consumption rate corresponding to each actual cooling rate with geometric transmission constraints and through correlation analysis algorithms.

[0013] Specifically, the control mapping module also includes a control adjustment unit;

[0014] The control adjustment unit is used to construct an energy consumption adjustment coefficient based on the reciprocal of the comfort score corresponding to the control area of ​​each target device, and map the energy consumption adjustment coefficient to the second correlation mapping to obtain the second correlation feedback mapping;

[0015] The simulation module includes an adversarial prediction unit. The adversarial prediction unit is used to construct a generator based on the first correlation mapping and the second correlation feedback mapping, and combined with the real-time collected population density and real-time temperature of the single target device control area, predict the energy consumption rate and heating rate of the single target device control area at the current moment. At the same time, it performs comfort evaluation on each target device control area using a comprehensive fuzzy algorithm to obtain a comfort evaluation score for each target device control area.

[0016] Specifically, the simulation module includes a discrimination unit;

[0017] The identification unit is used to construct adversarial errors based on the predicted energy consumption rate and corresponding heating rate of all target equipment control areas at the current moment, as well as the deviation between the real-time energy consumption rate and the heating rate. It constructs a multi-objective optimization function to minimize the overall energy consumption of all target equipment control areas and maximize the comfort score. Using geometric transmission constraints, equipment energy consumption constraints, equipment coverage area, and target temperature range as the objective constraint set, it performs synchronous adversarial simulation with the target building's 3D area segmentation model through a simulation algorithm. Simultaneously, a dynamic multi-objective evolutionary algorithm is used to solve for the optimal energy consumption rate adversarial problem for each target equipment control area. The comfort evaluation score and the deviation between the real-time temperature and the upper limit of the target temperature range obtained from each round of simulation are fed back to the generator to correct the first and second association mappings, ensuring that the overall energy consumption and comfort score of the target equipment meet preset thresholds.

[0018] Specifically, the process of constructing the three-dimensional region segmentation model of the target building includes:

[0019] The physical parameters, structural parameters, and target equipment layout parameters of the target building are obtained and then processed hierarchically using an analytical algorithm. The physical parameters are stored hierarchically according to the type of building components, the structural parameters are hierarchically associated according to spatial hierarchy, and the target equipment layout parameters are hierarchically mapped according to the equipment function type, thus obtaining a hierarchical parameter set.

[0020] Based on the hierarchical parameter set, the geometric information and physical parameters of building components are bound together by the BIM and point cloud data fusion algorithm. A three-dimensional model of the target building foundation containing thermal properties is constructed by the three-dimensional mesh generation algorithm, so that each building component in the target building foundation three-dimensional model is associated with the corresponding thermal conductivity coefficient of the building envelope.

[0021] Based on the distribution of target equipment in the 3D model of the target building foundation, the weighted Voronoi diagram algorithm is used for initial mesh segmentation, with the rated power of the equipment as the weight factor. At the same time, the functional coverage range of a single target equipment and the regional barrier characteristics are used as constraints to obtain the initial mesh model of equipment association.

[0022] Specifically, the process of constructing the three-dimensional region segmentation model of the target building also includes:

[0023] Extract the thermal conductivity coefficient of the building envelope from the geometric transport constraints, and use the thermal resistance gradient analysis algorithm to calculate the gradient distribution of the thermal conductivity coefficient in the building space;

[0024] Based on the gradient distribution results, an adaptive mesh refinement algorithm is used to increase the mesh density change rate along the gradient direction of the thermal conductivity coefficient in the initial mesh model, so that the mesh boundary of the region where the difference in thermal conductivity coefficient exceeds the preset difference threshold is consistent with the gradient direction, thus obtaining a thermal resistance adapted mesh model.

[0025] Specifically, the process of constructing the three-dimensional region segmentation model of the target building also includes:

[0026] The airflow path in the geometric transport constraints is extracted and combined with a CFD simulation algorithm to obtain the airflow velocity vector distribution on the path;

[0027] The thermal resistance adaptation mesh model is topologically corrected based on vector distribution combined with streamline tracing algorithm to ensure that the edge direction of the mesh is consistent with the mainstream airflow direction. At the same time, the mesh of the airflow intersection area is divided using polyhedral mesh rules to construct the airflow adaptation mesh model.

[0028] Specifically, the process of constructing the three-dimensional region segmentation model of the target building also includes:

[0029] The regional barrier features in the geometric transmission constraints are extracted, and a boundary recognition algorithm is used to distinguish between fixed barrier components and dynamic barrier components. For fixed barrier components, a boundary fitting algorithm is used to make the mesh boundary and the physical boundary of the component completely coincide. For dynamic barrier components, a virtual boundary algorithm is used to preset the deformable area of ​​the mesh within its movement range to obtain a barrier-adaptive mesh model.

[0030] Based on the thermal resistance adaptive mesh model, airflow adaptive mesh model and barrier adaptive mesh model combined with a multi-objective genetic algorithm, the optimization objective is to use the mesh division accuracy and the total energy consumption corresponding to the control area of ​​all target equipment as the optimization objective, and the thermal conductivity gradient, airflow path and regional barrier boundary as the constraints. The size, shape and topological relationship of the mesh are iteratively adjusted to obtain the three-dimensional regional segmentation model of the target building.

[0031] Specifically, the implementation process of the adaptive mesh densification algorithm includes: calculating the error distribution of the initial mesh in the heat conduction simulation using the finite element error estimation method; marking the region where the error exceeds the preset error threshold as the densification region; using a quadtree or octree subdivision algorithm to divide the densification region into meshes; and maintaining the proportion of shared edge or face lengths of adjacent meshes not exceeding the preset proportion during the division process.

[0032] Compared with the prior art, the beneficial effects of the present invention are:

[0033] This invention addresses the shortcomings of existing technologies by constructing a three-dimensional region segmentation model that accurately reflects the building's thermal environment and equipment coverage characteristics through a modeling module that combines building physics, structural, and equipment layout parameters with geometric transmission constraints. This provides a realistic framework for simulation. The control mapping module constructs a multi-level correlation mapping including pedestrian flow-cooling rate, cooling rate-energy consumption, and comfort adjustment, achieving precise correlation between pedestrian flow, temperature changes, and equipment energy consumption, allowing the control logic to adapt to the dynamic needs of the region. The simulation module uses the correlation mapping to construct a generator and discriminator, and combines a dynamic multi-objective evolutionary algorithm to conduct adversarial simulation, efficiently solving for the optimal energy consumption rate and balancing overall energy consumption and comfort. The feedback adjustment module compares actual data with preset thresholds to correct mapping parameters and algorithm weights, forming a closed-loop optimization that ensures that energy consumption and comfort continuously meet standards during long-term system operation. This significantly improves the accuracy and efficiency of building energy consumption optimization, greatly enhances indoor environmental comfort, and reduces operation and maintenance costs. Attached Figure Description

[0034] Figure 1 This is a block diagram of a digital twin-based building energy consumption dynamic optimization system according to the present invention;

[0035] Figure 2 This is a flowchart illustrating the construction process of the three-dimensional region segmentation model of the target building in this invention. Detailed Implementation

[0036] In large-scale building energy management, achieving synergistic improvement in both dynamic energy consumption optimization and indoor comfort is a key challenge. Traditional control systems, often based on static models, cannot effectively address the impacts of variations in the thermal conductivity of the building envelope, the complexity of airflow paths, and dynamic zoning characteristics. Simultaneously, real-time fluctuations in pedestrian density, changes in equipment energy consumption, and temperature rate differences require systems with rapid response and adaptive adjustment capabilities. Existing methods lack integration of high-precision 3D modeling and multi-level correlated control mapping, leading to inaccurate simulations and limited effectiveness of optimization strategies. Therefore, please refer to [link to relevant documentation]. Figure 1 The present invention provides an embodiment of a digital twin building energy consumption dynamic optimization system, comprising: a modeling module, a control mapping module, a simulation module, and a feedback adjustment module;

[0037] The modeling module is used to acquire the physical parameters, structural parameters, and layout parameters of the target building and the target equipment. It then constructs a 3D model of the target building using a 3D modeling algorithm. Based on the distribution of the target equipment in the 3D model, the functional coverage of a single target equipment, and the geometrical transmission constraints of the target building, a mesh segmentation algorithm is used to segment the area, resulting in a 3D area segmentation model of the target building. Geometrical transmission constraints include the thermal conductivity coefficient of the building envelope, airflow paths, and regional barrier characteristics. It should be further noted that the target equipment in this embodiment includes, but is not limited to, air conditioners.

[0038] It should be further explained that the process of obtaining the thermal conductivity coefficient of the building envelope in this embodiment includes:

[0039] S1. Based on the national standard material thermal property database, obtain the thermal conductivity, specific heat capacity and density parameters of the building envelope material. At the same time, extract the material type, thickness and construction sequence information of the building envelope from the building information model. Combine the thermal performance data measured by the on-site infrared thermal imager and heat flow meter, and use the Kalman filter algorithm to fuse the standard parameters and measured data to obtain a complete material thermal property dataset.

[0040] S2. Based on the structural parameters of building components, the composite enclosure structure is decomposed into single material layers using the hierarchical analysis method. By establishing a bidirectional index table of component codes and material layer codes, a mapping relationship between the thermal properties of components and material layers is formed. Special structural nodes are processed separately using a region growth algorithm to ensure that each material layer has complete thermal property parameter definitions.

[0041] S3. For layered enclosure structures, the overall thermal resistance is calculated using a series thermal resistance model, and the equivalent thermal conductivity coefficient is calculated based on the total thickness of the structure. For heterogeneous structures, the area-weighted average method is used to calculate the zoned thermal resistance, and the overall thermal resistance is synthesized through a parallel thermal resistance model. For cavity structures, the equivalent thermal conductivity coefficient method is used to comprehensively consider the effects of air convection and radiation heat transfer to obtain an accurate equivalent thermal conductivity coefficient.

[0042] S4. Based on temperature, humidity and solar radiation intensity data from meteorological stations at the building's location, a multiple linear regression model is used to establish the correction relationship between environmental parameters and material thermal properties, and to dynamically correct the equivalent thermal conductivity coefficient, including corrections for the influence of temperature on thermal conductivity, the influence of humidity on the performance of thermal insulation materials, and the solar radiation thermal effect.

[0043] S5. The nearest neighbor matching algorithm is used to spatially associate the corrected equivalent thermal conductivity coefficient with the building envelope components in the 3D model. Through component type identification and geometric feature matching, the accurate thermal property parameters of each building envelope surface are ensured, and finally, a 3D model of the target building foundation containing complete thermal parameters is constructed.

[0044] In this embodiment, the regional barrier features include fixed barrier component features and dynamic barrier component features. The fixed barrier component features cover component type, material properties, physical boundary parameters, and structural integrity parameters. The dynamic barrier component features include component type, mobility properties, variable boundary parameters, and state switching parameters. Through feature extraction and definition, a complete parameter basis is provided for constructing a barrier adaptation mesh model.

[0045] It should be further explained that, see [link / reference] Figure 2 The construction process of the target building 3D region segmentation model in this embodiment includes:

[0046] A1. Obtain the physical parameters, structural parameters, and target equipment layout parameters of the target building. Combine the analytical algorithm to process the parameters in layers. Physical parameters are stored in layers according to building component type, structural parameters are associated in layers according to spatial hierarchy, and target equipment layout parameters are mapped in layers according to equipment function type to obtain a layered parameter set. In this embodiment, the physical parameters, structural parameters, and target equipment layout parameters of the target building include: physical parameters, which refer to the inherent properties of the materials constituting the building envelope and each component, mainly including thermal conductivity, density, specific heat capacity, and solar radiation absorptivity; structural parameters, which refer to the entity information defining the division and geometric shape of the building space, mainly including the geometric dimensions, spatial coordinates, topological relationships, and functional zoning of floors and rooms of components; and target equipment layout parameters, which refer to the spatial configuration and performance data of energy consumption systems such as HVAC and lighting, mainly including equipment type, installation location, rated power, operating conditions, and service area coverage.

[0047] A2. Based on a layered parameter set, a BIM and point cloud data fusion algorithm is used to bind the geometric information and physical parameters of building components. A 3D mesh generation algorithm is used to construct a 3D model of the target building foundation containing thermal properties, so that each building component in the 3D model of the target building foundation is associated with the corresponding thermal conductivity coefficient of the building envelope. It should be further noted that in this embodiment, the geometric information of building components in the layered parameter set is imported into the BIM platform to generate an initial BIM model. A model alignment algorithm is used to spatially align the global point cloud model with the initial BIM model, so that the component outline in the BIM model coincides with the component surface point cloud in the point cloud model. It should be further noted that in this embodiment, the physical parameters of building components are extracted from the layered parameter set, including the thermal conductivity coefficient of the building envelope and the specific heat capacity of the material. A hash table-based key-value mapping algorithm is used to associate and bind the physical parameters with the corresponding component objects in the BIM model.

[0048] The fused BIM-point cloud 3D model is meshed using the Delaunay triangulation algorithm. During the meshing process, a feature-preserving mesh optimization method is used to ensure the consistency between the mesh boundary and the geometric boundary of the building components.

[0049] During the mesh generation process, an attribute inheritance mechanism based on spatial location query is adopted to establish a relationship between each mesh cell and the physical parameters of its building components, including thermal conductivity and specific heat capacity parameters, and finally construct a three-dimensional model of the target building foundation containing complete thermal properties.

[0050] Specifically, the feature-preserving mesh optimization method includes: adopting a curvature-adaptive mesh subdivision strategy to increase mesh density in areas with significant geometric features, while simultaneously using a Laplacian mesh smoothing algorithm to optimize mesh quality;

[0051] The attribute inheritance mechanism specifically includes: determining the building component to which the grid cell belongs through a ray casting spatial positioning algorithm, and then assigning the corresponding thermal property parameters to the grid cell by searching a pre-established component-physical parameter mapping table.

[0052] Based on the distribution of target equipment in the 3D model of the target building foundation, the weighted Voronoi diagram algorithm is used for initial mesh segmentation, with the rated power of the equipment as the weight factor. At the same time, the functional coverage range of a single target equipment and the regional barrier characteristics are used as constraints to obtain the initial mesh model of equipment association.

[0053] A3. Extract the thermal conductivity coefficient of the building envelope from the geometrical transport constraints, and use the thermal resistance gradient analysis algorithm to calculate the gradient distribution of the thermal conductivity coefficient in the building space; it should be further noted that in this embodiment, the thermal conductivity coefficient of each building envelope of the target building is extracted from the geometrical transport constraints, and a correspondence table of building envelope and thermal conductivity coefficient is established.

[0054] A4. Based on the gradient distribution results, an adaptive mesh refinement algorithm is used to increase the mesh density change rate along the gradient direction of the thermal conductivity coefficient in the initial mesh model, so that the mesh boundary of the region where the difference in thermal conductivity coefficient exceeds the preset difference threshold is consistent with the gradient direction, thus obtaining a thermal resistance adapted mesh model.

[0055] It should be further explained that the construction process of the thermal resistance adaptation mesh model in this embodiment includes:

[0056] A4.1 Based on the spatial relationship between grid cells and enclosure structures in the three-dimensional model of the target building foundation, assign the thermal conductivity coefficient of each enclosure structure in the preset enclosure structure-thermal conductivity coefficient correspondence table to the corresponding grid cell to obtain the thermal conductivity coefficient value of each grid cell.

[0057] A4.2 Using the thermal conductivity coefficient value of each grid cell as input, a gradient calculation algorithm based on the central difference method is adopted. By calculating the difference in thermal conductivity coefficient between adjacent grid cells, the gradient vector of each grid cell is obtained, and all gradient vectors are arranged in spatial coordinate order to generate the thermal conductivity coefficient gradient distribution in the building space.

[0058] A4.3. Based on the aforementioned gradient distribution of the thermal conductivity coefficient, an adaptive mesh refinement method based on posterior error estimation is used to refine the initial mesh model, specifically including:

[0059] A4.3.1 Based on the spatial distribution data of the thermal conductivity coefficient, the gradient vector field is calculated using the Aobel gradient operator. The gradient vector field has a local gradient direction and gradient magnitude at any point in space.

[0060] A4.3.2 Based on a preset difference threshold, a region growing algorithm is used to identify continuous spatial regions where the gradient magnitude exceeds the threshold, and these regions are defined as high gradient change regions.

[0061] A4.3.3 For each of the high gradient change regions, a corresponding mesh densification strategy is generated based on a quadtree or octree subdivision method, wherein the local gradient direction in the region is the dominant densification direction, and a mesh density change rate higher than that in other directions is set in the dominant densification direction.

[0062] A4.3.4 Execute the encryption strategy, use the Delaunay triangulation algorithm to reconstruct and optimize the initial mesh model in the high gradient change region, and adjust the geometric orientation of the new mesh boundary using the Laplace smoothing algorithm;

[0063] A4.3.5 Iteratively execute A4.3.3 to A4.3.4 until the geometry of all grid cells in the high gradient change region meets the preset orientation accuracy requirements, and output the thermal resistance adapted grid model;

[0064] A5. Extract the airflow path from the geometric transport constraints and combine it with a CFD simulation algorithm to obtain the airflow velocity vector distribution on the path;

[0065] A6. Based on vector distribution and streamline tracing algorithm, the thermal resistance adaptation mesh model is topologically corrected so that the edge direction of the mesh is consistent with the mainstream airflow direction. At the same time, the mesh of the airflow intersection area is divided using polyhedral mesh rules to construct the airflow adaptation mesh model.

[0066] It should be further explained that the construction process of the airflow adaptation mesh model in this embodiment includes:

[0067] Based on the target building's basic 3D model, an edge detection algorithm is used to extract the spatial coordinates and cross-sectional dimensions of all airflow paths in the geometric transmission constraints. The airflow paths include fresh air ducts, return air channels, and door and window opening gaps. Based on the extraction results, a non-uniform rational B-spline curve reconstruction algorithm is used to construct a 3D geometric model of the airflow paths.

[0068] Boundary conditions based on the Realizable k-ε turbulence model are set for the three-dimensional geometric model, wherein the airflow inlet is set as a velocity inlet condition and the velocity magnitude and direction are specified, the outlet is set as a pressure outlet condition and a static pressure value is given, and the thermal conductivity coefficient of the building envelope is used as a thermal environment parameter input into the computational domain.

[0069] The computational fluid dynamics simulation algorithm based on the finite volume method is used to discretize and solve the three-dimensional geometric model, calculate the magnitude and direction of the airflow velocity at each spatial point in the airflow path, record the airflow velocity vector data according to the coordinates of the spatial grid nodes, and obtain the airflow velocity vector distribution on the path.

[0070] Based on the airflow velocity vector distribution, a streamline distribution map is generated using a fourth-order Runge-Kutta streamline tracing algorithm. The thermal resistance adaptation mesh model is then topologically corrected, and the mesh edge direction is adjusted using a mesh reconstruction algorithm to ensure that it is consistent with the mainstream airflow direction.

[0071] For the airflow convergence region, the Voronoi diagram partitioning method based on Euclidean distance is used to repartition the grid. While maintaining the grid quality, the consistency between the grid edges and the airflow direction is ensured, and finally an airflow-adaptive grid model that matches the airflow distribution characteristics is obtained.

[0072] Specifically, the mesh reconstruction algorithm includes: adopting a mesh deformation method based on spring theory, treating the mesh edges as a spring system, and iteratively optimizing with the airflow direction as the target direction until the average angle between the mesh edge direction and the mainstream airflow direction is less than a preset threshold.

[0073] A7. Extract the regional barrier features from the geometric transmission constraints, and use a boundary recognition algorithm to distinguish between fixed barrier components and dynamic barrier components. For fixed barrier components, a boundary fitting algorithm is used to make the mesh boundary and the physical boundary of the component completely coincide. For dynamic barrier components, a virtual boundary algorithm is used to preset the deformable area of ​​the mesh within its movement range to obtain a barrier-adaptive mesh model.

[0074] A7.1 Based on the aforementioned geometric transmission constraints, an edge detection algorithm is used to extract the barrier features of all areas within the building space, specifically including the following steps:

[0075] A7.1.1 Perform Gaussian filtering on the 3D building model and smooth the image using a Gaussian kernel with a preset standard deviation;

[0076] A7.1.2 Calculate the gradient magnitude and direction of an image using gradient operators;

[0077] A7.1.3. Employ a non-maximum suppression algorithm to refine the gradient magnitude and retain local maximum points;

[0078] A7.1.4. Apply a dual-threshold detection algorithm, and determine the final edge features by setting high and low thresholds and performing hysteresis threshold processing;

[0079] A7.2. Based on the extracted regional barrier features, a boundary recognition algorithm based on morphological analysis is used to distinguish between fixed barrier components and dynamic barrier components, specifically including:

[0080] A7.2.1. The extracted edge image is processed using morphological opening operations, and structuring elements of a specific shape are used to eliminate fine noise and smooth the boundaries.

[0081] A7.2.2 Based on edge curvature analysis, calculate the average curvature of the boundary curve, and determine the component type according to the curvature magnitude threshold;

[0082] A7.2.3 For fixed barrier components, a constrained triangulation algorithm is used to generate triangular meshes that satisfy the boundary constraints on the boundary vertex set, ensuring that the mesh boundary completely coincides with the physical boundary of the component.

[0083] A7.2.4 For dynamic barrier components, a mesh deformation algorithm based on spring simulation is adopted, the mesh edges are regarded as a spring system, a reasonable spring stiffness coefficient is set, and the mesh deformable area is preset within the component movement range.

[0084] A7.3. Based on the above processing results, the mesh in the blocked area is smoothed using a mesh quality optimization algorithm:

[0085] A7.3.1. The Laplace smoothing algorithm is used to iteratively optimize the internal vertex positions and update the vertex positions using a preset smoothing coefficient.

[0086] A7.3.2. Set up a grid quality evaluation index system, including the aspect ratio threshold, minimum interior angle threshold, and maximum interior angle threshold of grid cells;

[0087] A7.3.3 Through multiple iterations of optimization, all grid cells meet the preset quality index requirements, and finally a barrier-adapted grid model that is precisely adapted to the building barrier characteristics is obtained.

[0088] A8. Based on the thermal resistance adaptive mesh model, airflow adaptive mesh model and barrier adaptive mesh model combined with a multi-objective genetic algorithm, the optimization objective is to use the mesh division accuracy and the total energy consumption corresponding to the control area of ​​all target equipment as the optimization objective, and the thermal conductivity gradient, air flow path and regional barrier boundary as constraints. The size, shape and topological relationship of the mesh are iteratively adjusted to obtain the three-dimensional regional segmentation model of the target building.

[0089] The control mapping module is used to collect real-time pedestrian density, energy consumption status of the target equipment, temperature change rate, target temperature range and comfort score within the control area of ​​the target equipment, and construct a multi-level correlation control mapping for each target equipment control area through correlation analysis algorithm;

[0090] The simulation module constructs a generator based on the multi-level associated control mapping of each target device control area, and constructs a discriminator based on the prediction results of the multi-level associated control mapping of all target device control areas, combined with a multi-objective optimization algorithm. Through the generator, discriminator and simulation algorithm, combined with the three-dimensional area segmentation model of the target building, the module performs simulation and outputs the cooperative control strategy of the target device in real time.

[0091] The feedback adjustment module is used to collect the actual operating data and comfort score feedback data of the target equipment under the collaborative control strategy, compare the actual operating data and comfort score feedback data with preset thresholds, and adjust the parameters of the multi-level associated control mapping and the constraint weights of the multi-objective optimization algorithm according to the comparison results, so that the overall energy consumption and comfort score of the target equipment meet the preset thresholds.

[0092] It should be further noted that the control mapping module in this embodiment includes a people flow-rate correlation unit, a rate-energy consumption conversion unit, and a control adjustment unit;

[0093] It should be further explained that the people flow-rate correlation unit in this embodiment is used to obtain the first correlation mapping corresponding to the initial people flow density and cooling rate by combining the different people flow densities and corresponding actual cooling rates and population density time distribution functions collected in each target equipment control area under the preset target comfort temperature range, and by combining the multiple regression algorithm; the population density time distribution function is used to construct the population per unit area in each target equipment control area in each preset time period by combining the geometric probability distribution.

[0094] It should be further explained that the specific implementation process of constructing the first association mapping in this embodiment includes:

[0095] Step 1: Based on the thermal resistance adaptation mesh characteristics of the target building's 3D region segmentation model, specifically the thermal conductivity gradient, the mesh is divided into multiple thermal comfort zones using a clustering algorithm. Then, based on the building envelope type and functional scene label of each thermal comfort zone, combined with the PMV-PPD comfort model, the comfort temperature range is calculated using a weighted average algorithm to obtain a scene-based target comfort temperature range set.

[0096] Step 2: Based on the grid node layout of the target building's 3D area segmentation model, the nearest neighbor matching algorithm is used to bind the pedestrian density sensor and temperature sensor to the grid node one by one; then, based on the density level of the thermal resistance-adaptive grid, a time series sampling algorithm is used to set a differentiated data acquisition frequency to obtain a sample dataset of pedestrian density and cooling rate with grid identifiers.

[0097] Step 3: Based on the historical population per unit area of ​​each grid, generate an initial population density time distribution function using a Gaussian mixture model; then, based on the functional scene type of each grid, including office, meeting, and rest types, obtain scene weights using the analytic hierarchy process (AHP); simultaneously, based on the deformable grid region of the dynamic barrier components, obtain the barrier influence coefficient using the boundary element analysis method; superimpose the scene weights and barrier influence coefficients, and obtain the population density time distribution function adapted to the dynamic scene using the least squares fitting method.

[0098] Step 4: Based on the sample dataset of population density and cooling rate with grid labels and the thermal conductivity coefficient gradient of the thermal resistance adaptation grid, regression weights are assigned to samples with different thermal conductivity coefficient gradients using the inverse distance weighting method; combined with the population density time distribution function of dynamic scene adaptation as a regression variable, a preliminary first correlation mapping is obtained by fitting the data using the ridge regression algorithm; then, based on the consistency between the edge direction of the airflow adaptation grid and the mainstream airflow direction, the correlation coefficient of the preliminary first correlation mapping is adjusted by the principal component analysis method to obtain the first correlation mapping of dual adaptation of grid gradient and airflow characteristics.

[0099] Step 5: Based on the first correlation mapping of dual adaptation of grid gradient and airflow characteristics, and the real-time grid state of the target building's 3D region segmentation model, including the grid deformation state of dynamic barrier components and the real-time airflow velocity vector of the airflow-adapted grid, the Kalman filter algorithm is used to collect the population density deviation and cooling rate deviation data of each grid. Based on the deviation data, the causal inference algorithm is used to analyze the source of the deviation, which includes changes in the barrier boundary and changes in airflow. Then, for the source of the deviation, the correlation mapping parameters of the corresponding grid are adjusted using the gradient descent method to obtain the real-time optimized first correlation mapping. It should be further noted that the rate-energy consumption conversion unit in this embodiment is used to obtain the second correlation mapping between the cooling rate and the target energy consumption by combining the covariate constructed by the target equipment energy consumption rate corresponding to each actual cooling rate with geometric transmission constraints through a correlation analysis algorithm.

[0100] It should be further explained that the process of obtaining the second association mapping in this embodiment includes:

[0101] Step 1: Based on the thermal conductivity coefficient gradient of each grid unit in the 3D building model, a clustering algorithm is first used to clearly divide all grids into three levels: high, medium, and low, according to the gradient magnitude. Secondly, based on this division, a correlation analysis algorithm between thermal resistance and energy consumption is used to accurately calculate the thermal resistance influence coefficient corresponding to each level of grid region. Finally, this coefficient is determined as the core variable and integrated to construct fundamental covariates reflecting the thermal characteristics of the building foundation.

[0102] Step 2: Based on the geometric edge direction and spatial node coordinates of the airflow mesh in the 3D building model, a spatial coordinate matching algorithm is used to map the airflow velocity vector obtained from the computational fluid dynamics simulation to the corresponding airflow mesh element. On this basis, by analyzing the offsetting effect of airflow on cooling energy consumption, the airflow offset coefficient of each mesh element is calculated. This coefficient is defined as an adjustment variable, and through a variable coupling algorithm, it is superimposed on the basic covariate obtained in Step 1 to obtain a two-factor coupled covariate that simultaneously considers thermal resistance and airflow effects.

[0103] Step 3: Based on the deformation characteristics of the grid area where the dynamic barrier components are located in the 3D building model, the specific state of these components, including closed, half-open, and fully open, is continuously obtained through a real-time boundary state monitoring algorithm. Based on this real-time state data, a special algorithm for the energy consumption impact of barrier components is used to calculate the barrier impact coefficient of adjacent grid areas. This coefficient is defined as a dynamic variable and superimposed on the two-factor coupled covariate obtained in Step 2 to finally generate a dynamic covariate that can dynamically reflect the combined impact of thermal resistance, airflow, and barrier components.

[0104] Step 4: Based on the density distribution characteristics of the thermal resistance grid in the 3D building model, firstly, a weighting algorithm is used to assign different computational weights to the actual cooling rate and equipment energy consumption rate sample data collected from grid areas of different densities, following the principle of assigning high weights to high-density grid samples; secondly, the dynamic covariates obtained in Step 3 are used as core adjustment factors, and a weighted regression algorithm is used to fit all weighted sample data; simultaneously, a feature clustering algorithm is used to classify grid samples with similar thermal resistance gradients and airflow characteristics into several subsets, and the quantitative correlation coefficient between cooling rate and energy consumption rate within each subset is calculated; finally, a coefficient fusion algorithm is used to integrate the correlation coefficients of each subset into an energy consumption correlation mapping model based on the initially assigned grid weights.

[0105] Step 5: Based on the real-time operating data of the grid continuously acquired from the 3D building model, including the measured values ​​of the thermal conductivity coefficient of the thermal resistance grid, the real-time wind speed and direction of the airflow grid, and the instantaneous position of the dynamic barrier components, a deviation detection algorithm is used to continuously collect deviation data between the model's predicted energy consumption rate and the actual energy consumption rate for each grid cell. Based on the deviation data, a source analysis algorithm is used to accurately determine the source of the deviation. For different identified sources, such as fluctuations in thermal parameters, changes in airflow organization, or changes in component position, the corresponding parameter correction algorithm is activated to dynamically adjust and calibrate specific parameters in the energy consumption correlation mapping model, thereby ensuring that the model can continuously self-optimize and maintain prediction accuracy.

[0106] It should be further explained that the control adjustment unit in this embodiment is used to construct an energy consumption adjustment coefficient based on the reciprocal of the comfort score corresponding to the control area of ​​each target device, and to map the energy consumption adjustment coefficient to the second association mapping to obtain the second association feedback mapping;

[0107] It should be further explained that the simulation module in this embodiment includes an adversarial prediction unit and an identification unit. The adversarial prediction unit is used to construct a generator based on the first correlation mapping and the second correlation feedback mapping, and combine the real-time collected crowd density and real-time temperature of the single target device control area to predict the energy consumption rate and heating rate of the single target device control area at the current moment. At the same time, it performs comfort evaluation on each target device control area using a comprehensive fuzzy algorithm to obtain a comfort evaluation score for each target device control area.

[0108] It should be further explained that the workflow of the adversarial prediction unit in this embodiment includes:

[0109] B1. Based on the feedback parameters of the first and second association mappings, and combined with the thermal conductivity coefficient gradient of the thermal resistance adaptation mesh, the edge streamline direction of the airflow adaptation mesh, and the dynamic boundary state of the barrier adaptation mesh in the three-dimensional region segmentation model of the target building, a parameter binding algorithm is used to associate the association coefficients of the two types of mappings with the spatial coordinates and attribute labels of the corresponding meshes to construct a mesh anchoring generator.

[0110] B2. Based on the real-time acquisition of crowd density data and temperature data of a single target device control area, a spatial interpolation algorithm is used to match the crowd density data to the functional coverage area of ​​the initial grid associated with the device, and at the same time, the temperature data is associated with the high-density monitoring nodes of the thermal resistance adaptation grid to form a grid-level real-time input data matrix.

[0111] B3. Based on the grid anchoring generator and the grid-level real-time input data matrix, the weighted least squares method is used to perform grid weighted prediction, wherein the density value of the thermal resistance-adaptive grid is used as the weight factor. The cooling rate prediction value of each grid is calculated through the first correlation mapping, and the energy consumption rate prediction value is calculated through the second correlation mapping. Finally, the energy consumption rate prediction values ​​of all grids in the target equipment control area are accumulated to obtain the energy consumption rate prediction result of the area.

[0112] B4. Based on the real-time temperature data and grid energy consumption rate prediction values ​​in the grid-level real-time input data matrix, and combined with the thermal conductivity coefficient of the thermal resistance-adapted grid and the airflow velocity vector of the airflow-adapted grid, dynamic calculation is performed using the heat balance equation. The thermal resistance correction coefficient calculated based on the thermal conductivity coefficient gradient and the airflow regulation coefficient calculated based on the airflow velocity vector modulus are introduced to correct the basic heating rate calculation formula, obtain the heating rate prediction value of each grid, and summarize them to form the heating rate prediction result of the single target equipment control area.

[0113] B5. Based on the fuzzy comprehensive evaluation method, a three-dimensional evaluation system including thermal comfort, airflow comfort and humidity comfort is established. The thermal comfort dimension is associated with the temperature fluctuation value of the thermal resistance adaptation grid, the airflow comfort dimension is associated with the edge-stream direction deviation value of the airflow adaptation grid, and the humidity comfort dimension is associated with the boundary sealing parameter of the barrier adaptation grid. A grid feature association evaluation system is constructed.

[0114] B6. Based on the grid feature association evaluation system, the thermal comfort dimension uses the root mean square error algorithm to calculate the temperature deviation rate, the airflow comfort dimension uses the vector angle cosine algorithm to calculate the directional consistency, and the humidity comfort dimension uses the permeability calculation model to evaluate the boundary humidity exchange rate, thereby obtaining the basic evaluation values ​​of the three dimensions respectively.

[0115] B7. Based on the grid density level of the target building's three-dimensional region segmentation model, the analytic hierarchy process is used to assign weights to the basic evaluation values ​​of grids with different densities, with higher weights assigned to high-density grids, and grid adaptation fuzzy inference rules are established.

[0116] B8. Based on the basic evaluation values ​​of the three dimensions and the grid-adaptive fuzzy inference rules, a fuzzy comprehensive evaluation algorithm is used to calculate the preliminary comfort score, and the score is corrected by the dynamic barrier influence factor to finally obtain the comfort evaluation score of each target equipment control area.

[0117] The identification unit is used to construct adversarial errors based on the predicted energy consumption rate and corresponding heating rate of all target equipment control areas at the current moment, as well as the deviation between the real-time energy consumption rate and the heating rate. It constructs a multi-objective optimization function to minimize the overall energy consumption of all target equipment control areas and maximize the comfort score. Using geometric transmission constraints, equipment energy consumption constraints, equipment coverage area, and target temperature range as the objective constraint set, it performs synchronous adversarial simulation with the target building's 3D area segmentation model through a simulation algorithm. Simultaneously, a dynamic multi-objective evolutionary algorithm is used to solve for the optimal energy consumption rate adversarial problem for each target equipment control area. The comfort evaluation score and the deviation between the real-time temperature and the upper limit of the target temperature range obtained from each round of simulation are fed back to the generator to correct the first and second association mappings, ensuring that the overall energy consumption and comfort score of the target equipment meet preset thresholds.

[0118] It should be further explained that the process of constructing the adversarial error in this embodiment includes:

[0119] C1. Based on the predicted energy consumption rate and heating rate of all target equipment control areas at the current moment, as well as the real-time energy consumption rate and real-time heating rate measurements collected by sensors, calculate the energy consumption rate deviation and heating rate deviation of each area; wherein, the boundary of the target equipment control area is based on the fixed barrier component boundary and dynamic barrier component boundary of the barrier adaptation mesh in the three-dimensional area segmentation model of the target building.

[0120] C2. Based on the mesh topology of the target building's three-dimensional region segmentation model, the density zoning features of the thermal resistance adaptation mesh, the edge streamline direction features of the airflow adaptation mesh, and the equipment coverage domain boundary features of the initial mesh associated with equipment are extracted. A spatial autocorrelation analysis method is used to construct a regional adjacency matrix, which is used to describe the spatial adjacency relationship of each control area.

[0121] C3. Calculate the spatial coordination coefficient of the error between adjacent regions through spatial autocorrelation analysis. The weighting of the coordination coefficient is positively correlated with the length of the shared edge or the area of ​​the shared surface of the grid between adjacent regions. The longer the length of the shared edge or the larger the area of ​​the shared surface, the higher the corresponding coordination coefficient. Integrate the spatial coordination coefficient as a weighting factor into the error calculation of the corresponding region to obtain the spatial correlation error.

[0122] C4. An autoregressive integral moving average model is used to perform time series analysis on historical error data, extract the time series accumulation characteristics of errors, and generate time series accumulation weights, which reflect the evolution of errors in the time dimension.

[0123] C5. Based on the mesh size distribution characteristics optimized by the multi-objective genetic algorithm in the three-dimensional region segmentation model of the target building, adjust the calculation coefficient of the time-series cumulative weight, and combine the adjusted weight with the current region error to obtain the time-series cumulative error.

[0124] C6. The entropy weight method is used to weight and fuse spatial correlation error, temporal cumulative error and original deviation. During the fusion process, the polyhedral mesh characteristics of the airflow convergence area of ​​the airflow adaptation mesh are referenced to construct a multi-dimensional adversarial error that simultaneously reflects spatial correlation and temporal cumulative effects. The spatial distribution characteristics of the adversarial error adaptation mesh are also considered.

[0125] It should be further explained that the construction process of the multi-objective optimization function in this embodiment includes:

[0126] E1. Obtain the functional priority of each target equipment control area in the target building, wherein the functional priority is positively correlated with the weighted Voronoi diagram weight of the initial mesh associated with the equipment in the three-dimensional region segmentation model of the target building, and the weighted Voronoi diagram weight is determined by the rated power of the equipment.

[0127] E2. The functional priorities are quantified using the analytic hierarchy process to determine the energy consumption weight of each region and establish a positive correlation between the rated power of the equipment and the energy consumption weight.

[0128] E3. Decompose the comfort score into three evaluation dimensions: thermal comfort, airflow comfort, and humidity comfort, and collect historical evaluation data for each dimension; among them, the airflow comfort evaluation data is correlated with the edge-stream direction consistency of the airflow adaptation grid in the three-dimensional area segmentation model of the target building;

[0129] E4. Based on the historical data of the three evaluation dimensions, the weight coefficient of each dimension is calculated using the entropy weight method;

[0130] E5. Establish a multi-objective optimization model with minimizing overall energy consumption as the first optimization objective and maximizing comfort score as the second optimization objective. Substitute the energy consumption weights of each region into the overall energy consumption objective function and the weights of each comfort dimension into the comfort score objective function.

[0131] E6. A hierarchical multi-objective optimization function is constructed using the goal programming method, which reflects the differences in regional functional priority, the differences in comfort dimension weight, and the spatial attribute correlation characteristics of the three-dimensional grid during the optimization process.

[0132] It should be further explained that the target constraint set adaptation process in this embodiment includes:

[0133] F1. Real-time acquisition of the operational health status parameters of the target equipment, and simultaneous acquisition of heat flow data of the shared enclosure structure between the control areas of each target equipment, and calculation of the thermal coupling strength between the areas based on the heat flow data; wherein, the identification of the shared enclosure structure is based on the boundary of the fixed barrier component of the barrier adaptation mesh in the three-dimensional area segmentation model of the target building, and the acquisition points of the heat flow data are set at the mesh nodes that are in contact with the boundary of the fixed barrier component.

[0134] F2. The constraint priority sorting algorithm is used to evaluate the priority of the target constraint set. When the health status parameters of the equipment exceed the preset normal range, the priority of the corresponding equipment energy consumption constraint is increased, and the priority adjustment range is positively correlated with the initial mesh coverage range of the equipment in the target building three-dimensional region segmentation model.

[0135] F3. When the thermal coupling strength between regions is higher than the preset strength threshold, the spatial association conditions in the geometric transmission constraint are strengthened, and the degree of strengthening is positively correlated with the length of the grid boundary of the fixed barrier component shared by adjacent regions.

[0136] F4. An adaptive parameter adjustment algorithm is adopted to adjust the threshold range of equipment energy consumption constraints according to the degree of deviation of equipment health status parameters. At the same time, the gradient distribution of thermal conductivity coefficient of thermal resistance adaptation mesh in the three-dimensional region segmentation model of the target building is referenced. The higher the gradient value of thermal conductivity coefficient, the higher the adjustment accuracy of equipment energy consumption constraint threshold.

[0137] F5. Adjust the thermal conductivity coefficient correlation parameter in the geometric transmission constraint according to the inter-regional thermal coupling strength, and establish a linkage relationship between the adjustment of this parameter and the airflow velocity vector distribution of the airflow adaptation grid. The higher the airflow velocity value, the greater the correction range of the thermal conductivity coefficient correlation parameter, so that the target constraint set can be dynamically adapted to the equipment operating status, regional coupling relationship and three-dimensional grid spatial characteristics.

[0138] It should be further explained that the twin synchronization simulation in this embodiment is as follows:

[0139] Based on the 3D region segmentation model of the target building, a spatial coordinate transformation algorithm is used to establish an interaction channel between the physical space and the virtual model. The coordinates of sensor monitoring points in the physical space and the spatial coordinates of virtual grid nodes are mapped one-to-one using a kd-tree nearest neighbor search algorithm, enabling real-time synchronous updates of physical measurement parameters and virtual model parameters. In this embodiment, the kd-tree nearest neighbor search algorithm is used to establish a precise mapping between physical sensors and the virtual grid. The specific execution process is as follows: First, the 3D spatial coordinates of all virtual grid nodes in the 3D region segmentation model of the target building are used as input data to construct a kd-tree index structure, where the segmentation dimension... The selection of degree is based on the principle of maximizing the variance of coordinate values ​​in each dimension to ensure the balance of the tree structure. When a new physical sensor detects data, the system uses the three-dimensional coordinates of the sensor as the query point, recursively traverses from the root node of the kd tree, calculates the Euclidean distance between the query point and each segmentation hyperplane, and quickly eliminates subtrees that do not contain the nearest neighbor (i.e., virtual grid regions that are unrelated to this sensor). After traversing to the leaf node, the current nearest point is recorded, and backtracking is performed to check whether there are any closer nodes on the other side of other segmentation hyperplanes. Finally, the virtual grid node that best matches the spatial position of the physical sensor is output, completing the accurate injection of physical data into the virtual model.

[0140] Based on the mesh functional attributes of the aforementioned 3D region segmentation model, a dedicated computational sub-model is deployed using an MPI-based distributed simulation architecture: the building heat conduction sub-model is deployed in the thermal resistance-adaptive mesh region, the equipment operation sub-model is deployed in the equipment-associated initial mesh region, and the pedestrian flow dynamic sub-model is deployed in the pedestrian activity area of ​​the barrier-adaptive mesh. Data exchange between the sub-models is conducted using the HDF5 standard data format, which includes mesh node coordinates, thermal property parameters, and timestamp information, thus constructing a multiphysics coupled simulation system. In this embodiment, the MPI distributed simulation architecture is used to achieve parallel computation of multiphysics simulation. Specifically, the mesh functional attribute features of the target building's 3D region segmentation model are first extracted, and the entire computational domain is decomposed into multiple sub-regions accordingly. For example, the thermal resistance-adaptive mesh region is assigned to one group of processes running the building heat conduction sub-model, and the equipment-associated initial mesh region is assigned to another group of processes running the equipment operation sub-model. When each process calculates the internal state of its sub-region, it extracts the state features (such as boundary temperature and heat flux) of the sub-region boundary mesh in real time and sends these feature data to adjacent processes through the MPI message passing interface. At the same time, it also receives boundary feature data from adjacent processes to achieve data synchronization and coupled calculation between sub-models, thereby efficiently and collaboratively completing the dynamic simulation of the entire building.

[0141] Based on real-time acquired temperature data streams, a sliding window variance calculation method is used to calculate the rate of change of the thermal environment in the control area of ​​each target device. Monitoring points are preferentially deployed in regions where the gradient value of the thermal conductivity coefficient in the thermal resistance adaptation grid is higher than the threshold K1, and in polyhedral grid regions where the magnitude of the airflow velocity vector is higher than the threshold K2 in the airflow adaptation grid. The sliding window variance calculation method in this embodiment is used to quantify the thermal dynamic characteristics of the control area of ​​each target device in real time. The specific implementation process is as follows: For each virtual grid node bound to a sensor, the system continuously receives its temperature data stream and maintains a time window of fixed length L (e.g., 10 sampling periods). As new data arrives, the window slides forward, removing the oldest data points. At each calculation moment, the average value of all temperature data within the current window is calculated, and then the sum of squares of the differences between each data point and the mean is calculated. Finally, this sum is divided by the window length L to obtain the temperature variance of that time window. This variance value is output in real time as a direct measure of the rate of change of the thermal environment. When this value exceeds a preset threshold R1, a subsequent adaptive simulation step size adjustment mechanism is triggered to achieve refined simulation of areas with drastic thermal dynamics.

[0142] The variable-step-size Runge-Kutta algorithm dynamically adjusts the simulation calculation step size. Specifically, when the rate of change of the thermal environment in a certain region is higher than a preset threshold R1, the simulation step size is shortened proportionally according to the grid density value D1 of that region. The adjustment formula for the step size ΔT is ΔT=T0×(1-α×D1), where T0 is the baseline step size and α is the step size adjustment coefficient. When the rate of change of the thermal environment is lower than a preset threshold R2, the simulation step size is extended proportionally according to the grid functional priority P1. The adjustment formula for the step size is ΔT=T0×(1+β / P1), where β is the priority weight coefficient. In this embodiment, the variable-step-size Runge-Kutta algorithm is used to solve the building thermodynamic equations and realize adaptive control of the simulation step size. Its core is dynamic adjustment based on error estimation: In each simulation step, the algorithm first uses the current step size T0 to predict the temperature state of a specific grid region, and simultaneously uses embedded 4th and 5th order formulas to calculate the solution respectively. The difference between the two results is used to estimate the local truncation error. This error is used as a key control feature. If the error characteristic value exceeds the preset tolerance, it is determined that the current region's state has changed drastically. The algorithm extracts features with excessively large step sizes and immediately reduces the step size for recalculation. If the error is much smaller than the tolerance, features with excessive computational resources are extracted, and the step size is appropriately increased in the next step. This process enables the simulation to automatically use small step sizes in high-gradient areas such as windows and exterior walls to ensure accuracy, and large step sizes in flat areas such as the indoor core area to improve efficiency.

[0143] Through the aforementioned adaptive step size adjustment mechanism, a fine step size is used in regions with drastic changes in the thermal environment to ensure computational accuracy, while a loose step size is used in regions with gradual changes to improve computational efficiency, thereby achieving a dynamic balance between simulation accuracy and computational efficiency.

[0144] It should be further explained that the process of solving the optimal energy consumption rate adversarial problem for each target device control area in this embodiment includes:

[0145] Based on a dynamic multi-objective evolutionary algorithm, for a single target device control area, the boundary of this area is limited by the boundary of the barrier fitting mesh in the 3D region segmentation model of the target building; a non-dominated sorting genetic algorithm is used to perform local optimization within the area, with the energy consumption rate and heating rate of the area as optimization variables, and the initial search range of the optimization variables is positively correlated with the gradient of the thermal conductivity coefficient of the thermal resistance fitting mesh corresponding to the area; the sub-objectives of the corresponding area in the hierarchical multi-objective optimization function are used as the optimization direction to generate the Pareto optimal solution set within the area;

[0146] For adjacent target equipment control areas, the adjacency relationship is determined based on the shared edges / faces of the meshes in the 3D region segmentation model of the target building (the existence of shared edges or faces indicates adjacency). A co-evolutionary algorithm is used for inter-region coupling optimization to construct a region association operator. The parameters of this operator are linked to the length of the shared edges / faces, the gradient of the thermal conductivity coefficient, and the airflow velocity vector shared by adjacent regions. The local optimal solutions of adjacent regions are used as input parameters to transmit to the optimization process of the associated regions, realizing the synergy of optimization information between regions. The information transmission efficiency is positively correlated with the length of the shared edges / faces. Shared edges / faces refer to the non-overlapping and gapless common geometric boundaries between two or more adjacent meshes, analogous to the partition walls in a building: two adjacent rooms share a wall, which is both the boundary of the first room and the boundary of the second room. There are no gaps in the middle of the walls or overlapping of the two walls.

[0147] For example, suppose there is an office area grid X and a corridor area grid Y within the target building. One side boundary of grid X and one side boundary of grid Y completely coincide, forming a common geometric boundary, namely a shared surface area barrier feature F. Since grids X and Y have a shared surface area barrier feature F, the office area and the corridor area are determined to be adjacent target equipment control areas. Based on this adjacency relationship, inter-region coupling optimization is carried out to construct a region association operator related to the shared surface area barrier feature F.

[0148] An elite retention strategy is introduced to select the global Pareto optimal solution from the local optimal solution set of each region and the coupling optimization solution set between regions. During the selection process, the grid computation efficiency weight after multi-objective genetic algorithm optimization in the 3D region segmentation model of the target building is referenced. That is, the grid region with higher computation efficiency has a higher selection weight for its optimal solution.

[0149] The selected optimal solution is included in the global optimal solution pool. The optimal energy consumption rate of all target device control areas is solved through iterative calculation. The number of iterations is negatively correlated with the total number of grids, that is, the fewer the number of grids, the fewer the number of iterations.

[0150] It should be further explained that, in this embodiment, the process of feeding back the comfort assessment score and the deviation between the real-time temperature and the upper limit of the target temperature range obtained from each round of simulation to the generator to correct the first and second association mappings includes:

[0151] The deviation of the comfort assessment score and the deviation between the real-time temperature and the upper limit of the target temperature range are collected in each round of simulation. The deviation collection points correspond one-to-one with the grid nodes in the three-dimensional area segmentation model of the target building.

[0152] The deviation is divided into short-term and long-term deviations according to the time scale, and the time threshold for deviation division is based on the functional scenario of the grid coverage area. For example, the short-term deviation threshold of the office scenario is shorter than that of the rest scenario. Assume that there is an office functional scenario C (corresponding to the grid group area barrier features C1, C2, C3) and a rest functional scenario D (corresponding to the grid group area barrier features D1, D2) within the target building.

[0153] When decomposing deviations, the short-term deviation time threshold T_C set for office function scenario C is shorter than the short-term deviation time threshold T_D set for rest function scenario D. That is, when the duration of the deviation in office function scenario C does not exceed T_C, it is judged as a short-term deviation, while in rest function scenario D, the duration of the deviation must exceed T_D to be judged as a long-term deviation.

[0154] To address short-term deviations, a reinforcement learning algorithm is used to treat the deviation as a reward signal and adjust the real-time parameters of the first and second association maps. The magnitude of parameter adjustment is positively correlated with the density of the corresponding grid in the 3D region segmentation model of the target building (the higher the density, the more refined the parameter adjustment). Priority is given to adjusting the mapping parameters of regions where the streamline direction of the airflow adaptation grid deviates significantly from the actual airflow, thereby achieving short-term dynamic correction of the mapping.

[0155] To address long-term biases, a transfer learning algorithm is used to extract historical bias correction experiences and construct a bias correction knowledge graph. The construction of the knowledge graph is associated with the annual decay characteristics of the thermal conductivity coefficient of the thermal resistance adaptation mesh in the 3D region segmentation model of the target building. The more significant the decay, the higher the weight of the historical correction experience. The correction rules in the knowledge graph are then transferred to the current mapping correction process to optimize the basic association rules of the mapping.

[0156] The deviation attribution algorithm is used to analyze the source of deviation. In the attribution process, the mesh characteristics of the target building's 3D area segmentation model are combined: if the deviation is concentrated in the deformable mesh area of ​​the dynamic barrier component, it is determined to be caused by the change of the barrier boundary. For example, assuming that the mesh P of the conference room area in the target building is the deformable mesh area corresponding to the dynamic barrier component (movable partition), if the deviation is continuously concentrated in the range of mesh P in multiple rounds of simulation, and the deviation in other areas is dispersed, it is determined that the deviation is caused by the change of the barrier boundary of the movable partition (such as the change of the spatial range of the mesh area caused by the movement of the partition).

[0157] If the deviation is concentrated in the grid region with high thermal resistance gradient, it is determined to be caused by thermal conductivity coefficient drift. The correction direction is adjusted according to the attribution results: for deviations caused by equipment performance drift, the focus is on correcting the energy consumption parameters bound to the thermal resistance-adapted grid parameters in the second association mapping; for deviations caused by pedestrian flow prediction errors, the focus is on correcting the pedestrian flow association parameters bound to the initial grid coverage area associated with the equipment in the first association mapping. This completes the accurate correction of the first and second association mappings, and the correction results are deeply adapted to the 3D grid space features. Assuming that the grid Q of the target building's exterior wall area is a grid region with high thermal resistance gradient, if the deviation continues to be concentrated within the range of grid Q, it is determined that the deviation is caused by the thermal conductivity coefficient drift of the exterior wall envelope structure.

[0158] This embodiment constructs a high-fidelity 3D regional segmentation model of the target building through a full-process design involving layered parameter processing, BIM and point cloud fusion, and multi-round mesh adaptation. Specifically, firstly, physical, structural, and equipment parameters are processed in layers according to categories using analytical algorithms, avoiding model distortion caused by parameter mixing and ensuring accurate binding of geometric information and physical parameters of building components, including the thermal conductivity coefficient of the building envelope. Secondly, combined with BIM and point cloud data fusion algorithms, the virtual model is completely aligned with the outline and component positions of the building entity, solving the problem of the disconnect between geometric and physical attributes in traditional 3D modeling. Subsequently, the weighted Voronoi diagram algorithm was used to achieve initial mesh segmentation for device association. Combined with an adaptive mesh refinement algorithm based on thermal resistance gradient to refine high-error areas and control the shared edge / face ratio to ensure mesh quality, a streamline tracing algorithm to adapt mesh edges to airflow direction, and a virtual boundary algorithm to adapt to dynamic barrier components, a mesh model with thermal resistance, airflow, and barrier three-adaptation was constructed. This model can accurately reflect the differences in heat conduction in different areas of the building, such as the thermal resistance gradient between the exterior windows and the walls, airflow patterns, such as the airflow exchange between the fresh air duct and the interior, and dynamic spatial changes, such as the movement of movable partitions. This provides a basic framework for subsequent energy consumption prediction and optimization that is spatially perceptible and dynamically adaptable, avoiding optimization deviations caused by the uniform mesh ignoring regional characteristics in traditional modeling.

[0159] In terms of the control mapping module, this embodiment addresses the pain points of traditional energy consumption optimization, such as linearized parameter correlation and neglect of spatial dynamic characteristics, by constructing a multi-level correlation mapping of pedestrian density, cooling rate, equipment energy consumption, and environmental comfort. Specifically, the pedestrian density and cooling rate correlation unit, based on the thermal resistance grid characteristics in the three-dimensional area segmentation model of the target building, uses a clustering algorithm to divide thermal comfort zones, enabling the target comfort temperature range to adapt to the thermal resistance differences in different areas. For example, a narrower temperature range is set in areas with frequent heat exchange near exterior windows to enhance control accuracy, while a wider range is set in the indoor core area to improve system flexibility, thereby effectively avoiding local overcooling or overheating caused by uniform temperature settings. Furthermore, this unit constructs a first-level correlation mapping by binding sensor data to grid nodes, implementing high-frequency data acquisition in high-density grid areas, and combining a dynamic scene probability correction mechanism to integrate the influence of functional scene weights and dynamic barrier components. This mapping can accurately capture the dynamic relationship between pedestrian density and the required cooling rate in different grid areas. For example, high-pedestrian areas such as conference rooms correspond to higher cooling requirements, while low-pedestrian areas such as corridors correspond to lower cooling requirements.

[0160] Further introducing geometric transport constraints, this unit constructs a dynamic covariate that integrates three influencing factors: thermal resistance, airflow, and barrier properties. This is achieved by clustering thermal resistance grid levels, matching airflow velocity vectors using spatial coordinates, and monitoring the state of dynamic barrier components. Based on this, a second-level correlation mapping is established, comprehensively reflecting the impact mechanism of multiple factors on equipment energy consumption. For example, high thermal resistance areas can correspond to lower energy consumption due to good insulation performance, while high airflow areas can also reduce cooling energy consumption due to airflow-assisted heat dissipation. Building upon this, the control and adjustment unit introduces a comfort score to construct a feedback mechanism, forming a second-level correlation feedback mapping to achieve a dynamic balance between equipment energy consumption and environmental comfort. This multi-level mapping structure effectively overcomes the limitations of traditional linear correlations, deeply binding mapping parameters to the physical characteristics of the building space, providing a spatially differentiated and factor-integrated calculation basis for energy consumption prediction.

[0161] In terms of the simulation module, this embodiment adopts a dual-unit architecture combining adversarial prediction and discriminative optimization to achieve high-precision prediction of energy consumption rate and heating rate, and supports efficient solution of optimal control strategy. Specifically, the adversarial prediction unit, based on a grid anchoring generator, binds multi-level correlation mapping parameters to grid attributes, enabling real-time collected pedestrian density and temperature data to be accurately matched to the corresponding grid region. This unit combines a grid weighted prediction mechanism, assigning higher weights to high-density grids to improve overall prediction accuracy, and introducing correction coefficients for thermal resistance and airflow in the dynamic calculation of thermal balance, thereby ensuring that the predicted results of energy consumption rate and heating rate can accurately reflect regional characteristics. For example, the heating rate is slower in high thermal resistance grid regions, while high airflow grid regions can reduce cooling energy consumption due to airflow-assisted heat dissipation. In terms of comfort assessment, this unit uses a dimension-grid attribute binding mechanism to correlate thermal comfort with thermal resistance grid temperature fluctuations, airflow comfort with airflow grid direction deviations, and humidity comfort with barrier grid sealing performance, and combines a fuzzy rule weight allocation strategy to assign higher evaluation weights to high-density grids. This method effectively overcomes the problems of strong subjectivity and lack of regional differentiation in traditional comfort assessment, enabling the scoring results to accurately reflect the actual comfort status of each grid area.

[0162] The identification unit further constructs a multi-dimensional adversarial error fusion mechanism, integrating spatial correlation errors and temporal cumulative errors to adapt to the spatial distribution characteristics of the grid. Through a hierarchical multi-objective optimization function, this unit achieves priority adaptation of equipment power weights within the region, as well as dynamic matching of comfort dimension weights with airflow grid characteristics. Simultaneously, a dynamic constraint set ensures linkage between equipment operating status and grid coverage, and between thermal coupling strength and barrier grid boundaries. Based on this, a dynamic multi-objective evolutionary algorithm is employed to perform local optimization within a single region and implement collaborative optimization between regions. Combined with an elite retention strategy, the globally optimal solution is selected, thereby efficiently solving for the optimal energy consumption rate of each grid region. This optimization process ensures an effective balance between minimizing energy consumption and maximizing comfort, while fully adapting to the dynamic characteristics of the building space, avoiding the energy waste or insufficient local comfort caused by traditional globally unified optimization strategies that ignore regional differences.

[0163] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A digital twin-based dynamic building energy consumption optimization system, characterized in that, include: Modeling module, control mapping module, and simulation module; The modeling module is used to acquire the physical parameters, structural parameters, and layout parameters of the target building and the target equipment. It then constructs a 3D model of the target building using a 3D modeling algorithm. Based on the distribution of the target equipment in the 3D model, the functional coverage of a single target equipment, and the geometric transmission constraints of the target building, it performs region segmentation using a mesh segmentation algorithm to obtain a 3D region segmentation model of the target building. The geometric transmission constraints include the thermal conductivity coefficient of the building envelope, airflow path, and regional barrier characteristics. The control mapping module is used to collect real-time pedestrian density, target equipment energy consumption status, temperature change rate, target temperature range and comfort score within the target equipment control area, and construct a multi-level correlation control mapping for each target equipment control area through correlation analysis algorithm; The simulation module constructs a generator based on the multi-level associated control mapping of each target device control area, and constructs a discriminator based on the prediction results of the multi-level associated control mapping of all target device control areas combined with a multi-objective optimization algorithm. Through the generator, discriminator, and simulation algorithm, combined with the target building's three-dimensional area segmentation model, simulation is performed to output the collaborative control strategy of the target devices in real time. The control mapping module includes a human flow-rate correlation unit, a rate-energy consumption conversion unit, and a control adjustment unit. The human flow-rate correlation unit is used to obtain the first correlation mapping corresponding to the initial human flow density and cooling rate by fitting different human flow densities and corresponding actual cooling rates with the population density time distribution function collected in each target device control area under a preset target comfort temperature range, combined with a multiple regression algorithm. The population density time distribution function is constructed by combining the population per unit area in each target device control area for each preset time period with a geometric probability distribution. The rate-energy consumption conversion unit is used to obtain a second correlation mapping between cooling rate and target energy consumption based on the covariates constructed by combining the target equipment energy consumption rate corresponding to each actual cooling rate with the geometric transmission constraints, through a correlation analysis algorithm. The control adjustment unit is used to construct an energy consumption adjustment coefficient based on the reciprocal of the comfort score corresponding to the control area of ​​each target device, and map the energy consumption adjustment coefficient to the second association mapping to obtain the second association feedback mapping; The simulation module includes an adversarial prediction unit. The adversarial prediction unit is used to construct a generator based on the first correlation mapping and the second correlation feedback mapping, and combine the real-time collected crowd density and real-time temperature of the single target device control area to predict the energy consumption rate and heating rate of the single target device control area at the current moment. At the same time, it performs a comfort assessment on each target device control area using a comprehensive fuzzy algorithm to obtain a comfort assessment score for each target device control area. The simulation module includes an identification unit. This identification unit is used to construct adversarial errors based on the predicted energy consumption rate and corresponding heating rate of all target equipment control areas at the current moment, as well as the deviation between the real-time energy consumption rate and the heating rate. It constructs a multi-objective optimization function to minimize the overall energy consumption of all target equipment control areas and maximize the comfort score. Using geometric transmission constraints, equipment energy consumption constraints, equipment coverage area, and target temperature range as the objective constraint set, it performs synchronous adversarial simulation with the target building's three-dimensional area segmentation model through a simulation algorithm. Simultaneously, it uses a dynamic multi-objective evolutionary algorithm to solve for the optimal energy consumption rate adversarial problem for each target equipment control area. The comfort evaluation score and the deviation between the real-time temperature and the upper limit of the target temperature range obtained in each round of simulation are fed back to the generator to correct the first and second association mappings, ensuring that the overall energy consumption and comfort score of the target equipment meet preset thresholds.

2. The digital twin-based building energy consumption dynamic optimization system as described in claim 1, characterized in that, The system also includes a feedback adjustment module; the feedback adjustment module is used to collect the actual operating data and comfort score feedback data of the target device under the cooperative control strategy, compare the actual operating data and comfort score feedback data with preset thresholds, and adjust the parameters of the multi-level associated control mapping and the constraint weights of the multi-objective optimization algorithm according to the comparison results, so that the overall energy consumption and comfort score of the target device meet the preset thresholds.

3. The digital twin-based building energy consumption dynamic optimization system as described in claim 2, characterized in that, The process of constructing the three-dimensional region segmentation model of the target building includes: The physical parameters, structural parameters, and target equipment layout parameters of the target building are obtained and then processed hierarchically using an analytical algorithm. The physical parameters are stored hierarchically according to the type of building components, the structural parameters are associated hierarchically according to spatial hierarchy, and the target equipment layout parameters are mapped hierarchically according to the equipment function type, thus obtaining a hierarchical parameter set. Based on the hierarchical parameter set, the geometric information and physical parameters of building components are bound together by the BIM and point cloud data fusion algorithm. A three-dimensional model of the target building foundation containing thermal properties is constructed by the three-dimensional mesh generation algorithm, so that each building component in the target building foundation three-dimensional model is associated with the corresponding thermal conductivity coefficient of the building envelope. Based on the distribution of target equipment in the 3D model of the target building foundation, the weighted Voronoi diagram algorithm is used for initial mesh segmentation, with the rated power of the equipment as the weight factor. At the same time, the functional coverage range of a single target equipment and the regional barrier characteristics are used as constraints to obtain the initial mesh model of equipment association.

4. The digital twin-based building energy consumption dynamic optimization system as described in claim 3, characterized in that, The process of constructing the three-dimensional region segmentation model of the target building also includes: Extract the thermal conductivity coefficient of the building envelope from the geometric transport constraints, and use the thermal resistance gradient analysis algorithm to calculate the gradient distribution of the thermal conductivity coefficient in the building space; Based on the gradient distribution results, an adaptive mesh refinement algorithm is used to increase the mesh density change rate along the gradient direction of the thermal conductivity coefficient in the initial mesh model, so that the mesh boundary of the region where the difference in thermal conductivity coefficient exceeds the preset difference threshold is consistent with the gradient direction, thus obtaining a thermal resistance adapted mesh model.

5. A digital twin-based building energy consumption dynamic optimization system as described in claim 4, characterized in that, The process of constructing the three-dimensional region segmentation model of the target building also includes: The airflow path in the geometric transport constraints is extracted and combined with a CFD simulation algorithm to obtain the airflow velocity vector distribution on the path; The thermal resistance adaptation mesh model is topologically corrected based on vector distribution combined with streamline tracing algorithm to ensure that the edge direction of the mesh is consistent with the mainstream airflow direction. At the same time, the mesh of the airflow intersection area is divided using polyhedral mesh rules to construct the airflow adaptation mesh model.

6. A digital twin-based building energy consumption dynamic optimization system as described in claim 5, characterized in that, The process of constructing the three-dimensional region segmentation model of the target building also includes: The regional barrier features in the geometric transmission constraints are extracted, and a boundary recognition algorithm is used to distinguish between fixed barrier components and dynamic barrier components. For fixed barrier components, a boundary fitting algorithm is used to make the mesh boundary and the physical boundary of the component completely coincide. For dynamic barrier components, a virtual boundary algorithm is used to preset the deformable area of ​​the mesh within its movement range to obtain a barrier-adaptive mesh model. Based on the thermal resistance adaptive mesh model, airflow adaptive mesh model and barrier adaptive mesh model combined with a multi-objective genetic algorithm, the optimization objective is to use the mesh division accuracy and the total energy consumption corresponding to the control area of ​​all target equipment as the optimization objective, and the thermal conductivity gradient, airflow path and regional barrier boundary as the constraints. The size, shape and topological relationship of the mesh are iteratively adjusted to obtain the three-dimensional regional segmentation model of the target building.

7. A digital twin-based building energy consumption dynamic optimization system as described in claim 6, characterized in that, The implementation process of the adaptive mesh encryption algorithm includes: calculating the error distribution of the initial mesh in the heat conduction simulation using the finite element error estimation method, marking the region where the error exceeds the preset error threshold as the encryption region, and using a quadtree or octree subdivision algorithm to divide the encryption region into meshes, while keeping the proportion of shared edge or face length of adjacent meshes not exceeding the preset proportion during the division process.

Images