Physical building data visualization and simulation analysis system

Abstract

The invention relates to the technical field of building intelligent management and operation control, and discloses a solid building data visualization and simulation analysis system, which is characterized in that dynamic monitoring data is acquired by constructing a building three-dimensional model and deploying multiple types of sensors, and a tensor structure is generated through standardization and three-dimensional index fusion; mapping the tensor into a three-dimensional model to realize regional data visualization; historical sequence modeling and strategy-driven future simulation prediction are carried out based on a sliding window; a user can control system behaviors through model interaction, the system automatically monitors index changes and triggers strategy simulation and visual dynamic updating in a linkage mode, and visual analysis and intelligent response control of the building state are achieved. Through fusion of three-dimensional model mapping, tensor construction and strategy response prediction, building state multi-dimensional visualization, continuous intelligent prediction and dynamic feedback control are realized, and innovative technologies such as sliding window modeling and interactive strategy driving are provided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building intelligent management and operation control, and particularly relates to a physical building data visualization and simulation analysis system. BACKGROUND

[0002] At present, in the field of modern intelligent building operation management, how to realize deep perception of multi-source sensing data, accurate prediction of state trend and efficient response of strategy intervention has become a key requirement for improving building energy efficiency, safety and comfort. Especially in dealing with complex scenes such as energy consumption control, environmental regulation and personnel behavior guidance, the traditional control system based on static rules or single-step prediction is difficult to meet the management needs of dynamic evolution and linkage response. The present application is proposed in this background, aiming to realize the forward modeling and intelligent feedback control of building operation state by fusing tensor prediction modeling and strategy response mechanism.

[0003] In the prior art, some systems have begun to use time series models, neural network methods and the like to predict the building environment or energy consumption state, and combine preset strategies for scheduling control. Such systems generally construct input sequences based on historical observation data, output prediction values for one or a limited number of time steps in the future through the model, and evaluate the control strategy with the help of a rule base or optimizer. Some advanced systems also try to introduce user behavior prediction or external disturbance information to improve the accuracy of prediction and the adaptability of control. These technical routes have promoted the intelligent development of building management to a certain extent, and have feasibility and certain practical effect.

[0004] However, the existing technology still has some deficiencies in the linkage processing of multi-step trend modeling and strategy response feedback: first, most of the current prediction methods focus on single-step or short-term prediction, which is difficult to form a stable medium and long-term trend judgment, so that the system cannot fully support predictive control; second, the existing control strategies are mostly static injections, which lack the ability to model the continuous impact of strategy variables in the time evolution process, making it difficult for the system to evaluate the global impact of strategy intervention on future time series; in addition, due to the lack of full use of the multi-dimensional expression advantage of tensor structure, the existing model has limitations in dealing with the complex space-time-category data interaction relationship in the building, and cannot realize the collaborative modeling and response simulation of high-dimensional state. These problems limit the adaptive ability of intelligent building systems to dynamic changes, and also affect the control accuracy and deployment versatility of the system in complex operation scenarios. SUMMARY

[0005] The purpose of the present application is to provide a physical building data visualization and simulation analysis system, which solves the problems of insufficient multi-step trend prediction ability, dynamic evolution modeling of strategy intervention effect and limitation of collaborative prediction of complex space-time-category multi-dimensional data in the prior art.

[0006] To achieve the above object, the present application is realized by the following technical solutions: An entity building data visualization and simulation analysis system, the system comprises: A building three-dimensional model and data acquisition module, for constructing a three-dimensional geometric structure model of a building, and deploying multiple types of sensors based on the model to collect dynamic monitoring data covering spatial position, time point and data category dimension; A multi-dimensional data fusion and tensor construction module, for receiving the dynamic monitoring data, and performing standardized preprocessing on the dynamic monitoring data, and fusing and constructing a unified tensor data structure according to the three-dimensional indexes of space, time and category; A visualization mapping and three-dimensional projection module, for receiving the three-dimensional data tensor output by the tensor construction module, and mapping it into the building three-dimensional model to realize the visualization rendering of the data state of each region in the building space; A simulation prediction and strategy generation module, for performing historical time series modeling and sliding window prediction based on the tensor data, and generating corresponding future simulation data tensor based on input strategy variables; An interactive control and feedback linkage module, for managing user interaction behavior on the building three-dimensional model, scheduling visualization content, and monitoring whether each type of index in the tensor data exceeds the set threshold, and if it exceeds, triggering the simulation prediction module to generate strategy response data and feeding back to the visualization mapping and three-dimensional projection module to realize dynamic update display.

[0007] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application combines the tensor sequence constructed by the sliding window with the deep learning prediction function to realize multi-dimensional and continuous trend prediction of the building state. The system not only supports single-step prediction, but also has multi-step rolling output capability, meets the needs of medium and long-term operation evaluation, significantly enhances the early perception ability of future state changes, and is suitable for dynamic management scenarios of complex building systems.

[0008] 2. The present application introduces a strategy variable injection mechanism, which can simulate the state evolution process under different management strategies based on the prediction results. By fusing user intervention intention and system prediction path, a strategy response model is constructed, so that the system has the ability to quickly generate feedback responses based on hypothetical conditions, and is widely applicable to the fields of building energy consumption control, personnel evacuation guidance, environmental optimization adjustment, etc.

[0009] 3. By introducing a trend weighting mechanism based on convolution kernels, the present application enhances the modeling capability of short-term change trends in the time series modeling process, making the model more sensitive and stable when dealing with sudden, periodic or slow fluctuation phenomena. At the same time, this mechanism has good scalability and can be flexibly adapted to different types of building sensor data and system state variables.

[0010] 4. The present application constructs an interactive control and feedback linkage module based on policy injection, which supports users to adjust policy parameters through a graphical interface and obtain real-time simulation results, effectively building an intelligent operation environment for human-machine co-management. This mechanism can assist decision-makers in dynamically optimizing building dispatching strategies, realizing the transition from "passive monitoring" to "active intervention" in operation and management paradigm. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a schematic diagram of the system structure of the present application; Figure 2 is a schematic diagram of the module architecture of the building three-dimensional model and data acquisition module of the present application. DETAILED DESCRIPTION

[0012] The present application will be further described below in conjunction with the accompanying Figure 1 and the accompanying Figure 2 , to make a further detailed description of the present application.

[0013] The present application provides an entity building data visualization and simulation analysis system, which introduces a sliding window tensor structure, a trend convolution modeling mechanism and a policy variable response function, and constructs an extensible, multi-step prediction and feedback integrated building operation state simulation system, improving the intelligent level and practicality of building systems in dynamic prediction and policy response.

[0014] As shown in Figure 1 and Figure 2 , the entity building data visualization and simulation analysis system can include: a building three-dimensional model and data acquisition module for constructing a three-dimensional geometric structure model of a building and deploying multiple types of sensors based on the model to collect dynamic monitoring data covering spatial location, time point and data category dimensions; In this embodiment, the building three-dimensional model and data acquisition module constructs the geometric semantic structure of the entity building space, and on this basis, completes real-time acquisition and standardized structured processing of multiple types of environmental and operational data, supporting subsequent data fusion modeling and visualization analysis tasks.

[0015] To realize accurate geometric expression and data mapping reference of building space, this embodiment adopts multiple source modeling methods such as building information modeling (BIM), laser radar scanning (LiDAR) and unmanned aerial vehicle image reconstruction to construct a high-fidelity three-dimensional building three-dimensional model.

[0016] It is preferred to introduce architectural design drawing data, including plan, elevation and structural drawing, in the BIM modeling process. A readable and writable structural semantic model is generated through CAD tools and BIM platforms (such as Revit, ArchiCAD, etc.). The model has information such as spatial topological structure, component hierarchical relationship and material properties.

[0017] For buildings that have been built or historic buildings, laser radar scanning technology is combined to collect dense point cloud data sets through mobile or fixed platforms. Surface fitting and voxel reduction are completed with the help of point cloud processing software (such as CloudCompare or RealityCapture), and a three-dimensional entity structure that can be connected with the BIM model is output.

[0018] If the target building is a large-scale outdoor space or a special structural form, a high-resolution photographic device can be further used on a drone to obtain multi-angle overhead images, and a three-dimensional reconstruction of the images is realized through a structure light reconstruction algorithm (Structure-from-Motion, SfM) to obtain an external contour model for supplementing areas where the laser or BIM model is insufficient in precision.

[0019] The above modeling process finally generates a three-dimensional building geometric model in a unified coordinate system, which serves as a spatial framework for subsequent sensor deployment and data mounting. The model has clear spatial unit division capability (such as floors, rooms, corridors, open areas, etc.), and each unit is assigned a spatial index code s∈{1,2,...,m} for spatial registration in the data fusion stage.

[0020] After the construction of the three-dimensional building model, the embodiment further relies on the model to deploy multiple types of environmental and state sensors in the internal and external key areas of the building. The types of sensors include but are not limited to the following categories: Environmental perception: temperature sensor, humidity sensor, air quality monitor (for collecting PM2.5, carbon dioxide concentration, etc.); Energy consumption perception: electricity meter, air conditioning load meter, lighting power monitoring device; People flow monitoring: infrared counter, visual recognition module (such as video people estimation), Bluetooth / WiFi sniffing device; Structural state (optional): vibration sensor, displacement meter, etc., for detecting the physical state of building components.

[0021] The deployment of sensors is preferably based on building functional zoning and use density, with key locations such as entrances, public areas, mechanical and electrical rooms, conference rooms, corridor nodes being covered. Through the integration of BIM models and GIS information, visual deployment and spatial coding labeling of sensors can be achieved.

[0022] The sensor data collection process supports both timed collection and event-driven collection modes. All collected data is automatically attached with meta-information, including: a spatial location index s, which is bound to the three-dimensional position coordinates (x, y, z) in the three-dimensional model of the building; a collection timestamp t, which is based on the system master clock and has a uniform time resolution; a data category number c, which is used to distinguish the sources of various sensors; a data value d, which is the actual sampled observation value.

[0023] The final generated raw data set can be formally represented as a sequence of four-tuples: D = {(s i ,t i ,c i ,d i ) | i = 1,..., N}; where N is the number of data records collected within the current time window. This data structure ensures that each data record has a clear identification of the spatial, temporal, and semantic dimensions, facilitating subsequent data fusion processing.

[0024] To further standardize the data format and interface, the embodiment uses a standardized communication protocol (such as MQTT or RESTful API) to transmit sensor data in real time to the data center server. After receiving the data, the server side performs preliminary format specification and cache storage through the ETL module (Extraction-Transformation-Loading), forming an initial data table.

[0025] Considering the heterogeneity of building space and the complexity of data types, the embodiment sets up an edge computing node on the data collection side to perform basic data cleaning and compression operations locally, reducing redundant information and improving data quality.

[0026] In addition, to ensure accurate alignment between sensor positioning and the three-dimensional model, the system establishes a spatial mapping table that records the sensor device ID, the spatial index of the installation location, the position coordinates in the three-dimensional model, and the region code, forming a binding relationship between the data and the geometric model.

[0027] The multi-dimensional data fusion and tensor construction module is used to receive dynamic monitoring data and perform standardized preprocessing on the dynamic monitoring data, and to fuse and construct a unified tensor data structure according to the three-dimensional indexes of space, time, and category; In the embodiment, the multi-dimensional data fusion and tensor construction module is mainly used to receive the original multi-source dynamic monitoring data provided by the building three-dimensional model and the data collection module, and to perform standardized preprocessing on it. Subsequently, according to the spatial position, time point, and data category dimensions, a three-dimensional tensor structure is constructed for downstream visualization and simulation analysis.

[0028] To realize the multi-dimensional alignment and unified modeling of data, the module first performs format specification processing on the input data set. The received data is a heterogeneous data set collected by different types of sensors at different time nodes and different spatial locations. The original data usually contains a combination of spatial location index, collection timestamp, data category number, and observation value, which is formalized as a four-tuple sequence disclosed by the building three-dimensional model and data collection module.

[0029] The embodiment preferably introduces a data preprocessing process to improve data quality, regularize data structure, and ensure the consistency of subsequent tensor construction. The data preprocessing steps include but are not limited to the following aspects: First, perform format unification operation. For sensor data from different devices, protocols or manufacturers, convert them to the standard structure supported by the system through the data adapter module, and eliminate problems such as inconsistent units and non-standard naming.

[0030] Second, implement time alignment processing. Due to the differences in sensor sampling frequency, the system maps all data to the set time resolution through the main time axis resampling mechanism, and supplements the missing time nodes to ensure that multi-dimensional parallel computing can be performed at the same time step.

[0031] Next, perform missing value filling. For missing or lost data points, the system uses interpolation, forward filling or sliding mean value to complete the preliminary filling, and preferably uses a forward interpolation strategy with constraints to preserve the original trend while avoiding abnormal spread.

[0032] In addition, the embodiment also introduces an outlier detection mechanism. By using statistical threshold-based methods or anomaly detection models, data that deviates significantly from historical statistical distribution is identified, marked and removed to improve the overall stability of the data.

[0033] After the above preprocessing is completed, the system constructs a tensor model according to the three-dimensional index structure. Specifically, let m be the number of building space units, n be the total number of sampling time steps, and p be the number of collected data categories, then the data is organized and constructed into a three-dimensional tensor structure: Where each element in the tensor represents the observation value at spatial unit s, time point t, and category number c. is the three-dimensional tensor structure after construction.

[0034] To reduce the problem of incomplete structure caused by missing data, the embodiment further introduces a low-rank matrix decomposition completion mechanism to complete the missing value processing in the tensor. Specifically: For the slice of the tensor in a fixed category c It is unfolded into a matrix form And a low-rank approximation operation is performed. The method sets the target matrix Can be approximately expressed as the product of two low-rank matrices , that is, X≈UV T . The optimization goal is to minimize the reconstruction error, that is: Where ||·|| F F represents the Frobenius norm, which is used to measure the overall difference between matrices.

[0035] Low-rank decomposition can be realized based on singular value decomposition (SVD) or principal component analysis (PCA) method. This processing method fully excavates the low-dimensional structural characteristics of the existing data without introducing too many new assumptions, and can provide reasonable estimation in the data sparsity scenario.

[0036] To enhance the modeling capability of the time dimension, the embodiment can also optionally introduce a time trend extraction mechanism. This mechanism performs local convolution on the tensor in the time dimension in a sliding window manner to extract its short-term time trend, which is formally expressed as: Where K(τ) is the convolution kernel function in the time dimension; w is the set window width; is the tensor value after time trend convolution processing; is the historical observation value in the original input tensor; τ is the time delay index within the window; This way helps to capture short-period fluctuation characteristics and provides a stronger context basis for prediction modeling.

[0037] The finally constructed tensor structure has good spatial consistency, temporal continuity and category separation, and can serve as a unified data abstraction carrier to support parallel processing of multiple modules such as three-dimensional visualization mapping, time series simulation and prediction, and strategy response analysis.

[0038] In addition, to improve the scalability and data interface compatibility of the system, the embodiment suggests using a tensor interface specification based on key-value mapping. When tensor data is transmitted between system modules, the spatial index table, time axis definition and category label coding are explicitly defined to ensure the context integrity of the data when scheduling across modules.

[0039] The visualization mapping and three-dimensional projection module is used to receive the three-dimensional data tensor output by the tensor construction module, and map it to the building three-dimensional model to realize the visualization rendering of the data state of each region in the building space; In this embodiment, the visualization mapping and three-dimensional projection module is used to receive the three-dimensional multi-dimensional data tensor output by the tensor construction module, and map the space-time-category data state contained in the tensor to the three-dimensional building model, realizing the visualization rendering and dynamic presentation of the state changes of each space region in the building.

[0040] To achieve the above functions, the embodiment first maps and defines the visual attributes of each data unit in the tensor data . Specifically, each element in the tensor corresponds to a building space position index s, a time step t, a data category code c, and an observation value V in the category. The system uses a set of preset visual mapping functions M c (·) to convert it into a set of visual attributes that can be rendered by a graphics engine, denoted as: where V s,t,c represents the mapped visual attributes, which preferably include color coding, geometric deformation parameters, or particle path density, etc.

[0041] The construction of the mapping function M c is based on the physical meaning and perceptual characteristics of different category data. For example, for temperature and humidity data, a color gradient method is preferably used for continuous expression, such as mapping the temperature interval to a blue-red gradient color band; and for people flow density data, particle path or dynamic arrow direction display can be introduced to enhance the perception of flow.

[0042] After the visual attribute conversion is completed, the embodiment introduces a spatial position matching mechanism to ensure the alignment of the tensor index and the three-dimensional model space structure. This process relies on the Spatial Mapping Table, which records the binding relationship between the index s in the tensor and the space unit in the three-dimensional building model, including but not limited to floor number, room ID, geometric coordinate range, component level, etc.

[0043] The system iterates through the tensor data, and at each time t, the visual attributes V s,t,* corresponding to all space indexes s∈{1,...,m} are mounted to the corresponding geometric nodes in the model, and real-time rendering is completed through a graphics rendering engine (such as Unity3D, CesiumJS, or WebGL platform).

[0044] In the implementation process, each visual attribute element is attached to the model geometry node in the form of attribute binding, supporting GPU-accelerated material update and layer control. For example, the color gradient can be dynamically replaced as a texture mapping parameter to model material map; the deformation parameter can be bound to the node transformation matrix to affect its scaling and displacement attributes; and the particle path can be used as an overlay layer to calculate and dynamically generate in real time in the spatial path.

[0045] To support the dynamic expression of building state evolution over time, the embodiment introduces a tensor time frame slicing mechanism. A dynamic animation timeline is set The system extracts the tensor slice n at each time frame t The corresponding visual state is generated according to the above mapping method, and played in the graphics engine according to the time frame switching method. This method forms a spatial state time sequence animation based on the tensor time dimension, realizing the visual deduction of building operation state history playback or future prediction.

[0046] In addition, the embodiment also integrates an interactive user control interface, allowing users to perform the following operations in the three-dimensional building model: Control the observation view angle, including rotation, scaling, roaming, and sectioning; Select a specific data category c k and only show the visual attributes under this category; Select any time point t j through the TimeSlider and update the corresponding tensor slice mapping in real time; Adjust the mapping function parameters, such as color band range, threshold setting, and particle density, to meet the requirements of specific analysis scenarios.

[0047] The user interface and the visual engine interact and schedule through a unified visualization control protocol, ensuring synchronization and high responsiveness during data update, attribute transformation, and scene refresh.

[0048] The system preferably uses a double buffering mechanism during runtime to cache the tensor slice mapping of the next time frame on one side and perform rendering and display of the current frame on the other side, thereby realizing seamless switching between consecutive frames without lag.

[0049] To support the export and reproduction of subsequent analysis results, the embodiment can also save the user-selected view angle, time frame, category configuration, and mapping parameters as a configuration file (such as JSON or XML format) after the user operation is completed, and generate corresponding visualization snapshots or animation files, facilitating the presentation of achievements and external display.

[0050] In summary, the visualization mapping and three-dimensional projection module of the embodiment realizes the spatialization, graphization and dynamic expression of the multi-source operation state of the physical building by constructing the mapping bridge between the data and the model, and provides an operable and extensible basic support unit for the system to realize interactive analysis and simulation.

[0051] The simulation prediction and strategy generation module is configured to perform historical time series modeling and sliding window prediction based on the tensor data, and generate a corresponding future simulation data tensor based on an input strategy variable. In the embodiment, the simulation prediction and strategy generation module is configured to perform time series modeling and sliding prediction based on the three-dimensional dynamic data tensor output by the aforementioned tensor construction module, and generate a corresponding future simulation data tensor based on a strategy variable set by a user, so as to support forward-looking analysis and strategy decision support of the building operation state.

[0052] First, for the simulation prediction task, the embodiment adopts a time series extraction mechanism based on a sliding window, and sets a time window with a fixed width on the tensor to slice the historical sequence. Specifically, assuming that the sliding window width is w, at time step t, a historical sub-tensor segment is extracted and represented as: wherein, is a historical time window sequence tensor, representing the observation value sequence of the spatial unit s and the data category c in the past continuous w time steps (from t-w+1 to t), which is used to model the time series trend; is the observation value at the t-τ time point under the spatial index s and the category index c; w is the time window length, defined as the number of historical steps referred to for one prediction; s, t, and c are the spatial index, the time index, and the data category index, respectively.

[0053] The above time series sub-tensor is then input into a prediction function f θ (·) to generate the prediction value of the next time step t+1, and the specific form is as follows: wherein, is the tensor value generated by prediction; f θ is a trainable prediction model, and its parameter set is θ, which supports multiple modeling methods including linear regression, principal component regression (PCR), autoregressive moving average (ARMA), long short-term memory network (LSTM), etc.; is the input historical time series tensor, which is derived from the previous stage; f θ (·) is a forward prediction function.

[0054] To enhance the model's understanding of local temporal dynamics, the present embodiment preferably applies a temporal trend convolution mechanism on the input sequence. Specifically, for sequence χ s,t-w+1:t,c Applying a time dimension convolution operation: where K(τ) is a time domain convolution kernel function, used to strengthen the short-term trend and mutation signals in the input sequence; is the historical observation value of the original input tensor at time t-τ, spatial location s, and category c, reflecting the running state of the building system at a certain time point in the past; τ is the time delay index within the sliding window; X~X~ indicates the input tensor after trend enhancement, which helps to improve the response ability of the prediction model to the time pattern.

[0055] After completing the prediction process, the present embodiment further introduces a policy variable intervention mechanism to simulate future data under different running assumptions or control scenarios. This mechanism supports users to set a policy variable set and inject it into the prediction model in the form of parameters to adjust its internal weights or feature composition.

[0056] The intervention methods of policy variables include but are not limited to: Disturb the initial state or input sequence of the model; Adjust the output layer threshold or transformation function parameters of the model; Insert a control factor vector as an additional input dimension, such as a climate adjustment factor, an electricity use policy factor, a ventilation strategy factor, etc.

[0057] Specifically, let there be a policy control function Then the extended form of the prediction function is: where, is the simulated prediction value for spatial location s and data category c at time t+1 under the condition of policy variable , used to evaluate the running state after policy intervention; g φ (·,·) is the policy response function, whose internal parameters are φ, used to fuse the model prediction results and policy variables for output policy response results; is the policy variable set, including the intervention control parameters selected by the current user or recommended by the system, such as ventilation intensity adjustment value, personnel flow restriction parameter, equipment start-stop state, etc. This set is used to simulate the building running situation under different management strategies; φ is the set of trainable or rule-defined parameters in the policy simulation model, used to control the influence strength and direction of policy variables on the prediction results.

[0058] To support multi-step continuous prediction, the embodiment further constructs a rolling window iteration mechanism. After generating the prediction at time t+1, the system automatically updates the time window and iteratively generates the simulation tensor at subsequent time steps: wherein, is the prediction result of the model at time step t+2; is the new sliding window input sequence; until the future simulation sequence of the specified length is generated wherein, h is the simulation step; is the first step result in the prediction sequence; is the h-step prediction result generated under the recursive prediction mechanism.

[0059] The simulation tensor is finally output in a unified data interface format and passed to the aforementioned visualization mapping and three-dimensional projection module, supporting users to interactively explore, analyze and compare the future state trends.

[0060] In addition, the embodiment preferably designs a configurable model selection mechanism to support users to select appropriate prediction function types according to different scenario requirements. The system can automatically recommend the optimal model type and window parameters based on the historical data error performance at the initial stage of operation, improving the prediction reliability and adaptability.

[0061] To improve the running stability, the module adopts a sliding calibration mechanism during the prediction process, i.e., after generating the prediction result each time, if the user provides the posterior true observation tensor, it can be used to fine-tune and update the model, gradually improving the future deduction accuracy.

[0062] The embodiment supports the export and batch comparison of simulation results, users can set multiple strategy variable combinations to generate simulation tensors respectively, and perform horizontal comparison analysis on their key indicators to assist in operational decision-making and strategy optimization selection.

[0063] The interactive control and feedback linkage module is used to manage user interaction behavior on the building three-dimensional model, schedule the visualization content, and monitor whether various indicators in the tensor data exceed the set threshold. If it exceeds, the simulation prediction module is triggered to generate strategy response data, which is fed back to the visualization mapping and three-dimensional projection module for dynamic update display.

[0064] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A physical building data visualization and simulation analysis system, characterized in that, The system comprises: a building three-dimensional model and data acquisition module for constructing a three-dimensional geometric structure model of a building and deploying multiple types of sensors based on the model to collect dynamic monitoring data covering spatial positions, time points, and data category dimensions; a multi-dimensional data fusion and tensor construction module for receiving the dynamic monitoring data, standardizing and preprocessing the dynamic monitoring data, and fusing and constructing a unified tensor data structure according to spatial, temporal, and category three-dimensional indexes; a visualization mapping and three-dimensional projection module for receiving the three-dimensional data tensor output by the tensor construction module and mapping it into the building three-dimensional model to realize visualization rendering of the data state of each region in the building space; a simulation prediction and strategy generation module for performing historical time series modeling and sliding window prediction based on the tensor data and generating corresponding future simulation data tensors based on input strategy variables; an interactive control and feedback linkage module for managing user interaction behavior on the building three-dimensional model, scheduling visualization content, and monitoring whether each type of index in the tensor data exceeds a set threshold, and if so, triggering the simulation prediction module to generate strategy response data and feeding it back to the visualization mapping and three-dimensional projection module for dynamic update display.

2. The physical building data visualization and simulation analysis system of claim 1, wherein, The building three-dimensional model and data acquisition module comprises the following steps: constructing a three-dimensional geometric structure model of a building based on building information modeling, laser radar scanning, or unmanned aerial vehicle image reconstruction; relying on the three-dimensional model, deploying multiple types of sensors such as temperature and humidity, energy consumption, people flow, and air quality in key areas inside and outside the building; associating and processing the collected data according to its spatial position, collection time, and data type to construct an original multi-source dynamic monitoring data set.

3. The physical building data visualization and simulation analysis system of claim 1, wherein, The multi-dimensional data fusion and tensor construction module comprises the following steps: receiving the original data set from the building three-dimensional model and data acquisition module, and performing format unification, time alignment, missing value filling, and outlier removal operations on the data; After the preprocessing is completed, the data is organized and constructed into a tensor structure according to three-dimensional indexes of spatial position, time point and data category wherein, is the constructed three-dimensional tensor structure; m represents the spatial dimension; n represents the time dimension; and p represents the data category dimension.

4. The physical building data visualization and simulation analysis system of claim 3, wherein, The multi-dimensional data fusion and tensor construction module further comprises: for slices with missing values in the tensor structure, data completion is performed using low-rank matrix decomposition, specifically: The tensor under the fixed category dimension is sliced and expanded into a matrix X, and the following is solved and where U is the left singular matrix of the low-rank matrix decomposition; V is the right singular matrix of the low-rank matrix decomposition; ||·||F F is the Frobenius norm; realize the approximation filling of missing data, reduce the overall reconstruction error.

5. The physical building data visualization and simulation analysis system of claim 1, wherein, The visualization mapping and three-dimensional projection module comprises the following steps: Receiving the constructed tensor data Each tensor value is converted to a corresponding visual attribute, including color gradient, geometric deformation, or particle path, by a pre-set mapping function f(·). mapping and matching according to the spatial index s in the tensor and the spatial position M(x, y, z) in the three-dimensional building three-dimensional model, embedding visual attributes into the model structure, and completing the binding of data and space; mapping tensor slices at different time points to different time frames to realize time sequence animation visual display of the building space state.

6. The physical building data visualization and simulation analysis system of claim 5, wherein, The visualization mapping and three-dimensional projection module further comprises: providing a user interaction interface to allow users to control the observation angle on the three-dimensional model, select specific data categories for display, and visualize and update the tensor slices at any time point through a time slider.

7. The physical building data visualization and simulation analysis system of claim 1, wherein, The simulation prediction and strategy generation module comprises the following steps: Based on the tensor The sliding window width w is set, and a time sequence sub-tensor with a length of w is extracted. inputting the time series into a regression function φ(·) to predict the tensor value at the next time point and generate a predicted value: wherein, is the tensor value after time trend convolution processing at time t; s is a spatial index, indicating a specific unit in the building space; t is a time index, representing the current time step; c is a data category index; φ(·) is an input regression function; w is a sliding window width; Receiving a set of policy variables P = {pi, p2,..., p q} set by a user, adjusting parameters in a prediction process, generating a simulated scenario tensor Output analog tensor for the visualization module to present the simulation results.

8. The physical building data visualization and simulation analysis system of claim 7, wherein, The prediction function f θ As a trainable model, it takes a sequence of historical tensors as input and supports implementation using any of the following algorithms: linear regression, long short-term memory network, or autoregressive model.

9. The physical building data visualization and simulation analysis system of claim 1, wherein, The interactive control and feedback linkage module comprises the following steps: Setting the pre-warning threshold θ of different categories of data c comparing the predicted value of the next time in the tensor data; If the following condition is met then a warning mechanism is triggered. After the early warning trigger, the simulation prediction and strategy generation module is called in linkage, the simulation data tensor under the scene of coping with the strategy variable set is regenerated, and the result is fed back to the visualization module for real-time display.

10. The physical building data visualization and simulation analysis system of claim 9, wherein, The interactive control and feedback linkage module further comprises: A strategy switching console is provided to allow users to switch between multiple sets of strategy variable scenes and dynamically call the prediction engine to recalculate the simulation data; The original prediction and the visualization results after the strategy response are displayed in the console at the same time, and export of a comparative analysis report is supported for subsequent decision support.

Images