Property management prediction system based on BIM and digital twinborn and prediction method thereof

Abstract

The invention relates to a property management prediction system based on BIM (Building Information Modeling) and digital twinning and a prediction method thereof, which adopt mutual cooperation of a data acquisition module, a central processing unit and an AI processing unit and utilize a digital twinning engine to finally present a visual prediction result. The method has accurate prediction capability and can realize intelligent decision support. Moreover, during use, closed-loop optimization of data can be realized.

Description

Technical Field

[0001] The present invention relates to a property management prediction system and a prediction method thereof, and in particular to a property management prediction system and a prediction method thereof based on BIM and digital twins. Background Art

[0002] In the field of commercial real estate operations, traditional property management forecasting systems generally rely on a single data source (such as financial statements or historical rental data), which makes it difficult to accurately capture the multi-dimensional factors affecting tenant operating risks. Specifically, the existing technology has the following significant defects: 1. Data fragmentation and insufficient prediction accuracy. Existing solutions, such as the Chinese patent CN117951920A, disclose a method for building a digital twin platform based on BIM model simulation. It usually uses independent analysis models to process IoT data, video data and financial data, and lacks a multimodal fusion mechanism. For example: Data limitations: The linear regression model that only relies on financial statements cannot integrate spatial characteristics (such as tenant location accessibility) and behavioral characteristics (such as customer stay time), resulting in a risk prediction error rate of up to 20%-30%; Static strategy defects: Traditional rental strategies are based on fixed-cycle adjustments (such as annual evaluations) and cannot respond to market fluctuations in real time, resulting in property vacancy rates that remain at 8%-12% for a long time.

[0003] 2. Decision support tools are backward. Existing visualization systems, such as China's patent CN112883240A, a data-lightweight BIM building model management method based on digital twins and its management system, use two-dimensional charts to display data and lack dynamic interaction capabilities in the time and space dimensions. Due to information lag, managers cannot replay historical operating data and superimpose and compare predicted results on a timeline. Decisions rely on manual experience, and the response cycle exceeds 48 hours. Inefficient investment matching: Traditional investment matching systems rely on manual screening (matching degree <60%) and cannot analyze tenant business complementarity through AI. The average period for new tenants to move in is 3-6 months.

[0004] 3. Rigid model update mechanism. Existing technologies, such as the intelligent operation and maintenance management system and method based on BIM and digital twin technology in Chinese patent CN118761762A, mainly adopt offline batch training mode, and the model update cycle is as long as 1-3 months, which cannot adapt to sudden changes in the market.

[0005] At the same time, the data involved in the existing processing methods lack a closed loop, model iteration relies on manual intervention, and the optimization efficiency is low.

[0006] In view of the above-mentioned defects, the designer actively conducts research and innovation in order to create a property management forecasting system and its forecasting method based on BIM and digital twins, so as to make it more valuable for industrial use. Summary of the invention

[0007] In order to solve the above technical problems, the purpose of the present invention is to provide a property management prediction system and prediction method based on BIM and digital twins.

[0008] The property management prediction system based on BIM and digital twins of the present invention includes: a data acquisition module for real-time acquisition of multi-source data, including IoT device data, monitoring AI data, tenant public financial report data and BIM model data; a central processing unit for preprocessing data and constructing a data cube containing time, space and tenant ID dimensions, wherein the central processing unit eliminates noise carried during data transmission through a median filtering algorithm; an AI processing unit for extracting traffic and customer behavior characteristics from monitoring videos; a digital twin engine integrating a multimodal fusion model for generating tenant management risk scores and revenue forecast results; and a user interface module for visually displaying forecast results and dynamic rent adjustment strategies. In this way, tenant management risk scores and revenue forecast results can be generated by dynamically associating spatial topological data in the BIM model with tenant behavior time series data. The dynamic association includes constructing an adjacency matrix based on tenant spatial proximity relationships and extracting spatial weight features through a graph neural network (GCN).

[0009] Furthermore, in the above-mentioned property management prediction system based on BIM and digital twin, the data acquisition module includes: an IoT sensor connected to the central processing unit through a preset protocol, the IoT sensor adopts a cluster routing algorithm, and the cluster head selection rule is RSSI threshold = -110dBm. The surveillance camera transmits the video stream to the AI ​​processing unit through the RTSP protocol, and processes it in combination with the YOLO algorithm. The YOLO algorithm is at least a YOLOv5 algorithm, and its confidence threshold = 0.7. The anchor frame size of the YOLOv5 algorithm is preset to one of 10×10, 16×16, and 32×32. During implementation, anchor frame combinations of different sizes and proportions such as (10, 13), (16, 30), (33, 23), (30, 61), (62, 45), (59, 119), (116, 90), (156, 198), and (373, 326) can also be used.

[0010] Web crawlers are used to regularly capture tenants' public financial report data and store them in the BIM model database. During this period, most IoT sensor-like devices are based on the physical layer protocols RS-485 and RS-232. At the same time, the internal management system of the hardware involved is generally a private protocol set by the manufacturer. In order to facilitate docking with this system, the Modbus protocol is used for the fire control system; the BACnet protocol is used for HVAC, lighting, and energy management; the parking management system; and the http protocol is used for the video security system.

[0011] Furthermore, in the above-mentioned property management prediction system based on BIM and digital twins, the multimodal fusion model is a Transformer architecture, which is used to fuse time series data, spatial data and image data; the number of encoder layers of the Transformer architecture = 12, the number of multi-head attention heads = 8, the time series data is smoothed, the corresponding processing window is 7 days, and the spatial data is based on the three-dimensional path planning result of the BIM model.

[0012] Furthermore, in the above-mentioned property management prediction system based on BIM and digital twins, the multimodal fusion model is a hybrid architecture of Transformer, LSTM, and GCN, and the adjacency matrix construction rule constructed by the graph neural network model is the spatial proximity relationship of tenants.

[0013] Furthermore, in the above-mentioned property operation prediction system based on BIM and digital twins, the digital twin engine also includes a reinforcement learning module for generating dynamic rental strategies based on the PPO algorithm; a three-dimensional visualization module for risk warning prompts, and superimposing an operation heat map in the BIM model.

[0014] The property management prediction method based on BIM and digital twin of the present invention comprises the following steps: Collect multi-source data, including IoT device data, monitoring AI data, tenant public financial report data and BIM model data, and construct a data cube with time, space and tenant ID dimensions. Extract spatial features, behavioral features and financial features, and output tenant operating risk scores and revenue forecasts for the next three months. Generate dynamic rent adjustment strategies based on the forecast results and visualize them through the digital twin cockpit. The spatial features extract tenant location weights based on the BIM model; the behavioral features extract customer density and consumption behavior patterns from monitoring AI data; the financial features combine financial report data to calculate indicators such as tenant debt repayment ability and cash flow stability.

[0015] Furthermore, in the above-mentioned property management prediction method based on BIM and digital twins, the training step of the multimodal fusion model includes: using the Transformer architecture to fuse time series data, spatial data and image data. Combining the XGBoost model to output the tenant management risk score, the score range is 0-100 points.

[0016] Furthermore, in the above-mentioned property management prediction method based on BIM and digital twins, the dynamic rent adjustment strategy includes, when the risk score is greater than 70 points, it is recommended to reduce the rent or introduce complementary business formats; when the risk score is less than or equal to 70 points, it is recommended to maintain or increase the rent.

[0017] Furthermore, the above-mentioned property operation prediction method based on BIM and digital twins, wherein the method also includes an online learning mechanism with a sliding window size of 30 days, and updates the model parameters through real-time feedback data to adapt to market changes and fluctuations in tenant behavior.

[0018] Furthermore, in the above-mentioned property management prediction method based on BIM and digital twins, the BIM model integrates fire evacuation simulation data and adopts a hybrid algorithm. First, a static global escape route is generated using a conventional path algorithm and a floor plan. Every 30 seconds, the fire source location and the danger zone are updated in real time based on the path obstruction caused by the dynamic spread of the fire obtained by the sensor; then the D Lite algorithm is used to update the affected area; the parameters of the conventional path algorithm are: the heuristic function is Manhattan distance, the node expansion strategy is 8 fields, the cost function is the Euclidean distance between nodes, and the weight coefficient is 1; the parameters of the D Lite algorithm are consistent with the conventional path algorithm, and the incremental update threshold can be dynamically adjusted according to the sensor data and coordinates.

[0019] By means of the above scheme, the present invention has at least the following advantages: 1. Accurate prediction capability: Through multi-modal data fusion, the accuracy of tenant business risk prediction is improved, and dynamic rental strategies are used to reduce property vacancy rates.

[0020] 2. Intelligent decision support. The digital twin-based visualization cockpit supports playback of historical data by timeline and superimposition of forecast results, enabling property management personnel to view tenant operating status in real time and shorten decision-making cycles. In addition, through the AI-driven investment recommendation system, new tenants with a matching degree greater than 90% are recommended to improve investment efficiency.

[0021] 3. Data closed-loop optimization. The model is updated in real time through online learning to adapt to market changes and fluctuations in tenant behavior.

[0022] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of the implementation steps of the property management forecasting method based on BIM and digital twins. DETAILED DESCRIPTION

[0024] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0025] like Figure 1 The property management prediction system based on BIM and digital twins is unique in that it includes: The data collection module is used to collect multi-source data in real time. The multi-source data used includes IoT device data, monitoring AI data, tenant public financial report data and BIM model data.

[0026] The central processing unit is used to pre-process the data and construct a data cube containing time, space and tenant ID dimensions. The central processing unit uses the median filter algorithm to eliminate the noise carried during the data transmission process. During the implementation, other data cleaning methods can also be used. For example, the deletion method is used globally to remove abnormal data. Or, the isolation forest is used globally to detect outliers. Or, hashing is used globally to deduplicate data. In this way, the median filter algorithm can be used for time series data such as energy consumption, temperature and humidity that may have instantaneous noise; regular expressions can be used to extract and standardize text data for financial report data; and exponential smoothing can be used for short-term fluctuating data such as traffic.

[0027] AI processing unit, used to extract traffic and customer behavior characteristics from surveillance videos; Digital twin engine, integrating multimodal fusion models to generate tenant operating risk scores and revenue forecast results; The user interface module is used to visualize the prediction results and dynamic rent adjustment strategies. Specifically, the system constructed by the present invention can be built based on game engines such as UE, and its BIM model is manually finely modeled to meet the digital twin standard classification level L3. Compared with traditional BIM models, this system can access real-time data streams and run advanced analysis functions through TCP / IP. At the same time, it can provide functions such as first-person roaming to achieve rich dynamic interactive logic. In addition, the ray tracing technology of the game engine can be used to achieve realistic light and shadow, material texture and environmental details. In this way, with the help of the cross-platform capabilities of the game engine, multi-terminal applications can be achieved after lightweighting.

[0028] In combination with a preferred embodiment of the present invention, the data acquisition module used includes: an IoT sensor, which is connected to the central processing unit through a preset protocol. At the same time, the IoT sensor adopts a cluster routing algorithm, and the cluster head selection rule is RSSI threshold = -110dBm. During the implementation period, the data acquisition module used is compatible with a variety of different protocols, such as using the Modbus protocol to participate in the application of fire control systems, using the BACnet protocol to participate in HVAC, lighting, and energy management, and using the http protocol to participate in parking management systems and video security systems. In this way, different application scenarios can be met. During this period, the IoT sensor only needs to meet the physical layer protocol RS-485 and RS-232 communication, which is easy to deploy. In addition, the surveillance camera can be used to transmit the video stream to the AI ​​processing unit through the RTSP protocol, and combined with the YOLO algorithm for processing. During the implementation period, the YOLO algorithm is at least the YOLOv5 algorithm, and its confidence threshold = 0.7, and the anchor frame size of the YOLOv5 algorithm is preset to one of 10×10, 16×16, and 32×32. In view of the conventional warning situation, the YOLOv11 target detection algorithm can be used, and the model can be trained with the v11m pre-trained model and its own data, with a confidence threshold of 0.7. By setting up multi-threading, calling multiple thread algorithms, and setting the algorithm startup interval, it is used to avoid loading multiple models at the same time, which causes a lot of time to compete for resources. In this way, the training needs of the five algorithms of dangerous personnel warning, fireworks warning, knife, fork and stick warning, personnel gathering warning and personnel falling warning can be met.

[0029] At the same time, the financial report data disclosed by tenants can be regularly captured through web crawlers and stored in the BIM model database. Specifically, the financial report data is divided into two parts. One is unstructured data, which is mostly crawled PDF, HTML, etc., and is directly stored using the local file system. The other is structured data, one part of which is obtained by structuring the unstructured data, and the other part is the structured data directly crawled, which is stored in a relational database with clear fields. It should be noted that this system strictly abides by the Robots protocol and access frequency restrictions of the target website through web crawler technology to ensure the legality and compliance of data crawling behavior. The captured financial report data is only used for tenant business forecasting analysis. The system adopts encrypted storage and access control mechanisms to ensure data security and privacy. The crawler module adopts a distributed architecture, and avoids excessive access pressure on the target website through dynamic IP proxy and request delay mechanism.

[0030] Further, the multimodal fusion model adopted by the present invention is the Transformer architecture, which is used to fuse time series data, spatial data and image data. The Transformer architecture uses 12 encoder layers and 8 multi-head attention heads to smooth the time series data, and the corresponding processing window is 7 days. The spatial data used is based on the three-dimensional path planning results of the BIM model. At the same time, the multimodal fusion model is a hybrid architecture of Transformer, LSTM, and GCN, and the adjacency matrix construction rule constructed by the graph neural network model is the tenant space proximity relationship.

[0031] Specifically, a multimodal fusion strategy combining the Transformer architecture and LSTM is adopted, including: time series processing module, space processing module, image processing module and fusion module. The time series processing module uses LSTM to extract time series features (such as energy consumption, temperature and humidity); the space processing module builds a tenant space adjacency graph based on the BIM model, and uses GCN to extract spatial features (such as location weight and proximity relationship); the image processing module uses the ResNet-50 pre-trained model to extract image features (such as traffic flow and customer behavior patterns); the Transformer encoder fuses time series, space and image features to output the tenant business risk score.

[0032] The data processing method is as follows: The source of time series data is IOT devices, and a 7-day sliding window is used to calculate the mean, eliminate short-term fluctuations, and meet Z-score standardization. The source of spatial data is the BIM model. According to the tenant space distance (such as ≤50 meters), an adjacency matrix is ​​constructed to represent the spatial proximity relationship. The location weight of the tenant is extracted based on the BIM model (such as the weight close to the entrance and exit = 0.9). At the same time, the image data comes from monitoring, and YOLOv11 is used to count the flow of people and the length of customer stay. The detection results can be input into ResNet-50 to extract the image feature vector. The time series processing module is composed of 2 layers of LSTM, with the number of hidden units in each layer = 64, and Dropout = 0.2. The spatial processing module is a 2-layer graph convolutional network with the number of hidden units = 32. The image processing module ResNet-50 (pre-trained weights), remove the fully connected layer, and extract the features of the second to last layer. The number of layers in the fusion module = 12, and the number of heads = 8.

[0033] In practice, the digital twin engine also includes: a reinforcement learning module that generates a dynamic rental strategy based on the PPO algorithm. For ease of implementation, the reinforcement learning module can be adjusted and trained in accordance with the following table to meet the adjustment of the dynamic rental strategy.

[0034]

[0035] During the implementation period, the three-dimensional visualization module can be used to provide risk warning prompts and overlay the business heat map in the BIM model. Specifically, the spatial coordinates of each tenant in the BIM model are extracted, and the business indicators of each tenant (such as traffic, revenue, and risk score) are obtained, and the business indicators are normalized using Min-Max. After that, the normalized value is used as the thermal value, and a smooth thermal distribution is generated using the Gaussian kernel density estimation algorithm, and the thermal value is mapped to a color gradient from blue to red. In this way, a clear display of the business heat map can be achieved.

[0036] In order to better implement the present invention, a property management prediction method based on BIM and digital twin is first provided, which includes the following steps: First, use the aforementioned protocols to connect with the corresponding hardware and platforms to collect multi-source data, including IoT device data, monitoring AI data, tenant public financial report data and BIM model data. After that, build a data cube containing time, space and tenant ID dimensions. For the convenience of construction, the time dimension is the timestamp divided by day, week, month, etc., such as 2025-03-01; the spatial dimension is the tenant XYZ coordinates, such as (120.1, 112.2, 6.6); the tenant ID is the tenant ID defined by the system, and the tenant attributes are associated. Then, extract the spatial characteristics, behavioral characteristics and financial characteristics, and output the tenant's operating risk score and the revenue forecast results for the next three months. Then, generate a dynamic rent adjustment strategy based on the forecast results, and call the three-dimensional visualization module through the digital twin cockpit to realize the visualization display. During the implementation period, spatial characteristics, extract tenant location weights based on the BIM model, such as proximity to entrances and exits, floor height, etc. Behavioral characteristics, extract customer density, consumption behavior patterns and other data from monitoring AI data, and can be optimized for customers' high-frequency visit periods. Financial characteristics, combined with financial report data to calculate indicators such as tenants' debt repayment ability and cash flow stability.

[0037] During the implementation, the training steps of the multimodal fusion model used include the following methods: using the Transformer architecture to fuse time series data, spatial data, and image data. After that, the XGBoost model is combined to output the tenant's business risk score, with a score range of 0-100. In order to facilitate the acquisition of a more appropriate score, the sample data provides the risk score data of historical tenants for customers, and the feature data is the tenant's business indicators. The mean, standard deviation, skewness, kurtosis and other statistics of the risk score are calculated respectively. After that, the risk score is Z-score standardized, and the silhouette coefficient (Silhouette Score) is used to determine the optimal number of clusters. At the same time, the K-means algorithm is used to cluster the risk score, and the risk level is divided according to the cluster center. In this way, it can automatically test whether the means of different risk levels are significantly different. Furthermore, in order to improve the scoring angle, the classification threshold can be manually adjusted according to other indicators such as the market environment, tenant feedback, and customer wishes.

[0038] The dynamic rent adjustment strategies adopted include: when the risk score is greater than 70 points, it is recommended to reduce the rent or introduce complementary formats. The upper limit of rent reduction = 15% of the contract amount. It can be fine-tuned according to the business rules of the area. When the risk score is less than or equal to 70 points, it is recommended to maintain or increase the rent. The increase in rent = industry average rent × risk score / 100. Specifically, the following risk classification can be used: 0-30 is low risk, the rent can be increased, and the increase = industry average rent × risk score / 100. Upper limit = 10% of the contract amount. 31-70 is medium risk, the rent is maintained unchanged, or the rent is slightly increased (the increase is ≤ 5% of the contract amount). 71-100 is high risk, the rent can be reduced, the upper limit of the reduction = 15% of the contract amount, or complementary formats are introduced (such as catering tenants introducing retail formats). In addition, third-party data sources (such as market research reports) can be used to obtain the industry average rent, and the industry average rent data can be updated regularly to ensure the timeliness of the strategy.

[0039] In order to adjust the parameters of the model in time according to different environmental changes, the present invention may also include an online learning mechanism during implementation, using a sliding window size of 30 days. In this way, the model parameters are updated through real-time feedback data to adapt to market changes and fluctuations in tenant behavior. Specifically, the step size corresponding to the sliding window is once a day, and the data of the last 30 days is used for training each time it is updated. At the same time, the data update cycle is once a day, and it is updated every day at UTC+82:00AM. It mainly adds the operating data of the day. On specific dates, newly acquired monthly and annual data will be added, and old data from 30 days ago will be removed. The model parameters can be incrementally updated based on the data in the sliding window.

[0040] Considering the configuration needs of fire safety, the BIM model can be used to integrate fire evacuation simulation data. Specifically, this system adopts a hybrid algorithm. First, the conventional path algorithm and floor plan are used to generate a static global escape route. Every 30 seconds, the fire source location and danger zone are updated in real time based on the path obstruction caused by the dynamic spread of the fire obtained by the sensor. After that, the D Lite algorithm is used to update the affected area to adapt to the dynamically changing fire environment at a faster speed. The parameters of the conventional path algorithm are: the heuristic function is Manhattan distance, the node expansion strategy is 8 fields, the cost function is the Euclidean distance between nodes, and the weight coefficient is 1. The parameters of the D Lite algorithm are consistent with the conventional path algorithm, and the incremental update threshold can be dynamically adjusted according to the sensor data and coordinates.

[0041] The working principle of the present invention is as follows: First, the deployment location and key node definition, based on the spatial topology data of the BIM model, sensors are deployed at the following key nodes: Tenant shops: Install temperature and humidity sensors (model SHT35) at the entrance of each tenant shop (such as the front door of a retail store) to monitor the environmental data of customer flow routes. They can also be deployed in building lobbies, elevator waiting areas, fire escapes and rest areas.

[0042] In energy-intensive areas, energy consumption meters are installed at key energy consumption nodes such as power distribution rooms and air-conditioning rooms (smart telegraphs of model HPLC-DTS345 can be used).

[0043] After that, deploy the camera: For tenant stores, cameras are installed at the cashier counter and main shelf areas to monitor customers’ stay and product attention. For public areas, cameras are deployed at elevator entrances, stairwells and main passages to cover all entrances and exits.

[0044] Next, the multimodal fusion model training process is used to optimize spatial feature calculation. Weights can be assigned to key nodes. In other words, different weights are assigned to sensor data from tenant stores and public areas based on the importance of the deployment location. For reference, the store weight = 0.7 (main business area) and the public area weight = 0.3.

[0045] Then, the heat map rendering is performed. For tenant store priority, the color saturation of the heat map of the tenant store area can be increased by 20% in the digital twin cockpit to highlight its operating status. For public area warning signs, when the flow of people in the public area exceeds the safety threshold (such as >500 people / hour), a red flashing warning box is superimposed in the BIM model.

[0046] It can be seen from the above textual description and the accompanying drawings that the present invention has the following advantages: 1. Accurate prediction capability: Through multi-modal data fusion, the accuracy of tenant business risk prediction is improved, and dynamic rental strategies are used to reduce property vacancy rates.

[0047] 2. Intelligent decision support. The digital twin-based visualization cockpit supports playback of historical data by timeline and superimposition of forecast results, enabling property management personnel to view tenant operating status in real time and shorten decision-making cycles. In addition, through the AI-driven investment recommendation system, new tenants with a matching degree greater than 90% are recommended to improve investment efficiency.

[0048] 3. Data closed-loop optimization. The model is updated in real time through online learning to adapt to market changes and fluctuations in tenant behavior.

[0049] In addition, the indicated orientations or positional relationships described in the present invention are all based on the orientations or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the device or structure referred to must have a specific orientation or be operated with a specific orientation structure. Therefore, they cannot be understood as a limitation on the present invention.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. The property management prediction system based on BIM and digital twins is characterized by include: Data collection module, used to collect multi-source data in real time, including IoT device data, monitoring AI data, tenant public financial report data and BIM model data; A central processing unit, used to pre-process the data and construct a data cube containing time, space and tenant ID dimensions, wherein the central processing unit eliminates noise carried during data transmission by a median filtering algorithm; AI processing unit, used to extract traffic and customer behavior characteristics from surveillance videos; Digital twin engine, integrating multimodal fusion models to generate tenant operating risk scores and revenue forecast results; User interface module, used to visualize forecast results and dynamic rent adjustment strategies.

2. The property management prediction system based on BIM and digital twin according to claim 1 is characterized by: The data acquisition module includes: The IoT sensor is connected to the central processing unit through a preset protocol. The IoT sensor adopts a cluster routing algorithm, and the cluster head selection rule is RSSI threshold = -110dBm; The surveillance camera transmits the video stream to the AI ​​processing unit through the RTSP protocol, and processes it in combination with the YOLO algorithm. The YOLO algorithm is at least a YOLOv5 algorithm, and its confidence threshold is 0.

7. The anchor frame size of the YOLOv5 algorithm is preset to one of 10×10, 16×16, and 32×32; Web crawlers regularly capture tenants’ public financial report data and store them in the BIM model database.

3. The property management prediction system based on BIM and digital twin according to claim 1 is characterized by: The multimodal fusion model is a Transformer architecture, which is used to fuse time series data, spatial data and image data; the number of encoder layers of the Transformer architecture is 12, the number of multi-head attention heads is 8, the time series data is smoothed, and the corresponding processing window is 7 days. The spatial data is based on the three-dimensional path planning result of the BIM model.

4. The system according to claim 3, characterized in that: The multimodal fusion model is a hybrid architecture of Transformer, LSTM, and GCN, and the adjacency matrix construction rule constructed by the graph neural network model is the tenant space proximity relationship.

5. The property management prediction system based on BIM and digital twin according to claim 1 is characterized by: The digital twin engine also includes a reinforcement learning module that generates a dynamic rental strategy based on the PPO algorithm; The 3D visualization module is used for risk warning and overlaying the operation heat map in the BIM model.

6. Property management prediction method based on BIM and digital twins, characterized by The following steps are involved: Collect multi-source data, including IoT device data, monitoring AI data, tenant public financial report data and BIM model data, and build a data cube with time, space and tenant ID dimensions; Extract spatial characteristics, behavioral characteristics and financial characteristics, and output tenant business risk scores and revenue forecasts for the next three months; Generate dynamic rent adjustment strategies based on prediction results and visualize them through the digital twin cockpit; The spatial features extract the tenant location weight based on the BIM model; the behavioral features extract customer density and consumption behavior patterns from the monitoring AI data; the financial features calculate indicators such as the tenant's debt repayment ability and cash flow stability in combination with the financial report data.

7. The property management prediction method based on BIM and digital twin according to claim 6 is characterized by: The training steps of the multimodal fusion model include: Use the Transformer architecture to fuse time series data, spatial data, and image data; Combined with the XGBoost model, the tenant operating risk score is output, with a score range of 0-100.

8. The property management prediction method based on BIM and digital twin according to claim 6 is characterized by: The dynamic rent adjustment strategy includes: when the risk score is greater than 70 points, it is recommended to reduce the rent or introduce complementary business formats; when the risk score is less than or equal to 70 points, it is recommended to maintain or increase the rent.

9. The property management prediction method based on BIM and digital twin according to claim 6 is characterized by: The method also includes an online learning mechanism with a sliding window size of 30 days, which updates model parameters through real-time feedback data to adapt to market changes and fluctuations in tenant behavior.

10. The property management prediction method based on BIM and digital twin according to claim 6, characterized in that: The BIM model integrates fire evacuation simulation data and adopts a hybrid algorithm. First, a static global escape route is generated using a conventional path algorithm and a floor plan. Every 30 seconds, the fire source location and the danger zone are updated in real time based on the path obstruction caused by the dynamic spread of the fire obtained by the sensor. Then, the D Lite algorithm is used to update the affected area. The parameters of the conventional path algorithm are: the heuristic function is Manhattan distance, the node expansion strategy is 8 fields, the cost function is the Euclidean distance between nodes, and the weight coefficient is 1. The parameters of the D Lite algorithm are consistent with those of the conventional path algorithm, and the incremental update threshold can be dynamically adjusted according to the sensor data and coordinates.

Images