BIM (Building Information Modeling)-based intelligent regulation and control method for energy-saving transformation of existing building
By constructing a BIM-based digital twin and fusing real-time data, the problem of model-data disconnect in building energy-saving renovation has been solved, achieving a dynamic balance between building energy efficiency and comfort, and improving adaptive control capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA CONSTRUCTION SIXTH ENGINEERING DIVISION CO LTD
- Filing Date
- 2025-11-26
- Publication Date
- 2026-04-21
AI Technical Summary
In existing building energy-saving renovations, BIM models are disconnected from real-time data, control strategies are rigid, and there is a lack of accurate prediction and forward-looking response to the building's future load, outdoor environment, and human comfort. This makes it difficult to balance energy consumption and comfort, and information silos exist between systems, making it impossible to achieve continuous optimization and adaptive control of the building's overall energy efficiency.
By constructing a BIM building information model based on 3D laser scanning technology, integrating IoT sensor data to generate a digital twin, monitoring indoor environmental parameters in real time, establishing a dynamic range prediction model, predicting future loads, generating optimal operating strategies for energy-consuming equipment, and achieving adaptive control through a BMS system.
It achieves deep integration of physical buildings and virtual models, accurately generates equipment strategies that balance energy conservation and comfort, improves building energy efficiency and indoor environmental quality, and provides a data-driven intelligent decision-making integrated solution.
Smart Images
Figure CN121900144A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent building temperature control technology, specifically to an intelligent control method for energy-saving retrofitting of existing buildings based on BIM. Background Technology
[0002] In the current energy-saving renovation of existing buildings, traditional methods mostly rely on independent BIM static models or simple building automation systems, which have significant limitations. The models are disconnected from real-time data. BIM models are mostly static geometric representations at the completion stage and cannot be linked with IoT sensor data and equipment status during operation, resulting in a disconnect between static models and dynamic operation. The control strategies are rigid. Existing systems mostly adopt start-stop control based on fixed thresholds or preset schedules, lacking accurate prediction and forward-looking response to the dynamic needs of future building load, outdoor environment, and human comfort, making it difficult to balance energy consumption and comfort. Information silos are formed between systems. There is a lack of effective data fusion and intelligent decision-making loops between 3D scanning, BIM platforms and building automation systems, which keeps energy-saving control at a local, passive and experience-driven level, and cannot achieve continuous optimization and adaptive control of the overall building energy efficiency. Summary of the Invention
[0003] To address the aforementioned technical issues, this paper presents a BIM-based intelligent control method for energy-saving retrofitting of existing buildings, which solves the problems described above.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0005] A BIM-based intelligent control method for energy-saving retrofitting of existing buildings, comprising:
[0006] S1. Based on 3D laser scanning technology, acquire point cloud data of geometric coordinates of the target building, construct a BIM building information model, integrate IoT sensor data, and generate a digital twin of the target building;
[0007] S2. Based on multi-source heterogeneous sensors, monitor the indoor environmental parameters of the target building in real time, establish a dynamic range prediction model for indoor environmental comfort, couple external real-time meteorological data, and input the target building's digital twin to predict the dynamic load of the target building in future periods.
[0008] S3. Obtain historical energy consumption time series data of the target building, integrate dynamic load forecast for future periods, real-time external environmental parameters and dynamic range forecast of indoor environmental comfort, establish a collaborative optimization model for energy-consuming equipment of the target building, and generate the optimal operating strategy set for each energy-consuming equipment of the target building in the future period.
[0009] S4. Based on the generated optimal operating strategy set for each energy-consuming device in the target building for the future time period, link it to the BMS building equipment management system to realize adaptive control of building energy-consuming devices.
[0010] Preferably, step S1 specifically includes:
[0011] Based on 3D laser scanning technology, the original point cloud data of multiple stations of the target building is acquired in real time, and the original point cloud data is preprocessed.
[0012] Based on the acquisition of original point cloud data from multiple stations of the target building, the original point cloud data from multiple stations of the target building are initially aligned using the corners of the walls, columns, and building height of the target building as references, and the original point cloud data of the target building is divided into source point cloud and target point cloud respectively.
[0013] Using the source point cloud of the target building and the target point cloud as input, calculate the Euclidean distance from each data point in the source point cloud of the target building to the target point cloud, find the closest corresponding point in the Euclidean distance as the initial point pair, and remove invalid point pairs that are greater than the maximum Euclidean distance;
[0014] The optimal rotation and translation transformation matrix of the target building is obtained by calculating the minimum average sum of squared distances between all valid point pairs using the least squares method.
[0015] Based on the optimal rotation and translation transformation matrix of the target building, the source point cloud position is continuously updated iteratively until the maximum number of iterations is reached, thereby achieving automatic registration of the original point cloud data of multiple stations of the target building.
[0016] Preferably, step S1 further includes:
[0017] Based on the original point cloud data of multiple stations of the target building after automatic registration, the data is cut into several point cloud blocks of the target building with uniform size, and corresponding semantic labels are assigned using a labeling tool.
[0018] The corresponding semantic tags include: walls, floors, pipes, doors and windows, and furniture;
[0019] Using PyTorch as the framework, a PointNet++ model is constructed, which includes a Set Abstraction layer and a Feature Propagation layer to perform hierarchical feature learning on several target building point cloud blocks of uniform size.
[0020] For the Set Abstraction layer, M center points are selected from N point cloud data in several target building point cloud blocks of uniform size, and the farthest point of several target building point cloud blocks of uniform size is extracted.
[0021] Using each center point as the sphere center, a spherical region is constructed with a radius of r. The neighboring points are found, and the local regions of the neighboring points are grouped. Max pooling is introduced to aggregate the features of different numbers of points in several target building point cloud blocks of uniform size into fixed-dimensional local features. High-level local features of several target building point cloud blocks of uniform size are extracted.
[0022] For the Feature Propagation layer, the inverse distance weighted interpolation method is used to upsample the high-level local features extracted by the SetAbstraction layer, gradually restoring the spatial resolution of point cloud data in several target building point cloud blocks of uniform size, thereby realizing the global propagation and fusion of features of several target building point cloud blocks of uniform size.
[0023] An encoder is constructed by stacking three Set Abstraction layers, and a decoder is constructed by stacking corresponding number of FeaturePropagation layers. An encoder-decoder structure is established, and the cross-entropy loss function is used as the optimization objective. The PointNet++ model learning rate, batch size, network depth, and ball query radius hyperparameters are continuously adjusted through backpropagation to achieve semantic segmentation of target building point cloud data.
[0024] Based on the semantic segmentation of the target building point cloud data, the BIM building information software is imported, and all point clouds marked as walls, floors, pipes, doors, windows and furniture are selected respectively. The Revit wall, floor, pipe, door, window and furniture commands are used respectively to generate the target building wall, floor, pipe, door, window and furniture models according to the outline of the target building point cloud data.
[0025] Integrate the target building's wall, floor, pipe, door, window, and furniture models to create a BIM building information model;
[0026] Based on IoT sensor data, the target building data is acquired in real time, coupled with the BIM building information model, and the BIM building information model is converted into an efficient format using the Unity game engine to generate a digital twin of the target building.
[0027] Preferably, step S2 specifically includes:
[0028] Based on multi-source heterogeneous sensors and building automation systems, a fixed time interval of 5 minutes is set to monitor indoor environmental parameters of the target building in real time, acquire equipment operation data, and transmit the data to the central data server.
[0029] The indoor environmental parameters are: temperature, humidity, CO2 concentration, and light intensity.
[0030] Equipment operation data includes: air conditioning setpoint, valve opening, and fan status;
[0031] Based on the indoor environmental parameters of the target building, a time-series database of the indoor environment of the target building is constructed by aligning timestamps with unified data frequencies.
[0032] Using the PMV prediction average voting formula, the predicted human comfort values corresponding to historical and real-time environmental data are calculated.
[0033] Based on the predicted human comfort values corresponding to historical and real-time environmental data, the acceptable range of average votes for PMV prediction is defined.
[0034] The acceptable range for the average vote in the PMV prediction is: PMV ;
[0035] Using historical indoor environmental data of the target building, real-time indoor environmental parameters and time steps of the target building as inputs, and the upper and lower limits of comfort for each time step in the next 1-6 hours as prediction targets, a dynamic interval prediction model for indoor environmental comfort is established. The learning rate, batch size and network depth hyperparameters of the model are optimized and evaluated using cross-validation.
[0036] Preferably, step S2 further includes:
[0037] Real-time acquisition of outdoor weather data based on sensors;
[0038] The outdoor meteorological data includes: temperature, humidity, wind speed, and solar radiation;
[0039] Using the current real-time indoor environmental parameters of the target building as the initial state, and the predicted dynamic range of indoor environmental comfort and outdoor meteorological data as boundary conditions, the target building digital twin is associated and paired to predict the target building load at each time step in the next 24 hours.
[0040] Calculate the predicted values of cooling and heating loads for the target building to maintain the dynamic range of indoor environmental comfort, plot the dynamic load prediction curve for the target building in the future, and realize the dynamic load prediction for the target building in the future.
[0041] Preferably, step S3 specifically includes:
[0042] Acquire historical energy consumption data of the target building, preprocess the data, align the timestamps, and establish a time-series database of historical energy consumption of the target building.
[0043] Using historical energy consumption time series data of a target building at a certain moment, dynamic load forecast for future periods, real-time external environmental parameters, and dynamic range forecast of indoor environmental comfort as inputs, indoor set temperature, lighting brightness percentage, and variable air volume terminal opening as decision variables for each energy-consuming device, minimizing total operating energy consumption and satisfying the dynamic range of indoor environmental comfort as objective functions, and maintaining indoor environmental parameters within the dynamic range forecast model of indoor environmental comfort, physical operation constraints of each device, and system coupling constraints as constraints, a collaborative optimization model for energy-consuming devices of the target building is established, and the optimal decision variable sequence for each energy-consuming device of the target building in future periods is output.
[0044] By integrating the optimal decision variable sequence of each energy-consuming device in the target building for the future period, an optimal operating strategy set for each energy-consuming device in the target building for the future period is generated.
[0045] Preferably, step S4 specifically includes:
[0046] Based on generating the optimal operating strategy set for each energy-consuming device in the target building for the future time period, the BACnet communication protocol is selected to send the set to the BMS building equipment management system, converting the optimal operating strategy set for each energy-consuming device in the target building for the future time period into control commands and synchronizing the timestamp.
[0047] Based on the BMS building equipment management system, control commands are obtained, corresponding execution signals are generated, and the energy-consuming equipment in the target building is driven to adjust its operating status.
[0048] Preferably, step S4 further includes:
[0049] Real-time monitoring of actual energy consumption data and indoor environmental parameters of each energy-consuming device in the target building, compared with dynamic load forecasts for future periods and expected values of the optimal operation strategy set for each energy-consuming device in the target building for future periods, to calculate the deviation between actual and future energy consumption of each energy-consuming device and indoor environmental parameters;
[0050] Set preset deviation thresholds, including: energy consumption deviation threshold of 15% and indoor temperature deviation threshold of 1.5℃;
[0051] If any deviation between the actual and future energy consumption of each energy-consuming device and the indoor environmental parameters exceeds a preset deviation threshold and lasts for 15 minutes, the optimization process will be automatically triggered to generate a new set of optimal operating strategies for each energy-consuming device in the target building, thereby achieving adaptive control of the building's energy-consuming devices.
[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0053] This invention proposes a BIM-based intelligent control scheme for energy-saving renovation of existing buildings. By constructing a high-precision digital twin of the building, this scheme achieves deep integration between the physical building and the virtual model. Based on multi-source real-time data and dynamic load prediction, it can accurately generate optimal equipment operation strategies that balance energy saving and comfort. Furthermore, by leveraging a closed-loop adaptive control mechanism, it improves building energy efficiency and indoor environmental quality, providing a systematic solution that integrates data-driven, model-guided, and intelligent decision-making for energy-saving renovation of existing buildings. Attached Figure Description
[0054] Figure 1 This is a flowchart of a BIM-based intelligent control method for energy-saving retrofitting of existing buildings. Detailed Implementation
[0055] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0056] Reference Figure 1 As shown, a BIM-based intelligent control method for energy-saving retrofitting of existing buildings includes:
[0057] S1. Based on 3D laser scanning technology, acquire point cloud data of geometric coordinates of the target building, construct a BIM building information model, integrate IoT sensor data, and generate a digital twin of the target building;
[0058] Step S1 specifically includes:
[0059] Based on 3D laser scanning technology, the original point cloud data of multiple stations of the target building is acquired in real time, and the original point cloud data is preprocessed.
[0060] Based on the acquisition of original point cloud data from multiple stations of the target building, the original point cloud data from multiple stations of the target building are initially aligned using the corners of the walls, columns, and building height of the target building as references, and the original point cloud data of the target building is divided into source point cloud and target point cloud respectively.
[0061] Using the source point cloud of the target building and the target point cloud as input, calculate the Euclidean distance from each data point in the source point cloud of the target building to the target point cloud, find the closest corresponding point in the Euclidean distance as the initial point pair, and remove invalid point pairs that are greater than the maximum Euclidean distance;
[0062] The optimal rotation and translation transformation matrix of the target building is obtained by calculating the minimum average sum of squared distances between all valid point pairs using the least squares method.
[0063] Based on the optimal rotation and translation transformation matrix of the target building, the source point cloud position is continuously updated iteratively until the maximum number of iterations is reached, thereby achieving automatic registration of the original point cloud data of multiple stations of the target building.
[0064] Step S1 also includes:
[0065] Based on the original point cloud data of multiple stations of the target building after automatic registration, the data is cut into several point cloud blocks of the target building with uniform size, and corresponding semantic labels are assigned using a labeling tool.
[0066] The corresponding semantic tags include: walls, floors, pipes, doors and windows, and furniture;
[0067] Using PyTorch as the framework, a PointNet++ model is constructed, which includes a Set Abstraction layer and a Feature Propagation layer to perform hierarchical feature learning on several target building point cloud blocks of uniform size.
[0068] For the Set Abstraction layer, M center points are selected from N point cloud data in several target building point cloud blocks of uniform size, and the farthest point of several target building point cloud blocks of uniform size is extracted.
[0069] Using each center point as the sphere center, a spherical region is constructed with a radius of r. The neighboring points are found, and the local regions of the neighboring points are grouped. Max pooling is introduced to aggregate the features of different numbers of points in several target building point cloud blocks of uniform size into fixed-dimensional local features. High-level local features of several target building point cloud blocks of uniform size are extracted.
[0070] For the Feature Propagation layer, the inverse distance weighted interpolation method is used to upsample the high-level local features extracted by the SetAbstraction layer, gradually restoring the spatial resolution of point cloud data in several target building point cloud blocks of uniform size, thereby realizing the global propagation and fusion of features of several target building point cloud blocks of uniform size.
[0071] An encoder is constructed by stacking three Set Abstraction layers, and a decoder is constructed by stacking corresponding number of FeaturePropagation layers. An encoder-decoder structure is established, and the cross-entropy loss function is used as the optimization objective. The PointNet++ model learning rate, batch size, network depth, and ball query radius hyperparameters are continuously adjusted through backpropagation to achieve semantic segmentation of target building point cloud data.
[0072] Based on the semantic segmentation of the target building point cloud data, the BIM building information software is imported, and all point clouds marked as walls, floors, pipes, doors, windows and furniture are selected respectively. The Revit wall, floor, pipe, door, window and furniture commands are used respectively to generate the target building wall, floor, pipe, door, window and furniture models according to the outline of the target building point cloud data.
[0073] Integrate the target building's wall, floor, pipe, door, window, and furniture models to create a BIM building information model;
[0074] Based on IoT sensor data, the target building data is acquired in real time, coupled with the BIM building information model, and the BIM building information model is converted into an efficient format using the Unity game engine to generate a digital twin of the target building.
[0075] When using it, please refer to the above:
[0076] In existing technologies, the process of building a 3D laser scanning and BIM (Building Information Modeling) system often faces drawbacks such as low point cloud registration accuracy, reliance on extensive manual intervention for semantic segmentation, insufficient automation, and low efficiency in multi-source data fusion. These drawbacks result in long model building cycles, missing details, and difficulty in real-time updates. This step introduces automated point cloud registration and an optimized PointNet++ semantic segmentation model, achieving high-precision point cloud data processing and automatic component identification, thereby improving the automation level and accuracy of BIM building information model construction. Simultaneously, by combining IoT sensor data and the Unity engine, a dynamically visualized digital twin is constructed, effectively supporting real-time monitoring and intelligent management of the target building, demonstrating high efficiency, accuracy, and integration.
[0077] S2. Based on multi-source heterogeneous sensors, monitor the indoor environmental parameters of the target building in real time, establish a dynamic range prediction model for indoor environmental comfort, couple external real-time meteorological data, and input the target building's digital twin to predict the dynamic load of the target building in future periods.
[0078] Step S2 specifically includes:
[0079] Based on multi-source heterogeneous sensors and building automation systems, a fixed time interval of 5 minutes is set to monitor indoor environmental parameters of the target building in real time, acquire equipment operation data, and transmit the data to the central data server.
[0080] The indoor environmental parameters are: temperature, humidity, CO2 concentration, and light intensity.
[0081] Equipment operation data includes: air conditioning setpoint, valve opening, and fan status;
[0082] Based on the indoor environmental parameters of the target building, a time-series database of the indoor environment of the target building is constructed by aligning timestamps with unified data frequencies.
[0083] Using the PMV prediction average voting formula, the predicted human comfort values corresponding to historical and real-time environmental data are calculated.
[0084] Based on the predicted human comfort values corresponding to historical and real-time environmental data, the acceptable range of average votes for PMV prediction is defined.
[0085] The acceptable range for the average vote in the PMV prediction is: PMV ;
[0086] Using historical indoor environmental data of the target building, real-time indoor environmental parameters and time steps of the target building as inputs, and the upper and lower limits of comfort for each time step in the next 1-6 hours as prediction targets, a dynamic interval prediction model for indoor environmental comfort is established. The learning rate, batch size and network depth hyperparameters of the model are optimized and evaluated using cross-validation.
[0087] Step S2 also includes:
[0088] Real-time acquisition of outdoor weather data based on sensors;
[0089] The outdoor meteorological data includes: temperature, humidity, wind speed, and solar radiation;
[0090] Using the current real-time indoor environmental parameters of the target building as the initial state, and the predicted dynamic range of indoor environmental comfort and outdoor meteorological data as boundary conditions, the target building digital twin is associated and paired to predict the target building load at each time step in the next 24 hours.
[0091] Calculate the predicted values of cooling and heating loads for the target building to maintain the dynamic range of indoor environmental comfort, plot the dynamic load prediction curve for the target building in the future, and realize the dynamic load prediction for the target building in the future.
[0092] When using it, please refer to the above:
[0093] Current research on building environment monitoring and load forecasting based on multi-source sensors relies heavily on static comfort standards or fixed thresholds, lacking real-time adaptive forecasting of dynamic comfort ranges and external meteorological conditions. This results in insufficient accuracy and adaptability of the models under changing environments. This step establishes a dynamic range prediction model for indoor environmental comfort, coupled with real-time meteorological data and a digital twin, to achieve accurate prediction of the comfort range for the next 1-6 hours and the dynamic load of the building over 24 hours. This effectively improves the real-time performance, adaptability, and system energy efficiency of building energy management.
[0094] S3. Obtain historical energy consumption time series data of the target building, integrate dynamic load forecast for future periods, real-time external environmental parameters and dynamic range forecast of indoor environmental comfort, establish a collaborative optimization model for energy-consuming equipment of the target building, and generate the optimal operating strategy set for each energy-consuming equipment of the target building in the future period.
[0095] Step S3 specifically includes:
[0096] Acquire historical energy consumption data of the target building, preprocess the data, align the timestamps, and establish a time-series database of historical energy consumption of the target building.
[0097] Using historical energy consumption time series data of a target building at a certain moment, dynamic load forecast for future periods, real-time external environmental parameters, and dynamic range forecast of indoor environmental comfort as inputs, indoor set temperature, lighting brightness percentage, and variable air volume terminal opening as decision variables for each energy-consuming device, minimizing total operating energy consumption and satisfying the dynamic range of indoor environmental comfort as objective functions, and maintaining indoor environmental parameters within the dynamic range forecast model of indoor environmental comfort, physical operation constraints of each device, and system coupling constraints as constraints, a collaborative optimization model for energy-consuming devices of the target building is established, and the optimal decision variable sequence for each energy-consuming device of the target building in future periods is output.
[0098] By integrating the optimal decision variable sequence of each energy-consuming device in the target building for the future period, an optimal operating strategy set for each energy-consuming device in the target building for the future period is generated.
[0099] The specific expression for the objective function that minimizes total operating energy consumption and satisfies the dynamic range of indoor environmental comfort is as follows:
[0100]
[0101] In the formula, The total operating energy consumption of the target building at a certain moment. The total energy consumption of the HVAC system. Set the indoor temperature. For variable air volume terminal opening, This represents the dynamic load forecast for future periods. For external real-time environmental parameters, The total energy consumption of the lighting system, This represents a percentage of lighting brightness.
[0102] For comfort deviation penalty items, To predict the average vote, The dynamic range of indoor environmental comfort. The ideal center point for the dynamic range of indoor environmental comfort. The width is half the dynamic range of indoor environmental comfort. This is a tolerance adjustment parameter;
[0103] To minimize the total operating energy consumption and meet the dynamic range weighted sum of indoor environmental comfort, The total number of time steps for the predicted future period. For time step index, Energy consumption weighting coefficient For at any time Time-decision variable vector, For at any time Input variables at time, This is a comfort weighting coefficient. For at any time Predict the average vote count at that time.
[0104] When using it, please refer to the above:
[0105] In existing research, the optimization of building energy-consuming equipment usually relies on static load forecasting or fixed comfort ranges, lacking a synergistic consideration of future dynamic loads, real-time environmental parameters, and dynamic comfort ranges. This results in poor strategy adaptability, limited energy efficiency improvement, and insufficient comfort assurance. This step, by integrating multi-source dynamic data and establishing a collaborative optimization model, can improve the dynamic balance between building energy efficiency and indoor environmental comfort, generating optimal equipment operation strategies that take into account both economy and comfort, effectively overcoming the lag and rigid constraints of traditional methods.
[0106] S4. Based on the generated optimal operating strategy set for each energy-consuming device in the target building for the future time period, link it to the BMS building equipment management system to realize adaptive control of building energy-consuming devices;
[0107] Step S4 specifically includes:
[0108] Based on generating the optimal operating strategy set for each energy-consuming device in the target building for the future time period, the BACnet communication protocol is selected to send the set to the BMS building equipment management system, converting the optimal operating strategy set for each energy-consuming device in the target building for the future time period into control commands and synchronizing the timestamp.
[0109] Based on the BMS building equipment management system, control commands are obtained, corresponding execution signals are generated, and the energy-consuming equipment in the target building is driven to adjust its operating status.
[0110] Step S4 also includes:
[0111] Real-time monitoring of actual energy consumption data and indoor environmental parameters of each energy-consuming device in the target building, compared with dynamic load forecasts for future periods and expected values of the optimal operation strategy set for each energy-consuming device in the target building for future periods, to calculate the deviation between actual and future energy consumption of each energy-consuming device and indoor environmental parameters;
[0112] Set preset deviation thresholds, including: energy consumption deviation threshold of 15% and indoor temperature deviation threshold of 1.5℃;
[0113] If any deviation between the actual and future energy consumption of each energy-consuming device and the indoor environmental parameters exceeds a preset deviation threshold and lasts for 15 minutes, the optimization process will be automatically triggered to generate a new set of optimal operating strategies for each energy-consuming device in the target building, thereby achieving adaptive control of the building's energy-consuming devices.
[0114] When using it, please refer to the above:
[0115] In the current field of building energy management, existing technologies mostly rely on static preset or delayed response control, lacking the ability to optimize closed loops based on dynamic prediction. This results in a disconnect between energy consumption strategies and actual needs, lagging regulation, and limited energy efficiency improvement. This step, by linking the optimal operating strategy set for future time periods with the BMS building equipment management system in real time and introducing a deviation threshold triggering mechanism, achieves a leap from one-way command issuance to two-way dynamic optimization. This improves the adaptive regulation accuracy of building energy-consuming equipment and the overall energy efficiency of the system, ensuring the stability and controllability of indoor environmental quality.
[0116] Based on the above steps, the specific implementation method is as follows:
[0117] Using an office building in Shanghai built in 2000 as the subject of renovation, multiple 3D laser scanners were used to scan from various locations, including the building's internal corridors, offices, and external courtyard, to obtain massive amounts of raw point cloud data. After noise removal through point cloud preprocessing, the point clouds from different locations were initially aligned based on the building's four corner columns and floor height.
[0118] Using an improved algorithm based on nearest point iteration, the optimal rotation and translation matrix is automatically calculated. After multiple iterations, all point clouds are successfully registered with high precision into a complete 3D point cloud model of the building. This unified model is then cut into standard-sized blocks.
[0119] Using a pre-trained PointNet++ deep learning model, semantic labels such as walls, floors, pipes, doors and windows are automatically assigned to each point cloud block. These point clouds with semantic information are imported into Revit software to automatically generate a BIM building information model containing components such as walls, windows, and air conditioning ducts. Coupled with temperature, humidity and CO2 concentration sensor data deployed in the building, a dynamic digital twin that is updated synchronously with the physical building is built in the Unity engine.
[0120] Based on a dynamically updated digital twin that is synchronized with the physical building, a sensor network distributed throughout the office building continuously collects indoor temperature, humidity, CO2 concentration, and illuminance data for each floor every 5 minutes. Simultaneously, data such as air conditioning setpoints and fan operating status are obtained from the building automation system. This data is used to calculate the average voting index for PMV prediction in real time. Combined with historical data from the past 30 days, a dynamic range prediction model for indoor environmental comfort based on a long short-term memory network is trained. This model can predict the dynamic range of temperature and humidity that ensures over 90% of people feel comfortable within the next 6 hours, under the influence of natural factors and human activities, such as temperature fluctuations between 23℃ and 26℃. The model also incorporates real-time temperature, humidity, wind speed, and solar radiation data from the local meteorological bureau. These external parameters, along with the predicted comfort range, are used as boundary conditions and input into the digital twin for energy consumption simulation, accurately predicting the hourly cooling and heating load changes of the building over the next 24 hours.
[0121] Based on the predicted dynamic load curve for the next 24 hours, outdoor meteorological parameters, and indoor comfort dynamic range, a collaborative optimization model for energy-consuming equipment is established. This model uses the building's electricity consumption data for the past year as a benchmark, and takes the indoor set temperature of the air conditioning system, the opening degree of the variable air volume terminal, and the dimming percentage of the lighting system as decision variables. The optimization objective is to minimize the total operating energy consumption of the system while ensuring that the indoor environmental parameters always fall within the predicted comfort range.
[0122] By solving the model, an optimization strategy set is generated every 15 minutes for the next 24 hours. During the midday period when solar radiation is strong and the load is high, the air conditioning supply temperature is appropriately increased and the lighting brightness in non-primary areas is reduced. During the nighttime pre-cooling phase, the outdoor low-temperature air is fully utilized for free cooling. This strategy set specifies in detail the optimal operating state of each device at different time points.
[0123] Based on the generated optimal operating strategy set for the next 24 hours, the system is automatically sent to the building's existing BMS (Building Management System) via the BACnet communication protocol. The BMS then generates control commands to precisely adjust the operation of equipment such as chillers, fans, variable air volume terminals, and smart lighting switches. The system also monitors the actual energy consumption and indoor temperature of each device in real time and compares them with the expected values of the strategy. One afternoon, due to a large impromptu meeting that caused people to gather, the measured indoor temperature was 1.8°C higher than the expected value of the strategy for 15 minutes, exceeding the preset threshold of 1.5°C. The system immediately triggered the optimization process.
[0124] Based on new indoor environmental data, a new set of equipment operation strategies was quickly recalculated and generated, automatically adjusting the air conditioning air volume in relevant areas and rapidly bringing the temperature back to the comfortable range. This achieved adaptive control based on real-time feedback, ensuring a dynamic balance between energy efficiency and comfort.
[0125] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A BIM-based intelligent control method for energy-saving retrofitting of existing buildings, characterized in that, include: S1. Based on 3D laser scanning technology, acquire point cloud data of geometric coordinates of the target building, construct a BIM building information model, integrate IoT sensor data, and generate a digital twin of the target building; S2. Based on multi-source heterogeneous sensors, monitor the indoor environmental parameters of the target building in real time, establish a dynamic range prediction model for indoor environmental comfort, couple external real-time meteorological data, and input the target building's digital twin to predict the target building's dynamic load in future periods. S3. Obtain historical energy consumption time series data of the target building, integrate dynamic load forecast for future periods, real-time external environmental parameters and dynamic range forecast of indoor environmental comfort, establish a collaborative optimization model for energy-consuming equipment of the target building, and generate the optimal operating strategy set for each energy-consuming equipment of the target building in future periods. S4. Based on the generated optimal operating strategy set for each energy-consuming device in the target building for the future time period, link it to the BMS building equipment management system to realize adaptive control of building energy-consuming devices.
2. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 1, characterized in that, Step S1 specifically includes: Based on 3D laser scanning technology, the original point cloud data of multiple stations of the target building is acquired in real time, and the original point cloud data is preprocessed. Based on the acquisition of original point cloud data from multiple stations of the target building, the original point cloud data from multiple stations of the target building are initially aligned using the corners of the walls, columns, and building height of the target building as references, and the original point cloud data of the target building is divided into source point cloud and target point cloud respectively. Using the source point cloud of the target building and the target point cloud as input, calculate the Euclidean distance from each data point in the source point cloud of the target building to the target point cloud, find the closest corresponding point in the Euclidean distance as the initial point pair, and remove invalid point pairs that are greater than the maximum Euclidean distance; The optimal rotation and translation transformation matrix of the target building is obtained by calculating the minimum average sum of squared distances between all valid point pairs using the least squares method. Based on the optimal rotation and translation transformation matrix of the target building, the source point cloud position is continuously updated iteratively until the maximum number of iterations is reached, thereby achieving automatic registration of the original point cloud data of multiple stations of the target building.
3. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 2, characterized in that, Step S1 also includes: Based on the original point cloud data of multiple stations of the target building after automatic registration, the data is cut into several point cloud blocks of the target building with uniform size, and corresponding semantic labels are assigned using a labeling tool. The corresponding semantic tags include: walls, floors, pipes, doors and windows, and furniture; Using PyTorch as the framework, a PointNet++ model is constructed, which includes a Set Abstraction layer and a Feature Propagation layer to perform hierarchical feature learning on several target building point cloud blocks of uniform size. For the Set Abstraction layer, M center points are selected from N point cloud data in several target building point cloud blocks of uniform size, and the farthest point of several target building point cloud blocks of uniform size is extracted. Using each center point as the sphere center, a spherical region is constructed with a radius of r. The neighboring points are found, and the local regions of the neighboring points are grouped. Max pooling is introduced to aggregate the features of different numbers of points in several target building point cloud blocks of uniform size into fixed-dimensional local features. High-level local features of several target building point cloud blocks of uniform size are extracted. For the Feature Propagation layer, the inverse distance weighted interpolation method is used to upsample the high-level local features extracted by the Set Abstraction layer, gradually restoring the spatial resolution of point cloud data in several target building point cloud blocks of uniform size, and realizing the global propagation and fusion of features of several target building point cloud blocks of uniform size. An encoder is constructed by stacking three Set Abstraction layers, and a decoder is constructed by stacking corresponding number of FeaturePropagation layers. An encoder-decoder structure is established, and the cross-entropy loss function is used as the optimization objective. The PointNet++ model learning rate, batch size, network depth, and ball query radius hyperparameters are continuously adjusted through backpropagation to achieve semantic segmentation of target building point cloud data. Based on the semantic segmentation of the target building point cloud data, the BIM building information software is imported, and all point clouds marked as walls, floors, pipes, doors, windows and furniture are selected respectively. The Revit wall, floor, pipe, door, window and furniture commands are used respectively to generate the target building wall, floor, pipe, door, window and furniture models according to the outline of the target building point cloud data. Integrate the target building's wall, floor, pipe, door, window, and furniture models to create a BIM building information model; Based on IoT sensor data, the target building data is acquired in real time, coupled with the BIM building information model, and the BIM building information model is converted into an efficient format using the Unity game engine to generate a digital twin of the target building.
4. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 1, characterized in that, Step S2 specifically includes: Based on multi-source heterogeneous sensors and building automation systems, a fixed time interval of 5 minutes is set to monitor indoor environmental parameters of the target building in real time, acquire equipment operation data, and transmit the data to the central data server. The indoor environmental parameters are: temperature, humidity, CO2 concentration, and light intensity. Equipment operation data includes: air conditioning setpoint, valve opening, and fan status; Based on the indoor environmental parameters of the target building, a time-series database of the indoor environment of the target building is constructed by aligning timestamps with unified data frequencies. Using the PMV prediction average voting formula, the predicted human comfort values corresponding to historical and real-time environmental data are calculated. Based on the predicted human comfort values corresponding to historical and real-time environmental data, the acceptable range of average votes for PMV prediction is defined. The acceptable range for the average vote in the PMV prediction is: PMV ; Using historical indoor environmental data of the target building, real-time indoor environmental parameters and time steps of the target building as inputs, and the upper and lower limits of comfort for each time step in the next 1-6 hours as prediction targets, a dynamic interval prediction model for indoor environmental comfort is established. The learning rate, batch size and network depth hyperparameters of the model are optimized and evaluated using cross-validation.
5. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 4, characterized in that, Step S2 also includes: Real-time acquisition of outdoor weather data based on sensors; The outdoor meteorological data includes: temperature, humidity, wind speed, and solar radiation; Using the current real-time indoor environmental parameters of the target building as the initial state, and the predicted dynamic range of indoor environmental comfort and outdoor meteorological data as boundary conditions, the target building digital twin is associated and paired to predict the load of the target building at each time step in the next 24 hours. Calculate the predicted values of cooling and heating loads for the target building to maintain the dynamic range of indoor environmental comfort, plot the dynamic load prediction curve for the target building in the future, and realize the dynamic load prediction for the target building in the future.
6. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 1, characterized in that, Step S3 specifically includes: Acquire historical energy consumption data of the target building, preprocess the data, align the timestamps, and establish a time-series database of historical energy consumption of the target building. Using historical energy consumption time series data of a target building at a certain moment, dynamic load forecast for future periods, real-time external environmental parameters, and dynamic range forecast of indoor environmental comfort as inputs, indoor set temperature, lighting brightness percentage, and variable air volume terminal opening as decision variables for each energy-consuming device, minimizing total operating energy consumption and satisfying the dynamic range of indoor environmental comfort as objective functions, and maintaining indoor environmental parameters within the dynamic range forecast model of indoor environmental comfort, physical operation constraints of each device, and system coupling constraints as constraints, a collaborative optimization model for energy-consuming devices of the target building is established, and the optimal decision variable sequence for each energy-consuming device of the target building in future periods is output. By integrating the optimal decision variable sequence of each energy-consuming device in the target building for the future period, an optimal operating strategy set for each energy-consuming device in the target building for the future period is generated.
7. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 6, characterized in that, Step S4 specifically includes: Based on generating the optimal operating strategy set for each energy-consuming device in the target building for the future time period, the BACnet communication protocol is selected to send the set to the BMS building equipment management system, converting the optimal operating strategy set for each energy-consuming device in the target building for the future time period into control commands and synchronizing the timestamp. Based on the BMS building equipment management system, control commands are obtained, corresponding execution signals are generated, and the energy-consuming equipment in the target building is driven to adjust its operating status.
8. The intelligent control method for energy-saving retrofitting of existing buildings based on BIM according to claim 7, characterized in that, Step S4 also includes: Real-time monitoring of actual energy consumption data and indoor environmental parameters of each energy-consuming device in the target building, compared with dynamic load forecasts for future periods and expected values of the optimal operation strategy set for each energy-consuming device in the target building for future periods, to calculate the deviation between actual and future energy consumption of each energy-consuming device and indoor environmental parameters; Set preset deviation thresholds, including: energy consumption deviation threshold of 15% and indoor temperature deviation threshold of 1.5℃; If any deviation between the actual and future energy consumption of each energy-consuming device and the indoor environmental parameters exceeds a preset deviation threshold and lasts for 15 minutes, the optimization process will be automatically triggered to generate a new set of optimal operating strategies for each energy-consuming device in the target building, thereby achieving adaptive control of the building's energy-consuming devices.