Fabricated building self-optimization energy-saving system and method based on DRL and BIM

Abstract

The invention discloses a fabricated building self-optimization energy-saving system and method based on DRL and BIM, and relates to the technical field of building energy management and intelligent control. The system comprises a BIM data acquisition module used for storing and providing BIM data of a prefabricated building; the energy simulation module is used for constructing a digital twin model based on BIM data, providing a simulation environment for the DRL control module in an offline training stage, and assisting verification and retraining in an online stage; the environment monitoring and execution module is used for acquiring operation parameters and energy consumption data in real time through a sensor, an actuator and an interface butted with a transportation system and executing a control instruction; and the DRL control module obtains an optimization control strategy through simulation interaction learning, generates an instruction according to real-time data in an online stage, and issues and executes the instruction. Energy consumption and carbon emission in the whole process of the fabricated building can be optimized, the energy utilization efficiency and comfort are improved, and the self-adaptability and long-term stability of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy management and intelligent control, and particularly relates to a prefabricated building self-optimization energy-saving system and method based on DRL and BIM. BACKGROUND

[0002] Prefabricated building refers to a building that is constructed by transferring a large number of on-site operations in the traditional construction mode to a factory, processing and manufacturing building components and accessories (such as floor slabs, wall panels, staircases, balconies, etc.) in the factory, transporting them to the construction site, and assembling and installing them on site through reliable connection methods. Prefabricated buildings have gradually become the focus of the development of the construction industry due to their high construction efficiency, significant resource conservation effect, and small environmental impact. At the same time, building energy consumption continues to increase throughout the life cycle, and how to achieve energy optimization for prefabricated buildings has become the core issue of the industry.

[0003] Currently, the common approach is to use building information modeling (BIM) and artificial intelligence methods to predict building energy consumption and generate energy-saving control strategies through optimization algorithms. For example, the invention patent with application number CN202411380872.5 discloses a building energy consumption prediction and dynamic energy-saving optimization method based on BIM, which extracts building geometry, material, and construction parameters from the BIM model, combines environmental factors and historical energy consumption data, uses artificial intelligence models such as neural networks, support vector machines, or random forests for energy consumption prediction, and generates energy-saving optimization strategies through genetic algorithms, ultimately applied to ventilation, lighting, and heating, ventilation, and air conditioning systems to achieve dynamic energy-saving control.

[0004] However, existing solutions mainly focus on predicting and controlling energy consumption during the building use phase, focusing on dynamic adjustment of heating, ventilation, air conditioning, lighting, and ventilation subsystems. The energy consumption and carbon emissions of the component production, transportation, and construction phases specific to prefabricated buildings are not fully included in the optimization range, making it difficult to achieve energy-saving and emission-reduction goals throughout the life cycle. Existing technologies focus more on extracting geometric and operational parameters, and underutilize data such as component weight, process flow, and construction schedule contained in BIM models, making it difficult to achieve close coupling with the entire project process.

[0005] Although existing solutions combine AI models and optimization algorithms, their energy-saving control often relies on a single prediction model or a fixed optimization framework. Under external disturbances (such as climate change, fluctuations in user usage patterns) and system parameter uncertainty, existing technologies still have deficiencies in robustness and adaptability. Many focus on minimizing energy consumption, although some studies consider cost factors, the dynamic balance of energy consumption, comfort, carbon emissions, and equipment life is insufficient, which can lead to one-sided optimization strategies.

[0006] In summary, although the existing BIM-based building energy consumption prediction and optimization methods have made some progress, there are still obvious deficiencies in the full life cycle energy consumption coverage of prefabricated buildings, the depth of data utilization, the dynamic trade-off of multiple targets, and the safe evolution of strategies. Therefore, new technical solutions are still needed to achieve energy-saving optimization and carbon emission reduction management in the whole process and full life cycle of prefabricated buildings. SUMMARY

[0007] In view of the problems that the existing energy-saving optimization methods for prefabricated buildings are mostly limited to the operation stage, cannot cover the whole process of production, transportation and construction, lack of BIM data utilization, lack of self-adaptation and robustness in external disturbance and multi-objective coordination, and the control strategy lacks long-term evolution and security deployment mechanism, and it is difficult to realize energy saving and carbon emission reduction in the whole life cycle, the present application proposes a prefabricated building self-optimization energy-saving system and method based on DRL and BIM, which realizes multi-source information fusion in the whole process based on BIM data, and realizes comprehensive balance of energy consumption, comfort and carbon emission through intelligent control.

[0008] The present application realizes the above-mentioned purpose through the following technical solutions:

[0009] The prefabricated building self-optimization energy-saving system based on DRL and BIM comprises:

[0010] A BIM data acquisition module for storing and providing BIM data of prefabricated buildings, the BIM data comprising structural size and construction information of prefabricated buildings, thermal performance of envelope structure, material and mechanical and electrical equipment parameters, functional partition and use schedule, indoor comfort parameters, and production process parameters, volume and weight information of prefabricated components;

[0011] An energy simulation module for constructing a digital twin model of prefabricated buildings based on the BIM data, providing a simulation environment for the DRL control module in the offline training stage, simulating the energy consumption and carbon emission results under different control instructions, and assisting verification and retraining in the online operation stage;

[0012] An environment monitoring and execution module comprising environment sensors and device executors deployed in prefabricated buildings and construction sites, and a communication module interfaced with a transportation vehicle management system or a vehicle-mounted data interface, for collecting building operation parameters, construction equipment energy consumption and transportation state information in real time in the online operation stage, and adjusting the corresponding energy equipment or transportation scheduling according to the control instructions of the DRL control module;

[0013] The DRL control module comprises a trained DRL agent, is used for reading the calculation result of the energy simulation module in an offline training stage, learns an optimized control strategy for the prefabricated building through multiple rounds of interaction, reads real-time data collected by the environment monitoring and execution module as input in an online running stage, generates corresponding control instructions based on the optimized control strategy, and delivers the control instructions to the environment monitoring and execution module for execution.

[0014] As a preferred scheme of the present application, the energy simulation module is an interactive energy consumption and carbon emission simulation environment driven by the BIM data acquisition module and at least comprising:

[0015] Based on the structure size and construction information in the BIM data and the thermal performance of the envelope structure, the partition division, the spatial volume and the envelope heat transfer parameters of the digital twin model are determined, the wall heat storage performance, the solar heat gain coefficient and the efficiency and capacity limit of the heating ventilation and air conditioning system are determined based on the material and mechanical and electrical equipment parameters, and the internal heat dissipation, the personnel density and the schedule of each region are set as the boundary conditions according to the functional division and the use schedule;

[0016] Based on the indoor comfort parameters, the action feasible region and the comfort constraint of the DRL control module are generated, and the synchronous constraint on strategy exploration and reward calculation is implemented in the simulation and training process;

[0017] Based on the production process parameters, the volume and the weight information of the prefabricated component, an energy consumption and carbon emission calculation model of the component production and transportation stage is automatically established, and is linked with the reward function or the constraint condition of the DRL control module, so as to comprehensively evaluate the energy consumption and carbon emission influence of the production and transportation stage in the training process;

[0018] Through a parallel communication interface, the data interaction ability of training and evaluation is provided to the DRL control module to support distributed sampling and optimized training.

[0019] As a preferred scheme of the present application, the energy consumption and carbon emission calculation model of the component production stage establishes a dynamic process energy consumption function based on the BIM data, and the expression is as follows:

[0020]

[0021] In the formula, E prod is the total energy consumption in the production stage; i represents the production process step, and n represents the number of production processes; P i is the rated power of the process equipment; η i (λ i ) is the efficiency correction factor of the equipment under the load rate λ i ; t i is the process duration; δ iis a production batch scheduling factor, used to reflect the influence of staggered operation and task sequencing on energy consumption; N is the number of components;

[0022] C prod is the total carbon emission of the production stage; E prod,i is the energy consumption of a single process; θ i is the emission factor of the corresponding energy.

[0023] As a preferred scheme of the present application, the energy consumption and carbon emission calculation model of the component transportation stage establishes a dynamic transportation energy consumption function based on BIM data, and the expression is:

[0024]

[0025] In the formula, E trans is the total energy consumption of the transportation stage; j represents a transportation task, and m is the number of transportation tasks; g(W j , V j , L j ) is a vehicle energy consumption function, which is derived based on the component weight W j , volume V j and transportation distance L j ; ψ j is a loading and hoisting sequence factor, used to reduce invalid transportation and secondary handling; κ j is a traffic prediction correction factor, which dynamically adjusts the transportation energy consumption based on future congestion;

[0026] C trans is the total carbon emission of the transportation stage; E trans,j is the energy consumption of a single transportation task; φ j is the emission factor of the corresponding vehicle energy.

[0027] As a preferred scheme of the present application, the DRL control module adopts a multi-agent collaborative architecture, including a plurality of DRL agents that cooperate with each other, and each DRL agent is responsible for controlling different subsystems or control dimensions of the prefabricated building; the DRL agent at least includes:

[0028] an HVAC agent for controlling the temperature setting of the heating, ventilation and air conditioning of each functional partition of the building;

[0029] a fresh air agent for controlling the start-stop and air flow of the fresh air system;

[0030] a load side agent for controlling lighting, shading and adjustable load response;

[0031] Each DRL agent makes collaborative decisions based on a unified reward and punishment signal on the basis of a shared state space, and a collaboration mechanism arbitrates the control instructions of potential conflicts, so as to achieve the global optimization of energy saving and comfort targets.

[0032] As a preferred scheme of the present application, the reward function of the DRL control module simultaneously embeds three types of indicators of energy, comfort and carbon emissions, and imposes a penalty on the start-stop frequency of key equipment, expressed as:

[0033] r t =-w e E t -w s Φ(ΔT t )-w em (Em t +C prod +C trans )-w p E prod -w t E trans -w sw N sw,t ;

[0034] In the formula, r t represents the immediate reward value at time t; E t is the immediate energy consumption in the use stage;

[0035] ΔT t is the deviation of indoor temperature from the comfort range, ΔT t =max(T t -T max ,0)+max(T min -T t ,0), where T t is the indoor temperature at time t, T max , T min are the upper and lower limits of the comfort range allowed for indoor temperature, respectively; Φ(ΔT t ) represents an exponential nonlinear amplification function; Em t is the equivalent carbon emissions calculated according to the step electricity price or the energy carbon emission factor; N sw,t represents the start-stop times of key equipment within a time period; w e , w s , w em , w p , w t , w sw are weight coefficients, which are adaptively adjusted according to the electricity price, carbon intensity or occupancy rate.

[0036] As a preferred scheme of the present application, the DRL control module introduces adaptive perturbation and parallel training mechanism based on BIM data in the training process, including:

[0037] An adaptive disturbance generator is set in the energy simulation module, when a disturbance is applied to the input environmental conditions, according to the parameter sensitivity analysis result of the BIM data, the heat transfer coefficient of the enclosure structure, the efficiency of the heating ventilation and air conditioning system and the weight information of the prefabricated component are selected to be randomly sampled or noise superimposed;

[0038] The training process is run in parallel in multiple simulation instances, and different disturbance parameters and scene combinations are used to synchronously evaluate the control strategy;

[0039] The training task is implemented in a distributed parallel computing environment, and parallel sampling, parameter asynchronous updating, experience replay and policy delayed updating mechanisms are adopted.

[0040] As a preferred scheme of the present application, the DRL control module has a self-learning and evolutionary optimization mechanism in the policy updating and deployment process, including:

[0041] The integrated performance indicators in a preset period are continuously monitored, and the integrated performance indicators at least include energy consumption level, comfort level and carbon emission level, when the integrated performance indicators deviate from the threshold value, the simulation calibration and incremental retraining based on the latest running data collected by the environment monitoring and execution module are triggered;

[0042] Cross-generation elite sampling and evolution operations are introduced in the strategy search, and candidate strategies with evaluation performance better than a preset threshold in the historical training or running process are mutated and crossed to generate new strategy combinations, so that the control strategy is prevented from falling into local optimum;

[0043] Before the strategy is put into operation, a safety constraint filter based on BIM boundary conditions is used to limit the temperature and humidity, air volume and equipment capacity parameters, and a gray online and shadow deployment mechanism is used to verify the updated strategy in stages.

[0044] As a preferred scheme of the present application, the environment monitoring and execution module forms a multi-source state fusion mechanism with an external data interface, and the state vector includes:

[0045] Building operation parameters and equipment operation state data;

[0046] Space partition information and enclosure structure heat transfer parameters derived from BIM data;

[0047] Real-time or predicted meteorological data, including outdoor temperature, humidity and solar radiation information;

[0048] Traffic state and vehicle operation data, including transportation path, vehicle energy consumption and energy storage state;

[0049] Real-time occupancy number or occupancy rate and use schedule of each functional area;

[0050] The predictable external disturbance data includes weather forecast, traffic state prediction, and use schedule of each functional area, the state vector of the DRL control module is injected with the predictable external disturbance data in a look-ahead input manner, so that the control strategy can respond to external environmental changes in advance.

[0051] The environmental monitoring and execution module has a construction phase energy consumption monitoring function, which is used to collect energy consumption and peak load information of hoisting equipment and temporary facilities, and the DRL control module optimizes the hoisting sequence, operation period and temporary facility operation strategy based on the BIM construction progress and component hoisting plan, thereby reducing the energy consumption and carbon emissions in the construction phase.

[0052] The DRL and BIM-based self-optimizing energy-saving method for prefabricated buildings, the method comprises:

[0053] The BIM data of the prefabricated building is stored and provided, a digital twin model of the prefabricated building is constructed based on the BIM data, a simulation environment is provided for the DRL control module in the offline training stage, the energy consumption and carbon emissions under different control instructions are simulated and calculated, and the actual operation strategy is verified and retrained in the online operation stage;

[0054] Environmental sensors and equipment actuators are deployed in the prefabricated building and the construction site, and communicate with the transportation vehicle management system or the vehicle-mounted data interface, which is used to collect building operation parameters, construction equipment energy consumption and transportation state information in real time in the online operation stage, and adjust the corresponding energy equipment or transportation scheduling according to the control instruction;

[0055] The simulation calculation results are read in the offline training stage, and the optimized control strategy for the prefabricated building is obtained through multiple rounds of interactive learning; in the online operation stage, the collected real-time data is read as input, the corresponding control instruction is generated based on the optimized control strategy, and is sent to the environmental monitoring and execution step for execution.

[0056] The beneficial effects of the present application are: the multi-dimensional information of the structural size, the enclosure performance, the material parameters, the functional division, the comfort index and the prefabricated component production process of the fabricated building are uniformly analyzed and output, so that the energy simulation and control module can be directly called, the errors and repeated labor of traditional manual modeling are avoided, and the data of building design, construction and operation stage are connected; the building digital twin model is established based on the BIM data, which can provide an interactive energy consumption and carbon emission calculation environment for the DRL control module in the offline stage, so as to efficiently train the control strategy; in the online stage, it is used for auxiliary verification and retraining, which improves the verifiability and flexibility of the optimization strategy; through the communication interface of the sensor, the actuator and the transportation vehicle system, the building operation state, the construction equipment energy consumption and the transportation information can be collected in real time, and the corresponding adjustment can be made according to the control instruction, so as to realize the linkage management of the building use stage, the construction stage and the transportation stage. The DRL control module learns through repeated interaction of simulation data in the offline stage, forms the optimization control strategy for the fabricated building; in the online stage, the monitoring data are received in real time and the control instruction is output, so that the optimization control of building operation energy consumption can still be realized under complex disturbance and dynamic environment, the comfort degree is improved, and the carbon emission control is considered. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0058] Among them:

[0059] Figure 1 It is a schematic diagram of the modular structure of the system of the present application.

[0060] Figure 2 It is a method flowchart in the embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0062] As Figure 1As shown, it is one embodiment of the present application, which provides a DRL and BIM based prefabricated building self-optimizing energy saving system, comprising a BIM data acquisition module, an energy simulation module, an environment monitoring and execution module and a DRL control module.

[0063] (1) BIM data acquisition module

[0064] The BIM data acquisition module is used to store and provide the BIM (Building Information Modeling) data of the prefabricated building, which is derived from the BIM model file (such as IFC or Revit format) established in the architectural design stage and parsed into a structured parameter set through API or data interface. The BIM data includes:

[0065] Structural size and construction information of prefabricated building: floor height, open depth, enclosure component thickness and node connection form of the building, etc.

[0066] Thermal performance of enclosure structure: heat transfer coefficient (U value), air tightness parameters of external wall, roof, floor, door and window, etc.

[0067] Material and mechanical and electrical equipment parameters: thermal parameters of materials such as concrete, insulation layer and glass, and efficiency and capacity parameters of heating, ventilation and air conditioning, lighting and power transmission and distribution system;

[0068] Functional partition and use schedule: for example, personnel use rule, typical number of occupants and operation schedule of office area, dormitory area and warehouse area, etc.

[0069] Indoor comfort parameters: target range or allowable fluctuation range of indoor temperature, humidity and carbon dioxide concentration, etc.

[0070] Production process parameters, volume and weight information of prefabricated components: for example, factory production process, component weight and volume data of beam and slab, wall panel, etc.

[0071] Through the analysis and output of the BIM data acquisition module, the energy simulation module can directly use these parameters to establish a digital twin model, avoiding the errors and repetitive labor caused by traditional manual modeling.

[0072] (2) Energy simulation module

[0073] The energy simulation module takes BIM data as input to build a digital twin model of prefabricated building:

[0074] In the offline training stage, the energy simulation module provides an interactive simulation environment for the DRL control module to simulate the energy consumption and carbon emission results under different control instructions and feedback to the DRL agent for strategy optimization;

[0075] In the online running phase, the energy simulation module is used to assist in verifying the actual running strategy and providing re-training simulation support when needed.

[0076] The energy simulation module is connected with the external environment parameter interface, which can import meteorological data (temperature, humidity, solar radiation) and internal load data (personnel occupancy, equipment load) in real time or preset, so as to ensure that the simulation process is consistent with the actual running conditions.

[0077] As one of the specific embodiments of the present application, the following will be described in detail how the energy simulation module realizes the parameterization construction of the energy consumption and carbon emission simulation environment based on BIM data driving.

[0078] 1) Digital twin model parameterization

[0079] According to the structure size and construction information provided by the BIM data, the building space is automatically partitioned, such as office area, dormitory area, and warehouse area, and the volume of each partition is calculated;

[0080] According to the thermal performance of the envelope structure (such as the heat transfer coefficient of the wall, roof, and window), the heat exchange equation between each partition and the external environment is established; according to the door and window parameters, the solar heat gain coefficient or equivalent transmission performance is determined, which is used to simulate the heat gain under different orientations and shading conditions;

[0081] According to the material parameters, the heat capacity of the wall and floor is determined to reflect the thermal inertia of the building; according to the parameters of the mechanical and electrical equipment, the efficiency, maximum refrigerating capacity and maximum heating capacity of the HVAC system are determined as the equipment operation boundary;

[0082] According to the functional partition and use schedule, the personnel density, lighting and equipment load curve are mapped into the internal load model, so that the simulation can reflect the actual use scene;

[0083] 2) Comfort parameter constraint

[0084] Based on the indoor comfort parameters, the action feasible region and comfort constraint of the DRL control module are automatically generated, for example, when the room temperature target range is 22-26℃, the humidity target range is 40%-60%, and the CO2 does not exceed 1000ppm, the DRL action space is automatically limited in this interval, ensuring that the strategy will not produce instructions that violate the comfort constraint in the training.

[0085] 3) Energy consumption and carbon emission modeling in production and transportation stage

[0086] According to the prefabricated component production process parameters (component type, material type, steam curing time, steel reinforcement welding amount, electric welding power, etc.) in the BIM data, the energy consumption and carbon emission calculation model of the production link is automatically established, such as concrete curing energy consumption, electric welding and cutting energy consumption;

[0087] Example: Energy consumption = ∑(energy consumption of each process step × duration of each process step × number of components)

[0088] Carbon emission = Energy consumption × emission factor of corresponding energy (e.g. kgCO2 / kWh for electricity, kgCO2 / m3 for natural gas) 3

[0089] Total energy consumption and carbon emission of each single component or batch production

[0090] Based on the volume and weight information of components, combined with the performance parameters of transportation vehicles (fuel consumption / energy consumption curve, load limit, transportation distance and road conditions), an energy consumption and carbon emission calculation model for the transportation stage is automatically established to optimize the transportation path and loading scheme.

[0091] Example:

[0092] Energy consumption = (energy consumption per mile of vehicle × distance) × (loading weight coefficient)

[0093] Carbon emission = Energy consumption × fuel / power emission factor

[0094] Output energy consumption and carbon emission of each single transportation task

[0095] The output results of the energy consumption and carbon emission calculation model of the production and transportation stages can be used as additional reward function items, so that the agent will automatically learn to consider the energy consumption and carbon emission optimization of the production and transportation stages while meeting the comfort requirements during the training process.

[0096] In a preferred embodiment, a multi-process energy consumption and carbon emission model for the component production stage is established in the energy simulation module.

[0097] The production process usually includes concrete mixing and molding, steam curing, stacking and handling, steel bar welding, etc. The energy consumption and emission of each process are not simply fixed, but are affected by the equipment operating state and production scheduling. The energy consumption of a single process is represented as:

[0098] E prod,i = P i · η i (λ i )· t i · δ i · N

[0099] The total energy consumption of the production stage is:

[0100] The total carbon emission of the production stage is:

[0101] E prod ​Total energy consumption of production phase; i represents production process step, n represents production process number; P i Process equipment rated power, such as mixer motor power, steam boiler thermal power; η i (λ i ) is the efficiency correction factor of equipment under load rate λ i ; t i is the process duration, which can be determined by process parameters in BIM model (such as curing time 48h); δ i is the production batch scheduling factor, which is used to reflect the influence of peak-shaving operation and task sequencing on energy consumption; N is the number of components;

[0102] C prod Total carbon emission of production phase; E prod,i is the energy consumption of a single process; θ i is the emission factor of corresponding energy, for example: electricity 0.65 kgCO2 / kWh, natural gas 2.1 kgCO2 / m 3 .

[0103] By introducing the efficiency correction factor and the production batch scheduling factor, this model can reflect the influence of nonlinear working conditions in actual production process and scheduling optimization on energy consumption and emission, which is different from traditional static LCA calculation. The final E prod , C prod are input into the reward function of the DRL control module, so that the agent learns to balance the energy consumption and carbon emission of the production link in the control strategy.

[0104] In another embodiment, transportation energy consumption and carbon emission are affected by multiple factors such as component volume, weight, loading sequence, transportation distance, road traffic, etc. The energy simulation module establishes a dynamic transportation model based on BIM data.

[0105] The energy consumption of a transportation task is represented as: E trans,j = g(W j , V j , L j ) · ψ j · κ j ;

[0106] The total transportation energy consumption is:

[0107] The total carbon emission of the transportation phase is:

[0108] In the formula, E trans is the total energy consumption of the transportation phase; j represents the transportation task, and m is the number of transportation tasks; g(W j , V j , Lj ) is the vehicle energy consumption function, based on the total weight of the transport component W j , the total volume of the transport component V j , and the transport distance L j , derived as, for example where a, b, g are fitting parameters, W max , V max are the weight and volume limits of the vehicle load respectively; y j is the loading and hoisting sequence factor, used to reduce ineffective transportation and secondary handling, taking value > 1 if the loading scheme requires secondary handling; k j is the traffic prediction correction factor, dynamically adjusting the transportation energy consumption based on future congestion, for example setting k j = 1.2 during peak hours;

[0109] C trans is the total carbon emission of the transportation phase; E trans,j is the energy consumption of a single transportation task; f j is the emission factor corresponding to the vehicle energy (diesel, electric, hydrogen).

[0110] This model incorporates the construction plan (loading sequence) and external environment (traffic prediction) into the transportation energy consumption calculation, and can dynamically reflect the impact of the transportation scheme on carbon emissions. The obtained E trans , C trans are also used as DRL training targets.

[0111] 4) Interactive interface and distributed training support

[0112] The energy simulation module is connected to the DRL control module through a parallel communication interface:

[0113] During the training process, multiple simulation instances can run simultaneously, supporting parallel sampling and improving training efficiency;

[0114] In the evaluation stage, the module can output multi-dimensional indicators such as partition energy consumption, carbon emission intensity, indoor comfort compliance rate, etc., for the DRL control module to comprehensively judge the pros and cons of the strategy.

[0115] Through the description of the embodiments, the energy simulation module realizes end-to-end parameterization based on BIM data, and can directly reflect the building physical characteristics, comfort constraints, production and transportation energy consumption in DRL training, thereby ensuring that the optimized strategy not only conforms to the engineering reality, but also has the ability to control energy consumption and carbon emissions across stages. This embodiment improves the precision and automation level of simulation, significantly shortens the modeling cycle, and realizes energy efficiency optimization throughout the production-transportation-construction-use process through DRL.

[0116] (3) Environmental monitoring and execution module

[0117] The environmental monitoring and execution module comprises:

[0118] Environmental sensors (temperature, humidity, CO2 concentration, illuminance, electricity meter, etc.) deployed indoors and outdoors of the fabricated building;

[0119] Energy consumption monitoring devices on the construction site (such as crane power sensors, temporary facility power metering devices);

[0120] A communication module interfaced with a transportation vehicle management system or a vehicle-mounted data interface for real-time collection of transportation vehicle energy consumption, mileage, and real-time location data;

[0121] During system operation, the environmental monitoring and execution module collects the above real-time data and transmits them to the DRL control module, and issues control instructions to the heating, ventilation, and air conditioning system, the fresh air system, the lighting device, the hoisting device, and the transportation scheduling system through the actuator interface, to realize real-time landing execution of the strategy.

[0122] The embodiment sets up a multi-source state fusion mechanism in the environmental monitoring and execution module, and cooperates with external data interfaces, so that the DRL control module can generate an optimized control strategy under more comprehensive data support. The state vector of the system not only contains running data from indoor sensors and device controllers, but also combines BIM data and external interface data to form a complete state input:

[0123] Building operation parameters and device operation state data: including indoor temperature, humidity, carbon dioxide concentration, illuminance, electricity meter readings, and device operation mode (start-stop state, power level), which are directly collected by environmental sensors and actuators;

[0124] Spatial partition information and envelope heat transfer parameters derived from BIM data: based on the building information model, automatically extract the spatial volume of each partition of the building, the partition boundary, and the envelope heat transfer energy parameters (such as the heat transfer coefficient U value of the wall and roof), for establishing a refined energy consumption model;

[0125] Real-time or predicted meteorological data: obtained through a meteorological data interface, including current and predicted outdoor temperature, humidity, and solar radiation intensity, for guiding the adjustment of HVAC and shading systems;

[0126] Traffic state and vehicle operation data: obtained through an interface with a transportation vehicle management system, including transportation path traffic conditions, vehicle energy consumption level, and fuel or battery remaining state (SOC), to provide inputs for transportation link scheduling and carbon emission assessment;

[0127] Real-time number of occupants or occupancy rate and use schedule of each functional area: actual occupancy is obtained through people flow sensors or an Occupancy interface, and combined with the use schedule information preset by BIM, to predict load changes and fresh air demand;

[0128] After the fusion of the above multi-source data, the input state vector is unified, which serves as the basis for decision-making of the DRL control module.

[0129] The system integrates weather forecasts, traffic state forecasts, and the use schedule of each functional area to form "predictable external disturbance data". These data are injected into the state vector through forward-looking input, that is, not only the current state but also the known disturbance trend in the future is considered when making a strategy decision. For example:

[0130] The hourly weather forecast curve for the next 24 hours is used as an additional input;

[0131] The predicted congestion period of the transportation path is used as a disturbance factor;

[0132] The schedule of each functional area is converted into a future occupancy rate forecast.

[0133] In this way, the DRL agent can learn and adapt to upcoming disturbances in advance, thereby generating a more forward-looking control strategy rather than relying solely on immediate state feedback.

[0134] During the construction phase, the environmental monitoring and execution module adds an energy consumption monitoring function to collect: power consumption and running time of hoisting equipment (such as cranes, hydraulic equipment); lighting, air conditioning, heating energy consumption of temporary facilities (such as temporary offices, dormitories, warehouses); peak load data of the construction site. After these data are input into the DRL control module, they are combined with the construction progress and component hoisting plan of BIM:

[0135] According to the construction progress, determine the component hoisting tasks to be completed each day;

[0136] According to the hoisting plan, generate alternative hoisting sequences and work periods;

[0137] The DRL control module selects the hoisting arrangement with lower energy consumption through optimization strategy and coordinates the operation mode of temporary facilities to reduce energy consumption and carbon emissions during the construction phase while ensuring construction progress.

[0138] (4) DRL (Deep Reinforcement Learning, Deep Reinforcement Learning) control module

[0139] During the offline training phase, the calculation results of the energy simulation module are read, and through multiple rounds of interactive learning (state input, action output, reward feedback), an optimized control strategy for prefabricated buildings is obtained;

[0140] In the online running phase, the real-time data collected by the environment monitoring and execution module are read as input, and the corresponding control instructions are generated based on the optimization control strategy, and are issued to the environment monitoring and execution module for execution.

[0141] The control targets include but are not limited to: reducing energy consumption per unit time, reducing carbon emissions, maintaining indoor comfort (temperature and humidity, air quality), optimizing transportation energy consumption and construction equipment energy consumption.

[0142] The DRL control module adopts a multi-agent architecture, and each agent focuses on a specific subsystem:

[0143] HVAC agent: input partition indoor and outdoor temperature, heat load, output set temperature;

[0144] Fresh air agent: input CO2 concentration and number of occupants, output air supply and start-stop instructions;

[0145] Load-side agent: input solar intensity and occupancy, output lighting brightness and sunshade angle.

[0146] Each agent shares state information and receives feedback through a unified reward function.

[0147] If the actions of the HVAC and fresh air agents conflict (for example, too much fresh air increases the cooling load), the arbitration mechanism automatically adjusts the priority according to the reward and punishment signals.

[0148] This collaboration ensures that local actions do not harm the overall energy efficiency goal.

[0149] The reward function takes into account energy consumption, comfort, carbon emissions, and equipment start-stop, and is defined as:

[0150] r t =-w e E t -w s Φ(ΔT t )-w em (Em t +C prod +C trans )-w p E prod -w t E trans -w sw N sw,t ;

[0151] In the formula, r t represents the immediate reward value at time t; E t is the instantaneous energy consumption in the use phase;

[0152] ΔT t is the deviation of indoor temperature from the comfort range, ΔTt = max(T t -T max , 0) + max(T min -T t , 0), where T t is the indoor temperature at time t, T max , T min are the upper and lower limits of the comfort range allowed for the indoor temperature; Φ(ΔT t ) represents an exponential nonlinear amplification function, e.g., Φ(x) = x γ , γ > 1; Em t is the equivalent carbon emission calculated according to the step electricity price or the energy carbon emission factor; N sw,t represents the number of start-stop times of the key equipment within the time period; w e , w s , w em , w p , w t , w sw are weight coefficients, which are adaptively adjusted according to the electricity price, the carbon intensity or the occupancy rate.

[0153] The reward function not only punishes the energy consumption and comfort deviation in the use stage, but also superimposes the energy consumption and carbon emission in the production and transportation stages;

[0154] The weight coefficients can be adjusted according to the electricity price, the carbon intensity or the building function, so that the optimization strategy is forward-looking;

[0155] The nonlinear amplification function Φ(·) is introduced to make the large-scale comfort violation be punished more severely, so as to ensure the priority of comfort.

[0156] Through this reward mechanism, the system can realize the optimal control of energy consumption and carbon emission in the whole life cycle.

[0157] In this embodiment, in order to enhance the robustness of the deep DRL control strategy and speed up the training convergence speed, the system introduces an adaptive disturbance and parallel training mechanism based on BIM data in the training process of the DRL control module. It includes:

[0158] 1) Construction of adaptive disturbance generator

[0159] An adaptive disturbance generator is set in the energy simulation module, which is used to apply disturbance to the environmental conditions of the training input.

[0160] First, the parameter sensitivity analysis result of the BIM data is obtained, and the key parameters which have significant influence on energy consumption and carbon emission are identified;

[0161] The key parameters with the highest sensitivity in this embodiment are selected, including the envelope structure heat transfer coefficient (U value, directly affecting the building heat load), the efficiency of the heating, ventilation and air conditioning system (COP, EER, determining the energy consumption of refrigeration and heating), and the weight information of the prefabricated component (affecting the transportation and hoisting energy consumption); the perturbation generator performs random sampling or noise superposition on the above key parameters during training:

[0162] For example, the heat transfer coefficient is disturbed using a Gaussian distribution: Wherein, U' is the disturbed envelope structure heat transfer coefficient; U is the original envelope structure heat transfer coefficient in the BIM data (unit: W / m 2 K); ∈ is the random disturbance applied to the heat transfer coefficient; is a normal distribution with a mean of 0 and a variance of σ 2 ; σ is the standard deviation of the disturbance amplitude, which determines the disturbance strength;

[0163] The heating efficiency parameter can be disturbed using a uniform distribution: Wherein, COP' is the disturbed heating, ventilation and air conditioning system performance coefficient (Coefficient of Performance); COP min is the lower limit value of the preset COP (such as the efficiency of the equipment under the worst working condition); COP max is the upper limit value of the preset COP (such as the efficiency of the equipment under the best working condition); [COP min , COP max ] is a uniform distribution in the interval σ;

[0164] The component weight is disturbed using a normal distribution or a proportional scaling: W' = W · (1 + δ), Wherein, W' is the disturbed prefabricated component weight; W is the original prefabricated component weight in the BIM data; δ is the weight disturbance factor; is a normal distribution with a mean of 0 and a variance of τ 2 ; τ is the standard deviation of the weight disturbance factor, which determines the disturbance amplitude;

[0165] In this way, different training rounds use simulation environments disturbed by parameters, thereby improving the adaptability of the strategy to uncertain conditions.

[0166] 2) Multi-scene parallel training

[0167] During the DRL training process, the system runs multiple simulation instances in parallel, and each instance loads different disturbance parameters and scene combinations. For example: use a higher envelope heat transfer coefficient, a lower COP, and a larger component weight in one instance; use the opposite combination in another instance. Each instance independently interacts with the DRL agent and generates samples, and finally the training results of multiple instances are unified into the experience pool for playback.

[0168] This method avoids overfitting caused by a single environment, enabling the control strategy to maintain stable performance under different operating conditions.

[0169] 3) Distributed parallel computing environment

[0170] The training task is implemented in a distributed computing environment:

[0171] Parallel sampling of multiple simulation instances using multi-core CPUs and GPU clusters;

[0172] Using a parameter server architecture to support asynchronous updating of agent policy network parameters;

[0173] Introducing the experience replay mechanism to uniformly store state-action-reward-next state sequences from different disturbance environments and provide them to the agent for training in random batches;

[0174] Using the policy delay update mechanism to distinguish the update frequency of the value network and the policy network, avoiding oscillation and divergence. The training task is implemented in a distributed parallel computing environment using parallel sampling, asynchronous parameter updating, experience replay, and policy delay updating mechanisms.

[0175] In one preferred embodiment of the present application, the DRL control module has a self-learning and evolutionary optimization mechanism during long-term operation to ensure that the control strategy remains energy-efficient and comfortable when the building operating conditions change.

[0176] The system calculates the following comprehensive performance indicators within a preset period (e.g., a week or a month):

[0177] Energy consumption level: based on real-time energy consumption E t , production energy consumption E prod , and transportation energy consumption E trans , the total energy consumption or energy consumption per unit area within the calculation period is calculated;

[0178] Comfort level: the proportion of time within the calculation period that the indoor temperature T t falls within the comfort range [T min , T max ] is taken as the comfort compliance rate;

[0179] Carbon emission level: based on the total carbon emissions Em t +C prod +C trans , the carbon intensity per unit energy consumption or per unit building area is calculated.

[0180] When the above comprehensive performance indicators deviate from the preset threshold (e.g., energy consumption exceeds 5%, comfort compliance rate is less than 90%, carbon intensity is higher than the limit), the system triggers the self-learning process.

[0181] After triggering, the DRL control module calls the energy simulation module and updates the initial state and parameters of the digital twin model based on the latest operating data collected by the environment monitoring and execution module, including indoor and outdoor temperature and humidity, equipment operating status, load level, transportation energy consumption, etc. On this basis, the agent uses the historical optimal strategy as a priori to perform incremental training, thereby shortening the convergence time and improving the retraining efficiency.

[0182] During the retraining process, the system establishes a candidate strategy library, performs cross-generation elite sampling on high-performance strategies that perform better than the preset threshold in historical operation, and applies evolutionary operations (including parameter crossover and mutation) to generate new candidate strategies.

[0183] These strategies are evaluated in a simulation environment, and the best ones are retained, thereby avoiding the control strategy from falling into a local optimum and improving the diversity and adaptability of the strategy.

[0184] Before the strategy goes online, the system performs two-level safety verification:

[0185] Safety constraint filter: use BIM boundary conditions to constrain temperature and humidity, air volume, and equipment capacity to ensure that the new strategy output does not exceed the design limit;

[0186] Gray online and shadow deployment: the new strategy first runs in "shadow mode", calculating the predicted control effect but not actually executing, and is verified by comparing with the old strategy; then it is gradually expanded in control range by gray online until it completely replaces the old strategy.

[0187] Through the mechanism of the embodiment, the system can automatically monitor energy saving and comfort effect in actual operation and trigger self-learning update in time when performance decreases. High-performance strategies in the strategy library are further optimized through cross-generation elite sampling and evolutionary operations, and gray online and safety filter ensure the controllability and safety of the strategy updating process. This design enables the system to have continuous evolution capability and adapt to changes in working conditions and equipment aging in long-term operation of buildings, and always maintain the optimal effect of energy saving and emission reduction.

[0188] As shown in Figure 2 Another embodiment of the present application provides a DRL and BIM-based prefabricated building self-optimizing energy saving method, which includes:

[0189] Store and provide BIM data of prefabricated buildings, build a digital twin model of prefabricated buildings based on BIM data, provide a simulation environment for DRL control module in offline training stage, simulate and calculate energy consumption and carbon emission results under different control instructions, and assist in verifying and retraining actual operation strategy in online operation stage;

[0190] In the prefabricated building and construction site deployment environment sensor and equipment executor, and with transport vehicle management system or vehicle-mounted data interface communication, for real-time collection of building operation parameters, construction equipment energy consumption and transportation state information in the online operation stage, and according to the control instruction, adjust the corresponding energy equipment or transportation scheduling;

[0191] The offline training stage reads the simulation calculation result, obtains the optimization control strategy for prefabricated buildings through multi-round interactive learning; in the online operation stage, the collected real-time data are read as input, the corresponding control instruction is generated based on the optimization control strategy, and is issued to the environment monitoring and execution step for execution.

[0192] In summary, the present application combines BIM data modeling, digital twin simulation, multi-source state monitoring and deep reinforcement learning control in the system architecture, not only improves the energy saving and comfort effect of prefabricated building in the operation stage, but also extends the energy saving optimization range to the construction and transportation process, realizes the energy consumption and carbon emission optimization control in the whole life cycle of building.

[0193] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of various changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A prefabricated building self-optimizing energy-saving system based on DRL and BIM, characterized in that, The system includes: The BIM data acquisition module is used to store and provide BIM data for prefabricated buildings. The BIM data includes the structural dimensions and construction information of the prefabricated buildings, thermal performance of the building envelope, material and electromechanical equipment parameters, functional zoning and usage schedule, indoor comfort parameters, and production process parameters, volume and weight information of prefabricated components. The energy simulation module is used to construct a digital twin model of the prefabricated building based on the BIM data. During the offline training phase, it provides a simulation environment for the DRL control module to simulate and calculate the energy consumption and carbon emission results under different control commands. During the online operation phase, it is used to assist in verification and retraining. The environmental monitoring and execution module includes environmental sensors and equipment actuators deployed at prefabricated buildings and construction sites, as well as a communication module that interfaces with a transportation vehicle management system or vehicle data interface. It is used to collect building operation parameters, construction equipment energy consumption and transportation status information in real time during the online operation phase, and adjust the corresponding energy-consuming equipment or transportation scheduling according to the control instructions of the DRL control module. The DRL control module includes a trained DRL agent, which reads the calculation results of the energy simulation module during the offline training phase and obtains an optimized control strategy for prefabricated buildings through multiple rounds of interactive learning. During the online operation phase, it reads real-time data collected by the environmental monitoring and execution module as input, generates corresponding control commands based on the optimized control strategy, and sends them to the environmental monitoring and execution module for execution.

2. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 1, characterized in that, The energy simulation module is an interactive energy consumption and carbon emission simulation environment, driven by the BIM data acquisition module, and includes at least: Based on the structural dimensions and construction information and the thermal performance of the building envelope in the BIM data, the zoning, spatial volume and heat transfer parameters of the building envelope of the digital twin model are determined. Based on the material and electromechanical equipment parameters, the heat storage performance of the wall, the solar heat gain coefficient and the efficiency and capacity limits of the HVAC system are determined. The internal heat dissipation, personnel density and time schedule of each area are set as boundary conditions according to the functional zoning and usage schedule. Based on the indoor comfort parameters, the feasible domain of action and comfort constraints of the DRL control module are generated, and synchronous constraints are implemented on strategy exploration and reward calculation during simulation and training. Based on the production process parameters, volume and weight information of the prefabricated components, an energy consumption and carbon emission calculation model for the component production and transportation stages is automatically established and linked with the reward function or constraint conditions of the DRL control module to comprehensively evaluate the energy consumption and carbon emission impact of the production and transportation stages during the training process. The parallel communication interface provides the DRL control module with the ability to exchange training and evaluation data to support distributed sampling and optimization training.

3. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 2, characterized in that, The energy consumption and carbon emission calculation model during the component production stage establishes a dynamic process energy consumption function based on BIM data, with the following expression: In the formula, E prod P represents the total energy consumption during the production phase; i represents the number of production steps, and n represents the number of production steps; i Rated power of the process equipment; η i (λ i ) represents the equipment at load factor λ i The efficiency correction factor below; t i This refers to the duration of the process. δ i This is the production batch scheduling factor, used to reflect the impact of off-peak operation and task sequencing on energy consumption; N is the number of components. C prod E represents total carbon emissions during the production phase. prod,i Energy consumption for a single production process; θ i This refers to the emission factor for the corresponding energy source.

4. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 3, characterized in that, The energy consumption and carbon emission calculation model for the component transportation phase is based on BIM data, establishing a dynamic transportation energy consumption function, the expression of which is: In the formula, E trans The total energy consumption during the transportation phase; j represents the transportation task, m represents the number of transportation tasks; g(W) j V j ,L j ) represents the vehicle energy consumption function, based on the component weight W. j Volume V j and transport distance L j ψ j κ is a loading and hoisting sequence factor used to reduce unnecessary transportation and secondary handling; j C is a traffic forecast correction factor that dynamically adjusts transportation energy consumption based on future congestion conditions. trans E represents total carbon emissions during the transportation phase. trans,j Energy consumption for a single transport mission; φ j This refers to the emission factor corresponding to the vehicle's energy.

5. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 1, characterized in that, The DRL control module adopts a multi-agent collaborative architecture, which includes several DRL agents that cooperate with each other. Each DRL agent is responsible for controlling different subsystems or control dimensions of the prefabricated building. The DRL agent includes at least: HVAC smart agents are used to control the temperature settings of HVAC systems in various functional zones of a building. The fresh air intelligent agent is used to control the start-up and shutdown of the fresh air system and the airflow. Load-side intelligent agents are used to control lighting, shading, and adjustable load response; Each DRL agent makes collaborative decisions based on a unified reward and punishment signal while sharing a state space. The collaborative mechanism arbitrates potentially conflicting control commands to achieve the global optimum of energy saving and comfort goals.

6. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 4, characterized in that, The reward function of the DRL control module simultaneously incorporates three types of indicators: energy, comfort, and carbon emissions, and imposes penalties on the start-up and shutdown frequency of key equipment, as expressed as: r t =-w e E t -w s Φ(ΔT t )-w em (Em t +C prod +C trans )-w p E prod -w t E trans -w sw N sw,t ; In the formula, r t E represents the instantaneous reward value at time t; t This refers to the immediate energy consumption during the usage phase. ΔT t ΔT represents the deviation of indoor temperature from the comfort range. t =max(T) t -T max ,0)+max(T min -T t ,0), where T t Let T be the indoor temperature at time t. max ,T min These represent the upper and lower limits of the permissible comfortable range for indoor temperature, respectively; Φ(ΔT) t ) represents an exponential nonlinear amplification function; Em t Equivalent carbon emissions calculated using tiered electricity pricing or energy carbon emission factors; N sw,t Indicates the number of times critical equipment is started and stopped within a time period; w e ,w s ,w em ,w p ,w t ,w sw The weighting coefficients are adaptively tuned based on electricity price, carbon intensity, or occupancy rate.

7. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 2, characterized in that, The DRL control module introduces an adaptive perturbation and parallel training mechanism based on BIM data during the training process, including: An adaptive disturbance generator is set in the energy simulation module. When disturbances are applied to the input environmental conditions, the heat transfer coefficient of the furnace structure, the efficiency of the HVAC system, and the weight information of the prefabricated components are randomly sampled or noise is superimposed based on the parameter sensitivity analysis results of the BIM data. The training process is run in parallel across multiple simulation instances, and the control strategy is evaluated synchronously using different perturbation parameters and scenario combinations. The training task is implemented in a distributed parallel computing environment, employing parallel sampling, asynchronous parameter updates, experience replay, and policy delayed update mechanisms.

8. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 1, characterized in that, The DRL control module possesses a self-learning and evolutionary optimization mechanism during policy updates and deployments, including: The system continuously monitors comprehensive performance indicators within a preset period. These comprehensive performance indicators include at least energy consumption level, comfort level, and carbon emission level. When a comprehensive performance indicator is detected to deviate from a threshold, simulation calibration and incremental retraining based on the latest operating data collected by the environmental monitoring and execution module are triggered. In the policy search, cross-generational elite sampling and evolution operations are introduced to mutate and cross candidate policies that have performed better than a preset threshold during historical training or operation to generate new policy combinations and avoid the control policy from getting trapped in local optima. Before the strategy is deployed online, a safety constraint filter based on BIM boundary conditions is used to limit temperature, humidity, air volume and equipment capacity parameters, and a gray-scale deployment and shadow deployment mechanism is used to verify the updated strategy in stages.

9. The prefabricated building self-optimizing energy-saving system based on DRL and BIM according to claim 1, characterized in that, The environmental monitoring and execution module forms a multi-source state fusion mechanism with the external data interface, and the state vector includes: Building operation parameters and equipment operation status data; Spatial zoning information and building envelope heat transfer parameters derived from BIM data; Real-time or forecast meteorological data, including outdoor temperature, humidity and solar radiation information; Traffic conditions and vehicle operation data, including transport routes, onboard energy consumption, and energy storage status; Real-time number of users or occupancy rate and usage schedule for each functional area; Among them, the predictable external disturbance data includes weather forecasts, traffic condition forecasts, and usage schedules of each functional area. The predictable external disturbance data is injected into the state vector of the DRL control module using a forward-looking input method, so that the control strategy can respond to changes in the external environment in advance. The environmental monitoring and execution module has the function of energy consumption monitoring during the construction phase, which is used to collect energy consumption and peak load information of hoisting equipment and temporary facilities. The DRL control module, based on the BIM construction progress and component hoisting plan, jointly optimizes the hoisting sequence, operation period and temporary facility operation strategy, thereby reducing energy consumption and carbon emissions during the construction phase.

10. A self-optimizing energy-saving method for prefabricated buildings based on DRL and BIM, applied to the self-optimizing energy-saving system for prefabricated buildings based on DRL and BIM as described in any one of claims 1-9, characterized in that, The method includes: It stores and provides BIM data of prefabricated buildings, constructs digital twin models of prefabricated buildings based on the BIM data, provides a simulation environment for the DRL control module during the offline training phase, simulates and calculates energy consumption and carbon emission results under different control commands, and assists in the verification and retraining of actual operation strategies during the online operation phase. Deploy environmental sensors and equipment actuators at prefabricated building and construction sites, and communicate with the transportation vehicle management system or vehicle data interface to collect building operation parameters, construction equipment energy consumption and transportation status information in real time during the online operation phase, and adjust the corresponding energy-consuming equipment or transportation scheduling according to control commands. During the offline training phase, simulation calculation results are read, and an optimized control strategy for prefabricated buildings is obtained through multiple rounds of interactive learning. During the online operation phase, real-time data collected is read as input, and corresponding control commands are generated based on the optimized control strategy and sent to the environmental monitoring and execution steps for execution.

Images