Model-data dual-driven building energy management intelligent agent training environment creation method

Abstract

The present invention belongs to the technical field of intelligent building energy management and control, and discloses a model-data dual-driven building energy management and control intelligent agent training environment creation method. The method is implemented by the following steps: First, a thermodynamic mechanism model of the building envelope structure and a simulation model of the HVAC system are constructed, and the two types of models are coupled through a joint simulation interface. Then, the existing building operation data is cleaned and repaired by time series alignment, missing value filling, outlier filtering, etc. to ensure data quality. Subsequently, the joint simulation model is run to generate output data such as hourly load and equipment status. Finally, through spatiotemporal label association and feature extraction, the cleaned and repaired measured data and the model output data are multi-source fused to construct a training database containing multi-dimensional features such as building thermal dynamic response and equipment operation characteristics, providing high-precision, full-scene training data support for the building environmental control intelligent agent.

Description

Technical Field

[0001] The present invention belongs to the technical field of building environment control, and specifically relates to a method for creating a building energy management and control intelligent agent training environment driven by a model and data. Background Art

[0002] The energy consumption of building operation and maintenance in my country accounts for more than 35% of the total social energy consumption. Existing buildings generally have problems such as backward operation and maintenance technology and low level of intelligence. They rely too much on manual experience and traditional rule-based control logic, and are unable to cope with complex and dynamic changes in indoor and outdoor environments, resulting in serious energy waste.

[0003] While artificial intelligence (AI) technology offers new opportunities for building environmental control, and AI-based agents in particular are increasingly being used in the management and control of building environmental control systems with promising results, current agent training environments are mostly purely data-driven, relying on large amounts of historical operating data. However, current operational data for both new and existing buildings often suffers from inconsistent standards, incomplete data types, low data quality, and insufficient data volume for agent learning. Crucially, while existing operational data is available, current agent training lacks the integration of specialized domain knowledge (such as thermodynamic mechanisms and equipment performance constraints). Building environmental control systems face a "data-mechanism" disconnect. Existing control strategies based on existing operational data are based on simple rule-based strategies, which are insufficient to provide high-quality learning data samples and decision-making basis for control agents. These issues make agent training difficult, data-intensive, and result in poor interpretability and generalization of the generated control strategies. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a model-data dual-driven building energy management intelligent agent training environment creation method, which is achieved through the following technical solutions:

[0005] A model-data dual-driven building energy management and control intelligent agent training environment creation method includes the following steps: using EnergyPlus to establish a thermodynamic mechanism model of a building envelope structure; using TRNSYS to establish a simulation model of a heating, ventilation and air-conditioning system; establishing a joint simulation interface between EnergyPlus and TRNSYS to couple the thermodynamic mechanism model of the building envelope structure with the simulation model of the heating, ventilation and air-conditioning system to obtain a joint simulation model; performing data cleaning and data repair on existing building operation data; running the joint simulation model to obtain model output data; and performing multi-source data fusion on the cleaned and repaired existing building operation data and the model output data to generate a database for training a building environmental control intelligent agent.

[0006] Compared with existing technologies, this invention offers the following advantages and benefits: It achieves precise coupling by establishing a joint simulation interface between the EnergyPlus building envelope thermodynamic mechanism model and the TRNSYS HVAC system simulation model. It also cleans, repairs, and calibrates existing building operation data, combines multi-source data fusion to generate a high-quality training database, and embeds expert knowledge graphs to optimize the DRL graybox model. This method significantly improves the joint simulation model's accuracy for both building thermal dynamic response and energy system operation, ensuring logical consistency and spatiotemporal alignment of data. It provides the building environmental control intelligent agent with reliable training data encompassing multi-dimensional features such as load distribution and equipment operating status, effectively enhancing the agent's learning and decision-making capabilities for complex building environments, and helping achieve the combined goals of optimizing building energy consumption, regulating indoor environmental comfort, and improving system operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0008] Figure 1 A flow chart of the method for creating a building energy management and control intelligent agent training environment driven by a model and data provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0009] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the examples. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention. The embodiments described below are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0010] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that these specific details are not necessarily required to practice the present invention. In other examples, well-known structures, materials, or methods are not specifically described to avoid obscuring the present invention. The materials, instruments, and reagents used in the following examples, unless otherwise specified, are commercially available. The techniques used in the examples, unless otherwise specified, are conventional techniques well known to those skilled in the art.

[0011] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0012] Example: A model-data dual-driven building energy management intelligent agent training environment creation method is proposed, including Figure 1 The following steps are shown:

[0013] Step 1: Use EnergyPlus to establish a thermodynamic mechanism model of the building envelope.

[0014] EnergyPlus is widely used in building performance analysis, HVAC system design, energy conservation assessment, and research. It simulates a building's energy consumption, indoor environmental quality, and the operation of various systems (such as HVAC, lighting, and water heating). The thermodynamic mechanism model of the building envelope is a mathematical model based on thermodynamic principles. It comprehensively considers the physical properties of the envelope (such as the thermal conductivity, thickness, and thermal resistance of the material) and external environmental factors (such as temperature, solar radiation, and wind speed). It is used to describe and predict the heat transfer behavior of the building envelope (such as exterior walls, roofs, floors, doors, and windows) under different environmental conditions.

[0015] This step specifically includes:

[0016] Step 1.1: Obtain the building's geometric parameters, functional zoning information, thermal parameters of each enclosure structure, and multiple thermal bridge areas by parsing the building drawings.

[0017] Among them, the geometric characteristic parameters include: plane dimensions, floor height, orientation, window-to-wall ratio and facade structure; the functional zoning information includes: multiple functional areas and the area, volume and adjacent interfaces of each functional area; the thermal parameters include: the thickness, thermal conductivity, specific heat capacity and density of each material structural layer.

[0018] This step is specifically implemented by parsing architectural drawings to extract geometric information and then deconstruct the building envelope. Extracting geometric information involves obtaining geometric parameters such as building dimensions, floor height, orientation, window-to-wall ratio, and facade structures (e.g., concave and convex balconies, sunshades), from the architectural drawings. Deconstructing the building envelope involves identifying functional areas (e.g., offices, conference rooms, bedrooms, corridors), and noting the area, volume, and adjacent interfaces (exterior walls, interior walls, floors, roofs, etc.) of each area.

[0019] Step 1.2: Collect operation and maintenance data.

[0020] Operation and maintenance data includes historical energy consumption data, indoor environmental data, and equipment operation data. Historical energy consumption data includes hourly electricity and gas consumption, peak load periods, and operating modes for each energy-consuming device. Indoor environmental data includes temperature, humidity, year-round temperature fluctuations for each functional area, and the control mode of the air conditioning system. Equipment operation data includes occupancy density, occupancy activity types, power consumption of each heat source device, type of each heat source device, distribution density of each heat source device, and operating mode of each heat source device. Furthermore, for historical energy consumption data, it is necessary to collect hourly energy consumption records for the building (such as electricity and gas consumption), distinguishing energy consumption by air conditioning, lighting, and equipment, and identifying peak load periods and typical operating modes. For indoor environmental data, it is necessary to organize temperature and humidity monitoring data (such as year-round temperature fluctuations in different areas) and analyze the existing air conditioning system control strategy (such as set temperature ranges and start and stop times). For equipment operation data, it is necessary to collect actual operating parameters of key heat source devices (such as occupancy density, equipment power, and lighting power density), including per capita office space, computer power, and lighting type and density.

[0021] Step 1.3: Perform the following steps in EnergyPlus:

[0022] Step 1.3.1: Based on the geometric parameters, functional zoning information, thermal parameters of each enclosure structure, and the linear thermal bridge coefficient of each thermal bridge area, thermal parameter modeling of the enclosure structure is performed to obtain the enclosure structure model.

[0023] The enclosure structure model includes the heat transfer coefficient and solar heat gain coefficient of each enclosure structure.

[0024] (1) For exterior walls, roofs or floors

[0025] Use the Construction module of EnergyPlus to input the parameters of each layer in the order of material layers, and the software will automatically calculate the total heat transfer coefficient (1), in formula (1), U represents the overall heat transfer coefficient, d i Indicates the material thickness, Represents thermal conductivity, R in Indicates the internal surface heat transfer resistance, R out Represents the external surface heat transfer resistance.

[0026] In addition, for thermal bridge locations such as balconies and corner walls, the total heat transfer coefficient is corrected by the linear thermal bridge coefficient. The correction formula is: (2), in formula (2), U total represents the corrected overall heat transfer coefficient of the enclosure structure, U baserepresents the heat transfer coefficient of the non-thermal bridge area, A base represents the area of ​​the non-thermal bridge area, A total Represents the overall area of ​​the enclosure structure, represents the linear thermal bridge coefficient, L Indicates the thermal bridge length.

[0027] (2) For exterior windows or glass

[0028] First, use the EnergyPlus WindowMaterial:Glass module to define the glass type (e.g., single-glazed glass, Low-E insulating glass) and input the visible light transmittance. Then, use the EnergyPlus Window:Frame module to set the window frame material (e.g., aluminum alloy / wood), frame ratio, and linear heat transfer coefficient. The software will automatically calculate the heat transfer absorption and solar heat gain coefficient for the entire window.

[0029] Step 1.3.2: Model the indoor load distribution based on geometric parameters, functional zoning information, historical energy consumption data, and equipment operation data to obtain an indoor load distribution model.

[0030] The indoor load distribution model includes a load distribution cloud diagram and a time series curve.

[0031] First, input the density of people in each area (e.g. office 5m 2 / person), activity type (sitting or light work), the software automatically calculates the sensible heat and latent heat release rates.

[0032] Then, use the ElectricEquipment module of EnergyPlus to define the equipment power density (such as 20W / m 2 、Printer 15W / m 2 ), distinguish between continuously running devices (such as servers) and intermittent devices (such as coffee machines), and associate device on and off schedules.

[0033] Next, enter the lighting power density (e.g. 15W / m² for office) through the Lights module of EnergyPlus. 2 ), calculate the heat dissipation ratio based on lamp efficiency (generally 80% of the heat dissipated by fluorescent lamps and 90% of the heat dissipated by LED lamps), and associate it with natural lighting control strategies (such as automatic dimming when the illumination is higher than 300 lux).

[0034] Finally, after running the EnergyPlus simulation, the hourly load data (sensible heat, latent heat, and total heat) of each area is extracted through the Output:Variable module to generate a load distribution cloud map or time series curve.

[0035] Step 1.3.3: Perform air conditioning zoning modeling based on geometric parameters, functional zoning information, indoor environmental data, historical energy consumption data, indoor environmental data, and equipment operation data to obtain an air conditioning zoning model.

[0036] First, use the Zone module of EnergyPlus to define each air conditioning zone, associate it with the room number in the geometric model, and set the zone type (such as "ExteriorZone_North" or "InteriorZone").

[0037] Then, for each zone, specify: the envelope construction (via the Construction association), the internal heat source definition (People / ElectricEquipment / Lights associated zones), and the HVAC system type (such as HVAC:Zone:Baseboard or HVAC:VariableRefrigerantFlow).

[0038] Finally, by simulating the load curves of different partitions, check whether there is significant thermal coupling between adjacent partitions (for example, if the temperature difference is greater than 2°C, the partition boundary needs to be adjusted).

[0039] Step 2: Use TRNSYS to build a simulation model of the HVAC system.

[0040] TRNSYS is widely used in building environmental control, HVAC system design, renewable energy system analysis, and building energy efficiency assessment. It implements device modeling through modular components (Type). To create a simulation model of an HVAC system using TRNSYS, you create device models for each energy system device by selecting modular components, define their performance curves, embed their control logic, and interconnect their device models. The following explains the specific implementation of this step using the examples of a chiller, a water pump, and a cooling tower.

[0041] (1) Establish equipment models for each energy system equipment by selecting modular components

[0042] For chillers: First, select a chiller model, such as Type 108 (piston / centrifugal chiller), Type 65 (custom performance curve chiller), or Type 557 (chiller model based on the EPRI database); then, enter model parameters, such as rated cooling capacity (kW), rated COP, refrigerant type, compressor type (centrifugal / screw, etc.), and design operating condensing temperature / evaporating temperature; finally, drag the chiller model into the TRNSYS graphical interface, enter the model parameters in the parameter dialog box, and the software will automatically build the chiller model.

[0043] For water pumps: First, select a water pump model, such as Type 31 (general water pump model) or Type 104 (variable speed water pump model); then, enter equipment parameters, such as rated flow (m³ / h), rated head (m), motor efficiency, impeller diameter, and pump curve polynomial coefficient (head-flow relationship); finally, drag the water pump model into the TRNSYS graphical interface, enter the model parameters in the parameter dialog box, and the software will automatically build the water pump model.

[0044] For cooling towers: First, select a cooling tower model, such as Type 303 (counterflow cooling tower model) or Type 212 (enthalpy difference-based cooling tower model); then, enter equipment parameters, such as circulating water volume (m³ / h), inlet water temperature (°C), outlet water temperature (°C), wet-bulb temperature (°C), fan power (kW), and filler characteristic parameters (such as enthalpy transfer coefficient); finally, drag the cooling tower model into the TRNSYS graphical interface, enter the model parameters in the parameter dialog box, and the software will automatically establish the cooling tower model.

[0045] (2) Define the performance curve of each energy system equipment

[0046] For chillers, define a curve showing how the chiller's cooling efficiency ratio (COP) changes with load factor. This can be achieved through polynomial fitting or piecewise linear interpolation. The specific method is: first, collect performance tables provided by the equipment manufacturer (e.g., COP at full load, COP at 75% / 50% / 25% load); then, fit the measured data (e.g., regressing the COP-load factor curve using historical operating data).

[0047] For water pumps, define the pump head-flow curve. The relationship between head (H) and flow (Q) can be expressed by a polynomial: H=H 0 -S×Q 2 (3), in formula (3), H Indicates the pump head, H 0 means rated lift, S Indicates the pipe resistance coefficient. In TRNSYS, enter the polynomial coefficients through the parameter column of the Type31 component.

[0048] For cooling towers, define a cooling tower heat dissipation curve to show how cooling efficiency varies with wet-bulb temperature and circulating water volume. Enter the cooling capacity attenuation coefficient at different wet-bulb temperatures in the Type303 component to calculate the actual outlet water temperature.

[0049] (3) Embed control logic of each energy system equipment

[0050] For chillers, the number of chillers started and stopped, as well as load distribution, must be dynamically adjusted based on real-time load. Common control logics include load percentage control and efficiency priority control. Furthermore, load percentage control logic: when the cumulative load exceeds 80% of a chiller's rated load, the next chiller is started; when the load falls below 20% of a chiller, a chiller is shut down. In TRNSYS, a Type2d component (logic controller) calculates the total load in real time (such as the hourly cooling load output by EnergyPlus), compares it with the capacity of each chiller, and outputs a start / stop signal (0 / 1). Efficiency priority control logic prioritizes the operation of chillers with the highest COP. By comparing the COPs of each chiller at the current load rate, a combination is selected to minimize total energy consumption. In TRNSYS, a Type155 component (optimization controller) is used to dynamically distribute load based on load distribution algorithms (such as the equal efficiency method).

[0051] The control objective for the water pump is to maintain a constant chilled water supply and return temperature difference (e.g., 5°C) and to vary the flow rate by adjusting the pump frequency. The control logic includes real-time monitoring of the supply and return temperature difference and calculation of the required flow rate. In TRNSYS, the pump frequency is adjusted using a Type 213 component (PID controller) to keep the actual flow rate close to the required flow rate.

[0052] For cooling towers, the control objective is to adjust the cooling tower fan frequency or start and stop it based on the condensing temperature to ensure the chiller's condensing pressure remains within the high-efficiency range. The control logic is: when the condensing temperature exceeds a set point (e.g., 32°C), the cooling tower air volume is increased (by raising the fan frequency or activating a backup tower); when it falls below a lower limit (e.g., 25°C), the air volume is reduced or the system is shut down. In TRNSYS, Type2d compares the real-time condensing temperature with a threshold and outputs a fan start / stop signal or frequency adjustment signal (0-10V analog).

[0053] Step 3: By establishing a joint simulation interface between EnergyPlus and TRNSYS, the thermodynamic mechanism model of the building envelope structure and the simulation model of the HVAC system are coupled to obtain a joint simulation model.

[0054] The co-simulation interface can be a TRNBuild interface, a dynamic link library, or a Socket interface based on the TCP transmission protocol.

[0055] First, a joint simulation architecture was established, encompassing the building, energy system, and interface sides. On the building side, EnergyPlus calculates the heat conduction of the building envelope, indoor loads, and the dynamic response of temperature and humidity. Data delivery targets output loads, temperature, and humidity. On the energy system side, TRNSYS simulates the operational control logic of various energy system devices (such as chillers, pumps, and cooling towers) and calculates energy consumption. Data exchange targets input device status data and device energy consumption data. On the interface side, EnergyPlus facilitates bidirectional data exchange, while TRNSYS handles time step synchronization, data format conversion, and exception handling. Data delivery targets bidirectional real-time interaction.

[0056] Next, select a coupling mode. These include loose coupling (file and interactive) and tight coupling (real-time data exchange). Loose coupling uses CSV / JSON files to periodically exchange data (time steps ≥ 1 minute) and is suitable for steady-state or quasi-dynamic simulations. Tight coupling uses Socket / TCP / IP or shared memory to synchronize data on a time-step basis (e.g., 1-second level), supporting dynamic feedback control.

[0057] Finally, the data interaction interface is set up. This involves defining the data interaction protocol, the time step synchronization mechanism, and the software interface. Defining the data interaction protocol includes defining core parameters and data format specifications. For example, building-side output parameters (hourly cooling / heating loads, indoor temperature and humidity, fresh / exhaust air volume, and solar heat gain) and energy system-side output parameters (chiller operating status, chilled / cooled water supply and return temperatures, pump / fan frequency / flow rate, and equipment energy consumption) are defined, and the data format specification is encapsulated in JSON format. Furthermore, the time step synchronization mechanism is defined, including either a master-slave mode or an event-driven mode. The master-slave mode prioritizes the EnergyPlus time step (typically 1 minute), with TRNSYS operating at the same step length and processing high-frequency data through interpolation (e.g., 1-second device response). The event-driven mode triggers data interaction when the building load change rate exceeds a threshold (e.g., ±5%), reducing ineffective communication. Furthermore, the software interface development includes: on the EnergyPlus side, defining output variables (such as adding Output:Variable in the .idf file to export key load data), starting the simulation through the EnergyPlusAPI or RunManager, and triggering the data sending function after each time step; on the TRNSYS side, in Type99 (user-defined module), writing a data receiving interface in Fortran / C++, parsing JSON data and driving the device model, and in Type213 (empty control module), linking the device operation logic with the building load in real time.

[0058] It should be noted that after obtaining the co-simulation model, this embodiment reserves a data-driven model extension interface for the co-simulation model. This is intended to enable the co-simulation model to flexibly integrate data-driven models (such as machine learning and deep learning models), addressing the limitations of traditional mechanism-based models in dynamic and complex scenarios and enhancing the ability to capture nonlinear characteristics of the building environment (such as sudden changes in user behavior and equipment failures).

[0059] The method for reserving the data-driven model extension interface is:

[0060] First, define the interface functional modules, including: data input interface, model call interface, and parameter feedback interface. The data access interface supports real-time data (such as subscribing to sensor data via MQTT) and batch data (such as importing CSV files), and is used to update model boundary conditions (such as real-time outdoor temperature and occupancy density). The model call interface allows the data-driven model to call the calculation functions of the co-simulation model (such as hourly load calculation in EnergyPlus) or obtain intermediate simulation results (such as air conditioning system energy efficiency parameters in TRNSYS). The parameter feedback interface allows the data-driven model to transfer optimization parameters (such as the heat transfer coefficient correction value of the building envelope and the air conditioning set temperature threshold) to the co-simulation model, achieving closed-loop simulation optimization.

[0061] Then, we develop interface adapters. We develop lightweight encapsulated components for data-driven models. For example, we package TensorFlow models trained in Python into Docker containers and provide prediction services through HTTP interfaces; we compile MATLAB optimization algorithms into dynamic link libraries (DLLs) and call them through the interface layer.

[0062] Step 4: Clean and repair the existing building operation data.

[0063] Existing building operation data includes: historical energy consumption data, indoor environment data and equipment operation data.

[0064] Data cleaning includes:

[0065] Use the Network Time Protocol to align the timing of existing building operation data. For example, select an NTP server and configure NTP for all sensors and devices that require time synchronization, enabling them to communicate with the selected NTP server and synchronize time.

[0066] Fill missing values ​​in existing building operation data using linear interpolation. For example, use data analysis tools (such as Pandas) to detect missing values ​​in chiller power data.

[0067] By setting mutation thresholds, outliers in existing building operation data can be screened and filtered. For example, data that exceeds the physical operating range of the equipment (such as chiller load rate > 100%) can be eliminated.

[0068] The Z-score standardization algorithm and Min-Max standardization algorithm are used to standardize the existing building operation data.

[0069] The data repair method is as follows: divide the existing building operation data into training set and test set; input the training set into the decision tree model or random forest model for model training, and output the false alarm detection model; input the test set into the false alarm detection model, and output the abnormal equipment status label; modify the abnormal equipment status label.

[0070] In addition, this method also uses the operation data of existing buildings after cleaning and repair to calibrate the model parameters of the joint simulation model.

[0071] Step 5: Run the co-simulation model to obtain model output data.

[0072] Step 6: Perform multi-source data fusion on the cleaned and repaired existing building operation data and the model output data to generate a database for training the building environmental control intelligent agent.

[0073] The purpose of this step is to break the "data silos" between building physical models, equipment operation data, and environmental monitoring data, and to build a closed-loop feedback system of "physical model + real-time data" to support accurate simulation, fault diagnosis, and optimized control of building energy systems.

[0074] The method of multi-source data fusion is:

[0075] Step 6.1: Add spatiotemporal labels.

[0076] Spatiotemporal tags are designed to accurately locate data and should be unique, hierarchical, and scalable. Uniqueness means each data point is assigned a unique label to avoid ambiguity (for example, distinguishing between "air supply volume of AC box A on the first floor" and "air supply volume of AC box B on the first floor"). Hierarchy refers to the hierarchical structure of data, supporting flexible retrieval (for example, using a five-level structure of "building-area-device-parameter-time"). Scalability means reserving fields to accommodate future equipment additions or parameter expansions (for example, using "AreaID" instead of fixed floor names).

[0077] In this embodiment, time and space include spatial dimension information, device identification information, parameter type information, and time range information. For example, in B1F_Chiller1_Power_2023Summer, BIF represents the time dimension information (e.g., representing the first underground floor), Chiller1 represents the device identification information (indicating chiller unit 1), Power represents the parameter type information (indicating electrical power), and 2023Summer represents the time range information (indicating the summer of 2023).

[0078] The method for adding time tags is: based on Python regular expressions or Excel pivot tables, automatically add tags to batch data, and create a "metadata" table in MySQL / PostgreSQL to store the mapping relationship between tags and data file paths.

[0079] Step 6.2: Extract feature information from the model parameters of the co-simulation model.

[0080] For example, the extracted feature information includes: equipment operation characteristics, indoor environment characteristics, energy flow characteristics, and meteorological coupling characteristics. Equipment operation characteristics include chiller load factor, COP, and pump efficiency; indoor environment characteristics include temperature and humidity distribution and PMV / PPD index; energy flow characteristics include hourly cooling load and proportion of cooling and heating transmission energy consumption; and meteorological coupling characteristics include outdoor temperature and humidity and solar radiation intensity.

[0081] Step 6.3: Establish a mapping relationship between the model parameters of the joint simulation model and the existing building operation data based on the spatiotemporal labels and characteristic information.

[0082] The specific method is:

[0083] Step 6.3.1: Create an initial mapping. Generate an initial mapping table through manual annotation or automatic association of BIM models (e.g., matching equipment IDs in Revit with component IDs in the simulation model).

[0084] Step 6.3.2: Dynamic Calibration. Use historical data to calculate the deviation between the model prediction value and the measured value (such as the root mean square error (RMSE)) and automatically adjust the mapping relationship (such as correcting the model area corresponding to the sensor installation location).

[0085] Step 7: Embed the expert knowledge graph into the gray-box model of DRL.

[0086] The goal of this step is to improve the model's interpretability and generalization capabilities. Specifically, deep reinforcement learning (DRL) models are typically black-box models, making their decision-making processes difficult to understand. However, expert knowledge graphs encompass specialized knowledge and experience in the field of building environmental control. Embedding them within the DRL gray-box model allows the model's decision-making process to be linked to domain knowledge. For example, in building environmental control, the expert knowledge graph contains knowledge about optimal temperature and humidity settings for different seasons and indoor and outdoor conditions. This knowledge can be used as a reference when the DRL model makes control decisions, making the decision-making process more transparent and explainable, making it easier for users to understand and trust the model's output. Furthermore, the building environment is complex and ever-changing, with different building types, locations, and usage scenarios all impacting environmental control systems. Expert knowledge graphs encompass a wealth of domain knowledge and strategies for various scenarios. Embedding them within the DRL gray-box model allows the model to learn from a wider range of knowledge and experience, enabling better generalization and making informed decisions when faced with new and unseen environments and working conditions. For example, when encountering buildings in different climate zones, the model can adjust based on the experience of similar areas in the expert knowledge graph to improve the adaptability of the control strategy.

[0087] The specific method is:

[0088] First, relevant knowledge is collected from multiple channels, including professional literature, specifications, standards, and expert experience in the field of building environmental control. For example, knowledge on the thermal performance of building envelopes, the operating principles and control strategies of HVAC systems, and the environmental requirements of different functional areas is collected.

[0089] The extracted knowledge is then represented in the form of triples (entity-relationship-entity), for example, ("air conditioning system", "optimal operating temperature range", "22℃-26℃").

[0090] Next, the represented knowledge is stored in a graph database (such as Neo4j) to build an expert knowledge graph.

[0091] Next, use graph embedding algorithms (such as Node2Vec, DeepWalk, etc.) to convert the nodes and relationships in the expert knowledge graph into low-dimensional vector representations.

[0092] It should be noted that these vector representations can preserve the structural and semantic information in the knowledge graph, making it easier to process in the DRL model. For example, nodes such as "air conditioning system" and "temperature setting" and the relationships between them can be converted into vectors, and then these vectors can be spliced ​​with the input features of the DRL model.

[0093] Finally, the model is optimized using joint training.

[0094] During training, both the DRL reward function and the constraints of the expert knowledge graph are considered. For example, when calculating rewards, not only the actual effects of building environmental control (such as reduced energy consumption and improved comfort) must be considered, but also whether the model's decisions are consistent with the knowledge in the expert knowledge graph.

[0095] It should be understood that the terms "system," "device," "unit," and / or "module" used in this specification are a method for distinguishing different components, elements, parts, portions, or assemblies at different levels. However, if other terms can achieve the same purpose, the terms may be replaced by other expressions.

[0096] As used in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not refer to the singular but also include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0097] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0098] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for understanding and reading by those familiar with this technology, and are not used to limit the conditions for implementation of the present invention. Therefore, they have no substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the efficacy and purpose of the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle", etc. quoted in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments in their relative relationships should also be regarded as the scope of implementation of the present invention without substantially changing the technical content.

Claims

1. A model-data dual-driven building energy management intelligent agent training environment creation method, characterized by: The following steps are involved: Use EnergyPlus to establish a thermodynamic mechanism model of building envelope structures; Use TRNSYS to build a simulation model of the HVAC system. By establishing a joint simulation interface between EnergyPlus and TRNSYS, the thermodynamic mechanism model of the building envelope and the simulation model of the HVAC system are coupled to obtain a joint simulation model. Perform data cleaning and data repair on existing building operation data; Run the co-simulation model to obtain model output data; The cleaned and repaired existing building operation data is integrated with the model output data to generate a database for training the building environmental control intelligent agent; Using EnergyPlus to establish a thermodynamic mechanism model of the building envelope includes the following steps: By analyzing architectural drawings, the building's geometric parameters, functional zoning information, thermal parameters of each enclosure structure, and multiple thermal bridge areas are obtained; Obtain historical energy consumption data, indoor environment data, and equipment operation data; Perform the following steps in EnergyPlus: Based on the geometric parameters, functional zoning information, thermal parameters of each enclosure structure and the linear thermal bridge coefficient of each thermal bridge area, the enclosure structure thermal parameter modeling is carried out to obtain the enclosure structure model; Indoor load distribution modeling is performed based on geometric parameters, functional zoning information, historical energy consumption data, and equipment operation data to obtain an indoor load distribution model; Air conditioning zoning modeling is performed based on geometric parameters, functional zoning information, indoor environmental data, historical energy consumption data, indoor environmental data and equipment operation data to obtain an air conditioning zoning model; The method of using TRNSYS to establish a simulation model of the HVAC system is as follows: in TRNSYS, the device model of each energy system device is established by selecting modular components, the performance curve of each energy system device is defined, the control logic of each energy system device is embedded, and the device models of each energy system device are interconnected; Embedding expert knowledge graph into a gray-box model for DRL.

2. The method for creating a building energy management and control agent training environment based on a model-data dual-driven approach according to claim 1 is characterized in that: The envelope structure model includes the heat transfer coefficient and solar heat gain coefficient of each envelope structure; the indoor load distribution model includes the load distribution cloud map and time series curve.

3. The method for creating a building energy management and control agent training environment based on a model-data dual-driven approach according to claim 1 or 2, characterized in that: Geometric parameters include: floor plan dimensions, floor height, orientation, window-to-wall ratio, and facade construction; Functional zoning information includes: multiple functional areas and the area, volume and adjacent interfaces of each functional area; Thermal parameters include: thickness, thermal conductivity, specific heat capacity and density of each material structural layer; Historical energy consumption data includes: hourly electricity consumption, hourly gas consumption, peak load period and operating mode of each energy-consuming device; Indoor environmental data includes: temperature, humidity, temperature fluctuations of each functional area throughout the year, and the control mode of the air conditioning system; The equipment operation data includes: personnel density, personnel activity type, power of each heat source equipment, type of each heat source equipment, distribution density of heat source equipment and operation mode of each heat source equipment.

4. The method for creating a building energy management and control agent training environment based on a model-data dual-driven approach according to claim 1 or 2, characterized in that: The joint simulation interface includes: TRNBuild interface, dynamic link library or Socket interface based on TCP transmission protocol.

5. The method for creating a building energy management and control agent training environment driven by both model and data according to claim 1 or 2, characterized in that: Data cleaning includes the following steps: using the network time protocol to align the time series of existing building operation data, filling missing values ​​in the existing building operation data through linear interpolation, screening and filtering outliers in the existing building operation data by setting a mutation judgment threshold, and standardizing the existing building operation data using the Z-score normalization algorithm and the Min-Max normalization algorithm; Data repair includes the following steps: dividing the existing building operation data into training sets and test sets; inputting the training set into a decision tree model or a random forest model for model training, and outputting a false alarm detection model; inputting the test set into the false alarm detection model, and outputting abnormal equipment status labels; and modifying the abnormal equipment status labels.

6. The method for creating a building energy management and control agent training environment based on a model-data dual-driven approach according to claim 1 or 2, characterized in that: Multi-source data fusion includes the following steps: adding spatiotemporal tags; the spatiotemporal tags contain: spatial dimension information, equipment identification information, parameter type information and time range information; extracting feature information from the model parameters of the joint simulation model; establishing a mapping relationship between the model parameters of the joint simulation model and the existing building operation data based on the spatiotemporal tags and feature information.

7. The method for creating a building energy management and control agent training environment driven by both model and data according to claim 1 or 2, characterized in that: After the joint simulation model is obtained, the following steps are also included: reserving a data-driven model extension interface for the joint simulation model.

8. The method for creating a building energy management and control intelligent agent training environment driven by both model and data according to claim 1 or 2, characterized in that: After the existing building operation data is cleaned and repaired, the following steps are also included: using the cleaned and repaired existing building operation data to calibrate the model parameters of the joint simulation model.

Images