Multi-modal fusion building energy consumption model construction method and energy consumption prediction system

Abstract

The present application relates to the field of energy consumption prediction, in particular to a multi-modal fusion building energy consumption model construction method and an energy consumption prediction system, according to the historical activity image and the environmental historical perception data in the building, the activity behavior labeled data set and the environmental perception labeled data set are generated, and are cross related with the building structure data, the multi-modal knowledge graph is obtained; according to the real-time perception information of the environment, the matched knowledge set is retrieved from the multi-modal knowledge graph, so as to generate the heat balance disturbance variable; according to the real-time perception information of the environment and the device configuration information, the building basic energy consumption model is formed; the heat balance disturbance variable is used to correct the building basic energy consumption model, the energy consumption space-time change trend is obtained, and the device operation constraint strategy is formed. The present application can form multi-modal and multi-directional energy consumption evaluation of the building by fusing building structure, building internal activity behavior and environmental perception parameters, and improve the prediction control flexibility and accuracy.

Description

Technical Field

[0001] This invention relates to the field of energy consumption prediction, and in particular to a method for constructing a multimodal fusion building energy consumption model and an energy consumption prediction system. Background Technology

[0002] With the advancement of urbanization, high-rise office buildings and other structures are experiencing rapid growth. Research indicates that building operation energy consumption accounts for a significant share of total social energy consumption. Accurate prediction of building energy consumption is crucial for energy conservation, emission reduction, and lowering operating costs. Advances in data acquisition technologies and high-precision sensors provide a comprehensive and accurate data foundation for building energy consumption prediction.

[0003] Building energy consumption mainly includes energy consumption for internal temperature regulation, which is affected by environmental factors such as sunlight. Existing building energy consumption modeling methods mainly include physical modeling, machine learning modeling, and statistical modeling. Among them, physical modeling uses parameters such as building BIM and material properties to conduct thermodynamic simulation and dynamic analysis, which is time-consuming and computationally expensive. Machine learning modeling uses artificial neural networks to perform nonlinear mapping predictions of buildings, which has strict requirements for the selection of modeling parameters. Inappropriate parameter selection can easily lead to overfitting or underfitting, affecting the accuracy of model predictions. Statistical modeling usually uses multiple regression models as empirical models, which require sufficient and accurate historical data as a basis. It has high data requirements and is only applicable to energy consumption prediction in the building BIM design stage, and cannot be applied to the entire life cycle of actual building operation. Therefore, there is an urgent need for a building energy consumption prediction method that can be applied to the entire life cycle of building operation, has low data requirements, and can be dynamically adjusted. Summary of the Invention

[0004] Considering the limitations of existing building energy consumption prediction methods in terms of application stage and modeling simulation time, as well as the high requirements for data type and accuracy, it is impossible to form a dynamic, controllable, and widely applicable energy consumption prediction method.

[0005] To address the above problems, this invention provides a method for constructing a multimodal fusion building energy consumption model, the method comprising the following steps:

[0006] S100: Perform semantic recognition of historical activity images inside the building to obtain a labeled dataset of activity behaviors; collect and process the building's environmental history perception data to generate an environmental perception labeled dataset.

[0007] S200: Cross-correlate the building structure dataset, the activity behavior labeled dataset, and the environmental perception labeled dataset to obtain a multimodal knowledge graph; retrieve a matching knowledge set from the multimodal knowledge graph based on the real-time environmental perception information of the building;

[0008] S300: Based on the real-time environmental sensing information and the building's equipment configuration information, form a basic building energy consumption model; generate thermal balance disturbance variables based on the knowledge set;

[0009] S400: Based on the thermal balance disturbance variables, perform spatiotemporal correction on the building foundation energy consumption model to obtain the spatiotemporal trend of the building's energy consumption, thereby forming an equipment operation constraint strategy.

[0010] Preferably, in S100, semantic recognition of historical activity images inside the building is performed to obtain a labeled dataset of activity behaviors; environmental historical perception data of the building is collected and processed to generate an environmental perception labeled dataset, specifically as follows:

[0011] Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections;

[0012] Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal balance in each subspace, and the activity behavior data has spatiotemporal labels;

[0013] Collect environmental condition sensing data of the building's surrounding environment; wherein, the environmental condition sensing data includes sunshine data and atmospheric airflow data;

[0014] Spatiotemporal calibration is performed on the environmental condition sensing data to generate an environmental sensing labeled dataset; wherein, the environmental sensing labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

[0015] Preferably, in S200, the building structure dataset, the activity behavior labeled dataset, and the environmental perception labeled dataset are cross-linked to obtain a multimodal knowledge graph; based on the real-time environmental perception information of the building, a matching knowledge set is retrieved from the multimodal knowledge graph, specifically as follows:

[0016] Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset;

[0017] Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain a first knowledge graph; wherein, the first knowledge graph includes spatiotemporal association knowledge between the activity behaviors affecting thermal balance within each subspace and the internal range of each subspace;

[0018] A second knowledge graph is obtained by performing spatiotemporal cross-association on the building structure dataset and the environmental perception label dataset; wherein, the second knowledge graph includes spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to;

[0019] The first knowledge graph and the second knowledge graph are spatiotemporally correlated and superimposed to obtain a multimodal knowledge graph;

[0020] The real-time environmental perception information of the building is parsed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field;

[0021] Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes activity behavior knowledge associated with the multidimensional environmental condition field in each subspace.

[0022] Preferably, in S300, a building basic energy consumption model is formed based on the real-time environmental sensing information and the building's equipment configuration information; and a thermal balance disturbance variable is generated based on the knowledge set, specifically:

[0023] Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental perception information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined.

[0024] Based on the equipment configuration information of the building, determine the dynamic characteristics of temperature control power consumption in each subspace inside the building;

[0025] A large model is constructed based on the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption to form a building basic energy consumption model.

[0026] Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects.

[0027] Based on the atmospheric environmental disturbances caused by the changes in the physical state, thermal balance disturbance variables are generated; wherein, the thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

[0028] Preferably, in S400, based on the thermal balance disturbance variable, a spatiotemporal correction is performed on the building foundation energy consumption model to obtain the spatiotemporal trend of the building's energy consumption, thereby forming an equipment operation constraint strategy, specifically:

[0029] The heat exchange flow variables included in the thermal balance disturbance variables are subjected to spatiotemporal dynamic transformation to obtain the potential heat flow field;

[0030] Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal trend of energy consumption during the restoration of the original thermal balance in each subspace inside the building.

[0031] Based on the spatiotemporal variation trend of energy consumption, the potential overload operation domain distribution characteristics of the temperature control equipment inside the building are predicted; based on the time domain distribution characteristics, an operation constraint strategy for the temperature control equipment is formed; wherein, the operation constraint strategy includes overload operation early warning and load stabilization strategy.

[0032] On the other hand, the present invention provides a multimodal fusion building energy consumption prediction system, the system comprising the following modules:

[0033] The first dataset generation module is used to perform semantic recognition of historical activity images inside buildings to obtain a labeled dataset of activity behaviors.

[0034] The second dataset generation module is used to collect and process the building's environmental history perception data to generate an environmental perception labeled dataset.

[0035] The knowledge graph generation module is used to cross-associate the building structure dataset, the activity behavior labeled dataset, and the environment perception labeled dataset to obtain a multimodal knowledge graph;

[0036] The retrieval module is used to retrieve a matching knowledge set from the multimodal knowledge graph based on the real-time environmental perception information of the building.

[0037] The model building module is used to form a basic building energy consumption model based on the real-time environmental sensing information and the building's equipment configuration information.

[0038] The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set.

[0039] The energy consumption prediction module is used to perform spatiotemporal correction on the building foundation energy consumption model based on the thermal balance disturbance variables, so as to obtain the spatiotemporal trend of the building's energy consumption and thereby form an equipment operation constraint strategy.

[0040] Preferably, the first dataset generation module is used to perform semantic recognition of historical activity images inside the building to obtain a labeled dataset of activity behaviors, specifically:

[0041] Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections;

[0042] Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal balance in each subspace, and the activity behavior data has spatiotemporal labels;

[0043] The second dataset generation module is used to collect and process the building's historical environmental perception data to generate an environmental perception labeled dataset, specifically:

[0044] Collect environmental condition sensing data of the building's surrounding environment; wherein, the environmental condition sensing data includes sunshine data and atmospheric airflow data;

[0045] Spatiotemporal calibration is performed on the environmental condition sensing data to generate an environmental sensing labeled dataset; wherein, the environmental sensing labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

[0046] Preferably, the graph generation module is used to cross-associate the building structure dataset, the activity behavior labeled dataset, and the environment perception labeled dataset to obtain a multimodal knowledge graph, specifically:

[0047] Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset;

[0048] Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain a first knowledge graph; wherein, the first knowledge graph includes spatiotemporal association knowledge between the activity behaviors affecting thermal balance within each subspace and the internal range of each subspace;

[0049] A second knowledge graph is obtained by performing spatiotemporal cross-association on the building structure dataset and the environmental perception label dataset; wherein, the second knowledge graph includes spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to;

[0050] The first knowledge graph and the second knowledge graph are spatiotemporally correlated and superimposed to obtain a multimodal knowledge graph;

[0051] The retrieval module is used to retrieve a matching knowledge set from the multimodal knowledge graph based on the real-time environmental perception information of the building. Specifically:

[0052] The real-time environmental perception information of the building is parsed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field;

[0053] Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes activity behavior knowledge associated with the multidimensional environmental condition field in each subspace.

[0054] Preferably, the model building module is used to form a building basic energy consumption model based on the real-time environmental sensing information and the building's equipment configuration information, specifically:

[0055] Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental perception information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined.

[0056] Based on the equipment configuration information of the building, determine the dynamic characteristics of temperature control power consumption in each subspace inside the building;

[0057] A large model is constructed based on the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption to form a building basic energy consumption model.

[0058] The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set, specifically:

[0059] Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects.

[0060] Based on the atmospheric environmental disturbances caused by the changes in the physical state, thermal balance disturbance variables are generated; wherein, the thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

[0061] Preferably, the energy consumption prediction module is used to perform spatiotemporal correction on the building foundation energy consumption model based on the thermal balance disturbance variable, to obtain the spatiotemporal trend of the building's energy consumption, thereby forming an equipment operation constraint strategy, specifically:

[0062] The heat exchange flow variables included in the thermal balance disturbance variables are subjected to spatiotemporal dynamic transformation to obtain the potential heat flow field;

[0063] Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal trend of energy consumption during the restoration of the original thermal balance in each subspace inside the building.

[0064] Based on the spatiotemporal variation trend of energy consumption, the potential overload operation domain distribution characteristics of the temperature control equipment inside the building are predicted; based on the time domain distribution characteristics, an operation constraint strategy for the temperature control equipment is formed; wherein, the operation constraint strategy includes overload operation early warning and load stabilization strategy.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] The multimodal fusion building energy consumption model construction method and energy consumption prediction system of this invention generate activity behavior labeled datasets and environmental perception labeled datasets based on historical activity images and environmental historical perception data inside the building, and cross-link them with building structure data to obtain a multimodal knowledge graph. Based on real-time environmental perception information, matching knowledge sets are retrieved from the multimodal knowledge graph to generate thermal balance disturbance variables. A basic building energy consumption model is formed based on real-time environmental perception information and equipment configuration information. The basic building energy consumption model is corrected using the thermal balance disturbance variables to obtain the spatiotemporal trend of energy consumption and to formulate equipment operation constraint strategies. By integrating building structure, internal building activity behavior, and environmental perception parameters to form a multimodal and multi-faceted energy consumption evaluation of the building, it can adapt to the energy consumption prediction needs of different types of buildings in different scenarios, improving the flexibility and accuracy of prediction and control.

[0067] The present invention has the following advantages:

[0068] First, this invention integrates multimodal data such as building structure, internal building activities, and environmental perception information to form a comprehensive element characterization of building energy consumption processes from aspects such as the building itself, human activities, and the natural environment, providing a reliable data foundation for energy consumption prediction.

[0069] Secondly, this invention forms a basic building energy consumption model based on building structure and environmental perception information, ensuring that the energy consumption model matches the energy consumption pattern generated by the building due to temperature control and other needs. When the building type and / or environmental scenario changes, the basic building energy consumption model can be adapted by simply changing the building structure and / or environmental perception information from the existing collected data, thereby improving modeling flexibility and reducing model change costs.

[0070] Third, this invention treats building interior activities as factors affecting building thermal balance. By leveraging the influence of environmental perception information on these activities, it retrieves activities from existing knowledge graphs that are compatible with real-time environmental perception information. This eliminates the need for additional resources to identify real-time activities and reduces costs, ensuring reliable thermal balance interference references even when visual monitoring equipment malfunctions, thus guaranteeing the continuity of energy consumption prediction. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0072] Figure 1This is a flowchart of the method for constructing a multimodal fusion building energy consumption model provided by the present invention.

[0073] Figure 2 It is the process of generating a dataset of labeled activity behaviors.

[0074] Figure 3 It is the environmental condition perception data of the building.

[0075] Figure 4 It is the process of generating a multimodal knowledge graph.

[0076] Figure 5 It is the retrieval process of a knowledge graph.

[0077] Figure 6 It is the process of constructing a building basic energy consumption model.

[0078] Figure 7 It is the process of generating the spatiotemporal trend of energy consumption.

[0079] Figure 8 This is a structural diagram of the multimodal fusion building energy consumption model construction system provided by the present invention. Detailed Implementation

[0080] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and not for limiting the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all structures. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention.

[0081] The terms "comprising" and "having," and any variations thereof, used in this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0082] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0083] Please see Figure 1 As shown, this invention provides a method for constructing a multimodal fusion building energy consumption model, which includes the following steps:

[0084] S100: Perform semantic recognition of historical activity images inside the building to obtain a labeled dataset of activity behaviors; collect and process the building's environmental history perception data to generate an environmental perception labeled dataset.

[0085] Furthermore, in S100, semantic recognition of historical activity images inside the building is performed to obtain a labeled dataset of activity behaviors; environmental historical perception data of the building is collected and processed to generate an environmental perception labeled dataset, specifically:

[0086] Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections;

[0087] Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal equilibrium in each subspace, and the activity behavior data has spatiotemporal labels;

[0088] Collect environmental condition perception data of the building's surrounding environment; the environmental condition perception data includes sunshine data and atmospheric airflow data;

[0089] Spatiotemporal calibration is performed on the environmental condition perception data to generate an environmental perception labeled dataset; the environmental perception labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

[0090] Buildings, especially high-rise buildings, have different internal layouts on each floor (such as the division of rooms, the location and size of doors and windows), as well as different solar conditions (such as solar radiation and angle of sunlight) and atmospheric airflow conditions (such as airflow direction and speed). This results in varying energy consumption for human activities (such as opening doors and / or windows) and maintaining the desired thermal balance on each floor. In practical forecasting, each floor can be treated as a separate space for energy consumption forecasting. This avoids significant prediction errors caused by the fluctuations in energy demand across different floors when forecasting the energy consumption of the entire building.

[0091] Please see Figure 2Cameras can be distributed across each floor of the building to capture historical activity images, which are then integrated into a historical activity image set. These historical activity images may include, but are not limited to, images of human activity on each floor during the building's historical operation. It is understood that these human activities encompass actions such as opening doors and / or windows, which affect indoor thermal balance. These actions are related to the sunlight and / or airflow conditions on each floor. For example, when sunlight intensity is weak or external wind speed is high, people tend to open doors and / or windows for ventilation. This disrupts the original thermal balance (temperature equilibrium) established by air conditioning and other temperature control equipment. The temperature control equipment then needs to consume additional energy to achieve another thermal balance. Therefore, indoor activity is a significant factor affecting building energy consumption. Based on the spatial index of each historical activity image (e.g., the coordinates of the camera that captured each historical activity image), the historical activity image set is divided into several sub-space historical activity image sets. Each sub-space historical activity image set may contain all historical activity images corresponding to the same floor.

[0092] Temporal semantic recognition is performed on all historical activity images within each subspace's historical activity image set. This identifies the activity behavior characteristics of each historical activity image at the temporal level. These activity behavior characteristics may include, but are not limited to, the temporal trajectory changes of actions performed by people within the subspace that affect thermal equilibrium, such as the trajectory changes over time during actions like opening a door or window. Then, temporal and spatial labels are added to these activity behavior characteristics to form a labeled activity behavior dataset. It can be understood that this labeled activity behavior dataset contains the spatial and temporal changes of each action affecting thermal equilibrium within the subspace.

[0093] The natural environment surrounding a building, including sunlight and atmospheric airflow, directly impacts its energy consumption. Specifically, at any given moment, the amount of sunlight and airflow received by each floor of a building differs. For instance, higher floors generally receive more solar radiation than lower floors, and higher floors typically experience higher wind speeds. In summer, greater solar radiation leads to greater heat accumulation and consequently higher energy consumption; conversely, higher wind speeds in summer result in faster heat dissipation and lower energy consumption. Therefore, the natural environmental conditions surrounding a building are a crucial factor influencing its energy consumption. To address this, several light sensors and wind speed sensors are deployed within the building to obtain environmental condition data. This data may include, but is not limited to, the solar radiation intensity and angle of sunlight on each floor, as well as the wind speed and direction on each floor. Please refer to [link to relevant documentation]. Figure 3The wind speed data corresponds to a specific floor of the building between 9 AM and 5 PM, showing that wind speeds vary at different times. Considering the large size of each floor, to further subdivide the sunlight and airflow conditions at different locations on each floor, each floor (e.g., the exterior facade) is divided into several 3D grids. The sunlight and airflow conditions for each floor are then mapped onto these 3D grids, yielding sunlight and airflow data for each grid. These data are then calibrated with time points and 3D grid spatial coordinates to create an environmental perception labeled dataset. This dataset allows users to retrieve sunlight and airflow data for each 3D grid on each floor at any given time of day. By generating activity behavior labeled datasets and environmental perception labeled datasets, a sufficient data foundation is provided for subsequently characterizing building conditions by linking activity behavior with natural environmental parameters.

[0094] S200: Cross-reference the building structure dataset, activity behavior labeled dataset, and environmental perception labeled dataset to obtain a multimodal knowledge graph; retrieve matching knowledge sets from the multimodal knowledge graph based on the real-time environmental perception information of the building.

[0095] Furthermore, in S200, the building structure dataset, activity behavior labeled dataset, and environmental perception labeled dataset are cross-linked to obtain a multimodal knowledge graph; based on the real-time environmental perception information of the building, a matching knowledge set is retrieved from the multimodal knowledge graph, specifically:

[0096] Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset;

[0097] Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain the first knowledge graph; wherein, the first knowledge graph includes the spatiotemporal association knowledge between the activity behaviors affecting the thermal balance within each subspace and the internal range of each subspace;

[0098] Spatiotemporal cross-association is performed on the building structure dataset and the environmental perception labeled dataset to obtain a second knowledge graph; the second knowledge graph includes the spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to.

[0099] By spatiotemporally associating and superimposing the first and second knowledge graphs, a multimodal knowledge graph is obtained.

[0100] The real-time environmental perception information of the building is analyzed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field.

[0101] Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes the activity behavior knowledge associated with the multidimensional environmental condition field in each subspace.

[0102] As discussed above, a building can be divided into several sub-spaces corresponding to each floor, and each sub-space is further divided into several rooms. The activities of people in each room and the influence of natural environmental conditions differ. Each room can be considered a relatively independent temperature control range. When the doors and windows of a room are closed, the room can be considered to be in thermal equilibrium, meaning that heat exchange between the room's interior and exterior is minimal. When the doors and windows are opened, the original thermal equilibrium is disrupted. Given that different rooms may exhibit different activities that affect thermal equilibrium, it is necessary to perform correlation labeling of these activities for each room. Please refer to [link to relevant documentation]. Figure 4 First, the three-dimensional structure of each subspace within the building is obtained. This three-dimensional structure may include, but is not limited to, the three-dimensional structures corresponding to the walls, doors, and windows of all rooms formed within each subspace. Three-dimensional spatial coordinate and size transformations are then performed on these structures to obtain a building construction dataset. Next, according to the spatiotemporal labels of each activity in the activity behavior-labeled dataset, each activity is mapped to the building construction dataset at both the temporal and spatial levels, resulting in a first knowledge graph. This first knowledge graph reflects the correlation between activity behaviors and the three-dimensional structure within the subspace at both the temporal and spatial levels, such as when and where activities affecting thermal balance occur within the subspace. Then, according to the spatiotemporal labels of each natural environmental parameter in the environmental perception-labeled dataset, each natural environmental parameter is mapped to the building construction dataset at both the temporal and spatial levels, resulting in a second knowledge graph. This second knowledge graph reflects the correlation between natural environmental parameters and the three-dimensional structure within the subspace at both the temporal and spatial levels, such as the effects of sunlight and atmospheric airflow at a specific location within the subspace at a specific time. Finally, using the spatiotemporal attributes of the building structure dataset in the first and second knowledge graphs as association attributes, the spatiotemporal associations of the first and second knowledge graphs are superimposed to form a multimodal knowledge graph. It can be understood that the above multimodal knowledge graph represents the association of three factors affecting building energy consumption: the building itself, human activities, and the natural environment.

[0103] As discussed above, the natural environmental conditions, such as sunlight and / or airflow, corresponding to each floor of a building's subspaces affect the activities of people within those subspaces. When sunlight intensity is low or external wind speeds are high, people typically open windows and / or doors for ventilation. This indicates a correlation between real-time environmental perception information and the activities of people within the building. In practice, a general model can be constructed based on a multimodal knowledge graph. This probabilistic model represents the probability that people in each room will open doors or windows under a series of historical environmental perception parameters during a historical period. These historical environmental perception parameters are combinations of different solar radiation and wind speed values ​​generated from historical environmental perception data. Then, confidence calculations, conflict resolution, and feedback verification are performed on the probabilistic relationships between the combinations of historical environmental perception parameters included in the probabilistic model and the activities of opening doors or windows in each room within the building to obtain a reliable probabilistic relationship between each combination of historical environmental perception parameters and the activities of opening doors or windows. The conflict resolution and feedback verification processes mentioned above can be implemented through a conventional hybrid forward and reverse reasoning mechanism, which will not be described in detail here.

[0104] To quickly infer the activities of people inside a building given real-time environmental sensing information such as sunlight and / or airflow, a multimodal knowledge graph is used as the knowledge base, the real-time environmental sensing information as the index, and the aforementioned reliable probability relationships are combined to infer the activities of people inside the building that match the real-time environmental sensing information through knowledge retrieval. It is understood that the more refined the knowledge distillation of the multimodal knowledge graph, the closer the retrieved knowledge will be to the actual activities triggered by the real-time environmental sensing information. Please refer to [link / reference]. Figure 5 The system analyzes and transforms real-time environmental perception information to form a multi-dimensional environmental condition parameter distribution map. This map can be, but is not limited to, the numerical distribution of multiple environmental condition parameters such as solar radiation and wind speed on the building facade. Then, based on the room layout and orientation distribution of each floor's sub-spaces within the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into multi-dimensional environmental condition action fields. These fields can be, but are not limited to, the action fields formed by the solar radiation and wind speed received by the facade of each room within each floor's sub-space. Furthermore, these multi-dimensional environmental condition action fields are converted into semantic vectors, which are then used as indexes to retrieve matching knowledge sets from a multimodal knowledge graph. This yields knowledge of the activity behaviors associated with the multi-dimensional environmental condition action fields in each sub-space, providing a basis for subsequently determining the thermal balance disturbances occurring in each room within the sub-space.

[0105] S300: Based on real-time environmental sensing information and building equipment configuration information, a basic building energy consumption model is formed; based on the knowledge set, thermal balance disturbance variables are generated.

[0106] Furthermore, in S300, a building basic energy consumption model is formed based on real-time environmental sensing information and building equipment configuration information; based on the knowledge set, thermal balance disturbance variables are generated, specifically:

[0107] Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental sensing information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined.

[0108] Based on the building's equipment configuration information, determine the dynamic characteristics of temperature control power consumption in each subspace within the building;

[0109] A large model was constructed to analyze the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption, thus forming a building basic energy consumption model.

[0110] Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects.

[0111] Based on the atmospheric environmental disturbances caused by changes in physical state, thermal balance disturbance variables are generated; among them, thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

[0112] As discussed above, the multi-dimensional environmental condition parameter distribution map corresponding to real-time environmental sensing information represents the numerical distribution of multiple environmental condition parameters such as solar radiation and wind speed on the building facade. It can be understood that the amount of solar radiation and wind speed received by the building facade directly affects the heat exchange between the building's interior and exterior spaces. Greater solar radiation increases the potential for heat transfer from the exterior to the interior, while lower wind speeds reduce heat flow in the exterior and make it easier for heat to accumulate in the interior. Please refer to [link / reference]. Figure 6 Based on the aforementioned multidimensional environmental condition parameter distribution map, the spatiotemporal characteristics of heat exchange between the building's internal and external spaces are determined, specifically the direction and flow rate of heat transfer between the intersecting areas of the internal and external spaces over time. Furthermore, the building's internal spaces are equipped with temperature control devices such as air conditioners. The operating power, quantity, and installation location of these devices directly determine the dynamic characteristics of temperature control power consumption in each room within each floor's subspace, i.e., the change in power consumption over time during temperature control in each room. Based on these spatiotemporal characteristics of heat exchange and dynamic characteristics of temperature control power consumption, a large-scale model such as LSTM is constructed to form the building's basic energy consumption model. It can be understood that this basic building energy consumption model represents the energy consumption changes required to maintain the desired temperature conditions inside the building, assuming no activities affecting thermal balance, such as opening windows or doors.

[0113] In actual building operation, activities that affect indoor thermal balance inevitably occur. To match the building's basic energy consumption model with the actual energy consumption during building operation, the activity knowledge corresponding to the aforementioned knowledge set is mapped to internal building entities. This determines how the activities alter the physical states of internal building entities such as doors and windows, including their opening and closing states, opening duration, and opening degree. Furthermore, based on the atmospheric environmental disturbances caused by these physical state changes (e.g., the airflow rate between the building's interior and exterior spaces during door and window opening), thermal balance disturbance variables are determined. These variables characterize the heat exchange flow rate variable triggered by the disruption of the original thermal balance by the corresponding activities, providing a basis for subsequent revisions to the building's basic energy consumption model.

[0114] S400: Based on the thermal balance disturbance variables, perform spatiotemporal correction on the building foundation energy consumption model to obtain the spatiotemporal trend of building energy consumption, thereby forming equipment operation constraint strategies.

[0115] Furthermore, in S400, based on the thermal balance disturbance variables, a spatiotemporal correction is performed on the building foundation energy consumption model to obtain the spatiotemporal variation trend of the building's energy consumption, thereby forming an equipment operation constraint strategy, specifically:

[0116] By performing a spatiotemporal dynamic transformation on the heat exchange flow variables included in the thermal balance disturbance variables, the potential heat flow field is obtained.

[0117] Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal variation trend of energy consumption during the restoration of the original thermal balance in each subspace of the building.

[0118] Based on the spatiotemporal variation trend of energy consumption, the potential overload operation characteristics of temperature control equipment inside buildings are predicted; based on the spatiotemporal distribution characteristics, operation constraint strategies for temperature control equipment are formed; among them, the operation constraint strategies include overload operation early warning and load stabilization strategies.

[0119] It is understandable that the aforementioned thermal balance disturbance variables correspond to the rate of heat exchange between the building's interior and exterior spaces after the original thermal balance is disrupted. This heat exchange between the interior and exterior spaces creates a heat flow field, which directly determines the energy consumption (e.g., total power consumption) required for the temperature control equipment to adjust the indoor space to the desired temperature (e.g., lowering the actual indoor temperature to a preset value). Please refer to [link to relevant documentation]. Figure 7To adapt the building's basic energy consumption model to the thermal imbalance caused by activities, the heat exchange flow variables included in the thermal balance disturbance variables are subjected to spatiotemporal dynamic transformation to obtain a potential heat flow field. This potential heat flow field characterizes the distribution of heat exchange flow rates between the building's internal and external spaces. Based on this potential heat flow field, the heat change of each room within each subspace of the building is generated. This is used to perform spatiotemporal correction on the building's basic energy consumption model, thereby obtaining the spatiotemporal trend of energy consumption changes in each room within each subspace of the building during the process of restoring the original thermal balance (i.e., the original temperature conditions). This allows for a complete prediction of the energy consumption changes of each room at both the temporal and spatial levels.

[0120] As the energy consumption of temperature control equipment increases, its operating load also increases, leading to overload and other abnormal situations. To prevent temperature control equipment from malfunctioning due to prolonged overload operation, based on the aforementioned spatiotemporal trends in energy consumption, the potential overload operation domain distribution characteristics of each temperature control device within the building are predicted (i.e., the time interval distribution of the temperature control device's subsequent overload operation state). This forms an overload operation early warning and load stabilization strategy for each temperature control device. The overload operation early warning refers to issuing a warning notification when the temperature control device is about to enter an overload operation state. The load stabilization can be, but is not limited to, setting an upper limit for the operating load of the temperature control device when it is detected to be in an overload operation state, and proactively changing the operating mode of the temperature control device when the actual operating load reaches the upper limit, thereby reducing the actual operating load and preventing the actual operating load from exceeding the upper limit.

[0121] Please see Figure 8 As shown, the present invention provides a multimodal fusion building energy consumption prediction system, which includes the following modules:

[0122] The first dataset generation module is used to perform semantic recognition of historical activity images inside buildings to obtain a labeled dataset of activity behaviors.

[0123] The second dataset generation module is used to collect and process the building's environmental history perception data to generate an environmental perception labeled dataset.

[0124] The knowledge graph generation module is used to cross-associate building structure datasets, activity behavior labeled datasets, and environmental perception labeled datasets to obtain a multimodal knowledge graph.

[0125] The retrieval module is used to retrieve matching knowledge sets from a multimodal knowledge graph based on real-time environmental sensing information of the building.

[0126] The model building module is used to form a basic building energy consumption model based on real-time environmental sensing information and building equipment configuration information;

[0127] The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set;

[0128] The energy consumption prediction module is used to perform spatiotemporal corrections on the building's basic energy consumption model based on thermal balance disturbance variables, thereby obtaining the spatiotemporal trend of the building's energy consumption and forming equipment operation constraint strategies.

[0129] Furthermore, the first dataset generation module is used to perform semantic recognition of historical activity images inside the building, resulting in a labeled dataset of activity behaviors, specifically:

[0130] Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections;

[0131] Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal equilibrium in each subspace, and the activity behavior data has spatiotemporal labels;

[0132] The second dataset generation module is used to collect and process the building's historical environmental perception data to generate an environmental perception-labeled dataset, specifically:

[0133] Collect environmental condition perception data of the building's surrounding environment; the environmental condition perception data includes sunshine data and atmospheric airflow data;

[0134] Spatiotemporal calibration is performed on the environmental condition perception data to generate an environmental perception labeled dataset; the environmental perception labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

[0135] Furthermore, the graph generation module is used to cross-associate the building structure dataset, activity behavior labeled dataset, and environmental perception labeled dataset to obtain a multimodal knowledge graph, specifically:

[0136] Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset;

[0137] Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain the first knowledge graph; wherein, the first knowledge graph includes the spatiotemporal association knowledge between the activity behaviors affecting the thermal balance within each subspace and the internal range of each subspace;

[0138] Spatiotemporal cross-association is performed on the building structure dataset and the environmental perception labeled dataset to obtain a second knowledge graph; the second knowledge graph includes the spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to.

[0139] By spatiotemporally associating and superimposing the first and second knowledge graphs, a multimodal knowledge graph is obtained.

[0140] The retrieval module is used to retrieve matching knowledge sets from a multimodal knowledge graph based on real-time environmental sensing information of the building. Specifically:

[0141] The real-time environmental perception information of the building is analyzed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field.

[0142] Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes the activity behavior knowledge associated with the multidimensional environmental condition field in each subspace.

[0143] Furthermore, the model building module is used to form a basic building energy consumption model based on real-time environmental sensing information and building equipment configuration information, specifically:

[0144] Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental sensing information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined.

[0145] Based on the building's equipment configuration information, determine the dynamic characteristics of temperature control power consumption in each subspace within the building;

[0146] A large model was constructed to analyze the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption, thus forming a building basic energy consumption model.

[0147] The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set, specifically:

[0148] Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects.

[0149] Based on the atmospheric environmental disturbances caused by changes in physical state, thermal balance disturbance variables are generated; among them, thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

[0150] Furthermore, the energy consumption prediction module is used to perform spatiotemporal corrections on the building foundation energy consumption model based on thermal balance disturbance variables, thereby obtaining the spatiotemporal variation trend of the building's energy consumption and forming equipment operation constraint strategies, specifically:

[0151] By performing a spatiotemporal dynamic transformation on the heat exchange flow variables included in the thermal balance disturbance variables, the potential heat flow field is obtained.

[0152] Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal variation trend of energy consumption during the restoration of the original thermal balance in each subspace of the building.

[0153] Based on the spatiotemporal variation trend of energy consumption, the potential overload operation characteristics of temperature control equipment inside buildings are predicted; based on the spatiotemporal distribution characteristics, operation constraint strategies for temperature control equipment are formed; among them, the operation constraint strategies include overload operation early warning and load stabilization strategies.

[0154] The operation and effect of the multimodal fusion building energy consumption prediction system of the present invention are consistent with the above-mentioned multimodal fusion building energy consumption model construction method, and the multimodal fusion building energy consumption prediction system will not be described again here.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of a necessary general-purpose hardware platform, or by a combination of hardware and software. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a computer product. The present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Other embodiments may also be used. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a multimodal fusion building energy consumption model, characterized in that, The method includes the following steps: S100: Perform semantic recognition of historical activity images inside the building to obtain a labeled dataset of activity behaviors; collect and process the building's environmental history perception data to generate an environmental perception labeled dataset. S200: Cross-correlate the building structure dataset, the activity behavior labeled dataset, and the environmental perception labeled dataset to obtain a multimodal knowledge graph; based on the real-time environmental perception information of the building, retrieve a matching knowledge set from the multimodal knowledge graph, specifically: Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset; Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain a first knowledge graph; wherein, the first knowledge graph includes spatiotemporal association knowledge between the activity behaviors affecting thermal balance within each subspace and the internal range of each subspace; A second knowledge graph is obtained by performing spatiotemporal cross-association on the building structure dataset and the environmental perception label dataset; wherein, the second knowledge graph includes spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to; The first knowledge graph and the second knowledge graph are spatiotemporally correlated and superimposed to obtain a multimodal knowledge graph; The real-time environmental perception information of the building is parsed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field; Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes activity behavior knowledge associated with the multidimensional environmental condition field in each subspace; S300: Based on the real-time environmental sensing information and the building's equipment configuration information, form a basic building energy consumption model; generate thermal balance disturbance variables based on the knowledge set; S400: Based on the thermal balance disturbance variables, perform spatiotemporal correction on the building foundation energy consumption model to obtain the spatiotemporal trend of the building's energy consumption, thereby forming an equipment operation constraint strategy.

2. The method according to claim 1, characterized in that, In S100, semantic recognition of historical activity images inside the building is performed to obtain a labeled dataset of activity behaviors; environmental historical perception data of the building is collected and processed to generate an environmental perception labeled dataset, specifically: Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections; Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal balance in each subspace, and the activity behavior data has spatiotemporal labels; Collect environmental condition sensing data of the building's surrounding environment; wherein, the environmental condition sensing data includes sunshine data and atmospheric airflow data; Spatiotemporal calibration is performed on the environmental condition sensing data to generate an environmental sensing labeled dataset; wherein, the environmental sensing labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

3. The method according to claim 1, characterized in that, In S300, a building basic energy consumption model is formed based on the real-time environmental sensing information and the building's equipment configuration information; and thermal balance disturbance variables are generated based on the knowledge set, specifically: Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental perception information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined. Based on the equipment configuration information of the building, determine the dynamic characteristics of temperature control power consumption in each subspace inside the building; A large model is constructed based on the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption to form a building basic energy consumption model. Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects. Based on the atmospheric environmental disturbances caused by the changes in the physical state, thermal balance disturbance variables are generated; wherein, the thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

4. The method according to claim 1, characterized in that, In S400, based on the thermal balance disturbance variables, a spatiotemporal correction is performed on the building foundation energy consumption model to obtain the spatiotemporal variation trend of the building's energy consumption, thereby forming an equipment operation constraint strategy, specifically: The heat exchange flow variables included in the thermal balance disturbance variables are subjected to spatiotemporal dynamic transformation to obtain the potential heat flow field; Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal trend of energy consumption during the restoration of the original thermal balance in each subspace inside the building. Based on the spatiotemporal variation trend of energy consumption, the potential overload operation domain distribution characteristics of the temperature control equipment inside the building are predicted; based on the time domain distribution characteristics, an operation constraint strategy for the temperature control equipment is formed; wherein, the operation constraint strategy includes overload operation early warning and load stabilization strategy.

5. A multimodal fusion building energy consumption prediction system, characterized in that, The system includes the following modules: The first dataset generation module is used to perform semantic recognition of historical activity images inside buildings to obtain a labeled dataset of activity behaviors. The second dataset generation module is used to collect and process the building's environmental history perception data to generate an environmental perception labeled dataset. The knowledge graph generation module is used to cross-associate the building structure dataset, the activity behavior labeled dataset, and the environment perception labeled dataset to obtain a multimodal knowledge graph, specifically: Obtain the three-dimensional structure of each subspace inside the building to form a building structure dataset; Spatiotemporal cross-association is performed on the building structure dataset and the activity behavior labeled dataset to obtain a first knowledge graph; wherein, the first knowledge graph includes spatiotemporal association knowledge between the activity behaviors affecting thermal balance within each subspace and the internal range of each subspace; A second knowledge graph is obtained by performing spatiotemporal cross-association on the building structure dataset and the environmental perception label dataset; wherein, the second knowledge graph includes spatiotemporal association knowledge between each subspace and the environmental conditions it is subject to; The first knowledge graph and the second knowledge graph are spatiotemporally correlated and superimposed to obtain a multimodal knowledge graph; The retrieval module is used to retrieve a matching knowledge set from the multimodal knowledge graph based on the real-time environmental perception information of the building, specifically: The real-time environmental perception information of the building is parsed and transformed to form a multi-dimensional environmental condition parameter distribution map; based on the spatial distribution of each subspace inside the building, the multi-dimensional environmental condition parameter distribution map is semantically segmented into a multi-dimensional environmental condition action field; Based on the multidimensional environmental condition field, a matching knowledge set is retrieved from the multimodal knowledge graph; wherein, the knowledge set includes activity behavior knowledge associated with the multidimensional environmental condition field in each subspace; The model building module is used to form a basic building energy consumption model based on the real-time environmental sensing information and the building's equipment configuration information. The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set. The energy consumption prediction module is used to perform spatiotemporal correction on the building foundation energy consumption model based on the thermal balance disturbance variables, so as to obtain the spatiotemporal trend of the building's energy consumption and thereby form an equipment operation constraint strategy.

6. The building energy consumption prediction system according to claim 5, characterized in that, The first dataset generation module is used to perform semantic recognition of historical activity images inside buildings to obtain a labeled dataset of activity behaviors, specifically: Based on the spatial index, the historical activity image collection inside the building is divided into several sub-space historical activity image collections; Temporal correlation semantic recognition is performed on the historical activity image sets of each subspace to obtain an activity behavior labeled dataset; wherein, the activity behavior labeled dataset includes activity behavior data that affects thermal balance in each subspace, and the activity behavior data has spatiotemporal labels; The second dataset generation module is used to collect and process the building's historical environmental perception data to generate an environmental perception labeled dataset, specifically: Collect environmental condition sensing data of the building's surrounding environment; wherein, the environmental condition sensing data includes sunshine data and atmospheric airflow data; Spatiotemporal calibration is performed on the environmental condition sensing data to generate an environmental sensing labeled dataset; wherein, the environmental sensing labeled dataset includes the solar radiation data and atmospheric airflow data of each three-dimensional grid under the building at each time point.

7. The building energy consumption prediction system according to claim 5, characterized in that, The model building module is used to form a basic building energy consumption model based on the real-time environmental sensing information and the building's equipment configuration information, specifically: Based on the multi-dimensional environmental condition parameter distribution map corresponding to the real-time environmental perception information, the spatiotemporal characteristics of heat exchange between the building and the outside world are determined. Based on the equipment configuration information of the building, determine the dynamic characteristics of temperature control power consumption in each subspace inside the building; A large model is constructed based on the spatiotemporal characteristics of heat exchange and the dynamic characteristics of temperature control power consumption to form a building basic energy consumption model. The disturbance determination module is used to generate thermal equilibrium disturbance variables based on the knowledge set, specifically: Perform operation mapping on the building's internal entity objects to the activity behavior knowledge corresponding to the knowledge set, and obtain the physical state changes of the building's internal entity objects. Based on the atmospheric environmental disturbances caused by the changes in the physical state, thermal balance disturbance variables are generated; wherein, the thermal balance disturbance variables include heat exchange flow variables caused by the disruption of the original thermal balance.

8. The building energy consumption prediction system according to claim 5, characterized in that, The energy consumption prediction module is used to perform spatiotemporal correction on the building foundation energy consumption model based on the thermal balance disturbance variables, to obtain the spatiotemporal variation trend of the building's energy consumption, thereby forming an equipment operation constraint strategy, specifically: The heat exchange flow variables included in the thermal balance disturbance variables are subjected to spatiotemporal dynamic transformation to obtain the potential heat flow field; Based on the potential heat flow field, the building foundation energy consumption model is spatiotemporally corrected to obtain the spatiotemporal trend of energy consumption during the restoration of the original thermal balance in each subspace inside the building. Based on the spatiotemporal variation trend of energy consumption, the potential overload operation domain distribution characteristics of the temperature control equipment inside the building are predicted; based on the time domain distribution characteristics, an operation constraint strategy for the temperature control equipment is formed; wherein, the operation constraint strategy includes overload operation early warning and load stabilization strategy.

Images