Central air conditioning system and environment regulation and control method thereof

Abstract

The invention discloses a central air conditioning system and an environment regulation and control method thereof. The method comprises the following steps: monitoring environmental data of a current actual scene in real time through a plurality of sensors, establishing a physical field model through a controller according to the environmental data and building topology data, and establishing a behavior occupancy field model according to user behavior data in the environment; generating a multi-target demand mapping operator based on the physical field model, and obtaining scene condition representation according to the multi-target demand mapping operator and the behavior occupancy field model; and generating the virtual sensing data for compensating the environment data based on the scene condition characterization. And compensating the environment data by using the virtual sensing data, and dynamically regulating and controlling the current actual scene according to the compensated environment data. The method is used for achieving the effects that the accuracy of virtual sensing data and the adaptability in a real scene are improved, and the regulation and control accuracy of the central air conditioning system on the building space environment is enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of central air conditioning, and in particular to a central air conditioning system and an environmental control method thereof. Background Art

[0002] Central air conditioning systems use sensors to collect environmental parameter data within the building space environment, enabling them to finely control the building space environment based on this high-density, high-precision environmental parameter data. If environmental parameter data is missing in some areas (e.g., due to sensor loss or failure), the central air conditioning system can compensate for the missing environmental parameter data using virtual sensor data to ensure data integrity, allowing the central air conditioning system to continue monitoring and controlling the building space environment.

[0003] Regarding the acquisition process of virtual sensor data, existing techniques involve predicting the sensor's outlier value when it is determined that the sensor data collected contains an outlier. This generates sensor prediction data, which is then augmented using a diffusion model to produce augmented virtual sensor data. However, in complex, dynamic scenarios with multiple performance indicators, the virtual sensor data generated by existing techniques is inaccurate and difficult to adapt to real-world scenarios, thus reducing the accuracy of the central air conditioning system's control of the building's spatial environment.

[0004] How to improve the accuracy of virtual sensor data and its adaptability in real scenarios, and enhance the precision of central air-conditioning systems in controlling building space environments has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The present application provides a central air-conditioning system and an environmental control method thereof.

[0006] In a first aspect, an embodiment of the present application provides a central air conditioning system, comprising:

[0007] Multiple sensors are used to monitor the environmental data of the current actual scene in real time;

[0008] and a controller configured to obtain virtual sensor data for compensating the environmental data based on the environmental data and the building topology data; compensate the environmental data using the virtual sensor data, and dynamically control the current actual scene based on the compensated environmental data;

[0009] Among them, virtual sensing data is obtained based on environmental data and building topology data, including:

[0010] A physical field model is established based on environmental data and building topology data, and a behavioral occupancy field model is established based on user behavior data in the environment; a multi-objective demand mapping operator is generated based on the physical field model, and a scene condition representation is obtained based on the multi-objective demand mapping operator and the behavioral occupancy field model; virtual sensor data for compensating environmental data is generated based on the scene condition representation.

[0011] In this application, a physical field model and a behavioral occupancy field model are established through a controller, and the multi-target requirements are integrated with the behavioral occupancy field model and the physical field model to obtain a scene condition representation. Based on the scene condition representation, virtual sensor data is generated. By integrating the scene model with the multi-target requirements, the generated virtual sensor data takes into account the multi-target requirements, providing a reliable and detailed environmental data supplement for the central air-conditioning management system, improving the accuracy of the virtual sensor data and its adaptability in real scenarios, improving the efficiency of environmental control, and enhancing the precise control of the central air-conditioning system on the building space environment.

[0012] In some embodiments of the present application, when the controller generates virtual sensor data for compensating environmental data based on the scene condition representation, the controller is specifically configured to:

[0013] Based on the scene condition representation, physical constraints for correcting diffusion trajectories are added to the inverse sampling of the conditional diffusion model. A noise weighting strategy is used to dynamically correct the noise corresponding to multi-objective requirements to generate an adaptively updated conditional diffusion model.

[0014] Data generation and processing are performed based on the adaptively updated conditional diffusion model to obtain preliminary virtual sensing data;

[0015] Virtual sensing data for compensating the environmental data is generated according to the preliminary virtual sensing data.

[0016] In this application, based on the existing conditional diffusion model, physical constraints for correcting the diffusion trajectory are added in the reverse sampling of the conditional diffusion model, and the noise corresponding to the multi-target requirements is dynamically corrected by using a noise weighting strategy. The reverse sampling process of the conditional diffusion model can be automatically adapted as the scene evolves, and the problem of the lack of physical consistency and multi-target balancing ability in the existing technology is solved, so that the generated virtual sensor data covers multiple requirements in the current actual scene, thereby improving the accuracy and reliability of the virtual sensor data.

[0017] In some embodiments of the present application, when the controller generates virtual sensor data for compensating environmental data based on preliminary virtual sensor data, the controller is specifically configured to:

[0018] Establish multiple scene conditions and determine the scene conditions corresponding to the current actual scene;

[0019] Based on the scene conditions, the preliminary virtual sensor data is subjected to multi-perspective feature extraction and fusion processing to obtain the multi-perspective collaborative features of the preliminary virtual sensor data under the scene conditions. The multi-perspectives include spatial perspective, temporal perspective, demand perspective and physical prior perspective.

[0020] The multi-view collaborative features are optimized using the cross-view coupling kernel function and cross-view coupling divergence to generate optimized virtual sensing data under scene conditions.

[0021] In this application, by performing feature separation on the preliminary virtual sensor data generated by the conditional diffusion model under multiple perspectives, the technical problem of feature confusion under multiple perspectives in the existing technology is solved, the accuracy of the optimized virtual sensor data is enhanced, and the feature weights under multiple perspectives are redistributed according to the scene conditions corresponding to the current actual scene, thereby achieving adaptive updating of the scene during the multi-perspective comparative learning process, improving the adaptability of the multi-perspective model, and enhancing the accuracy of the optimized virtual sensor data and its matching degree with the current actual scene.

[0022] In some embodiments of the present application, the controller is further configured to:

[0023] Perform scene matching detection based on the currently obtained virtual sensor data to obtain the matching difference value between the virtual sensor data and the current actual scene;

[0024] When the matching difference value exceeds a preset threshold, the virtual sensor data is corrected through the correction network and the corrected virtual sensor data is output.

[0025] In this application, the accuracy of the virtual sensor data is further improved by determining the matching difference value between the current virtual sensor data and the current actual scene, and triggering the correction network to correct the virtual sensor data when the matching difference value exceeds a preset threshold.

[0026] In some embodiments of the present application, after correcting the virtual sensor data, the controller is further configured to:

[0027] The corrected virtual sensor data is subjected to multi-target fusion processing to generate virtual sensor data that is used to compensate for environmental data and meet multi-target requirements.

[0028] In this application, the corrected virtual sensor data is further adjusted so that the adjusted virtual sensor data can achieve a demand balance between multiple target requirements, improve the adaptability of the virtual sensor data to the current actual scene, and enhance the accuracy of environmental control.

[0029] In some embodiments of the present application, the controller is further configured to:

[0030] Real-time monitoring of whether the current actual scene is switched;

[0031] When the current actual scene is switched, the scene conditions corresponding to the switched scene are determined, and based on the switched scene conditions, optimized virtual sensing data under the switched scene conditions are generated.

[0032] In this application, by monitoring whether the current actual scene switches, the rapid generation of virtual sensor data in different scenes is achieved, the generation efficiency of virtual sensor data is improved, and the efficiency of environmental control is improved.

[0033] In some embodiments of the present application, when the controller establishes a physical field model based on environmental data and building topology data, it is specifically configured to:

[0034] Based on the building topology data in the current actual scenario, determine the dynamic distribution information of the temperature sensor, humidity sensor, and carbon dioxide concentration sensor in the building topology data. The building topology data is determined based on the connectivity relationship between each room in the current actual scenario;

[0035] Generate a sensor distribution tensor based on the dynamic distribution information; determine sensor confidence mapping data based on the sensor distribution tensor. For the same area, the sensor confidence mapping data is positively correlated with the observation density in the area;

[0036] Generate a physics model based on environmental data and sensor confidence map data.

[0037] In this application, building topology data is generated according to the connectivity relationship between rooms / floors, and the dynamic distribution information of different types of sensors is determined based on the building topology data to generate a sensor distribution tensor. The sensor confidence mapping data under the building topology data is determined according to the sensor distribution tensor. After aligning the environmental data monitored by different types of sensors with the confidence mapping data, a physical field model related to the dynamic distribution of sensors is obtained, which improves the fit of the physical field model with the current actual scene and helps to improve the accuracy of virtual sensor data.

[0038] In some embodiments of the present application, when the controller generates a multi-objective demand mapping operator based on a physical field model, it is specifically configured to:

[0039] Determine sensor attention weight data based on the physical field model, where the sensor attention weight data of any region is inversely correlated with the corresponding sensor confidence mapping data;

[0040] The multi-target demand mapping operator corresponding to the multi-target demand is determined according to the sensor attention weight data.

[0041] In this application, the multi-target demand mapping operator corresponding to the multi-target demand is determined by the sensor attention weight data in the current actual scenario. This not only avoids generating virtual sensor data only when the environmental data is missing or abnormal, but also can adaptively generate matching virtual sensor data according to actual needs to supplement the sensor data monitoring, which helps to achieve precise regulation of environmental data between buildings.

[0042] In some embodiments of the present application, when the controller obtains the scene condition representation based on the multi-objective demand mapping operator and the behavior occupancy field model, it is specifically configured to:

[0043] By fusing and aligning the behavioral occupancy field model with the physical field model, a fused field model is obtained;

[0044] Determine the scene fusion vector corresponding to the multi-target requirements based on the fusion field model and the multi-target requirement mapping operator;

[0045] A scene condition representation is determined based on the scene fusion vector.

[0046] In this application, by fusing the behavioral occupancy field model and the multi-objective demand mapping under the physical field model, a scene fusion vector is obtained. By determining the mutual information component between the behavioral occupancy field model and the scene fusion vector, the impact of human activities on the environmental state and multi-objective demands is further highlighted, thereby improving the authenticity and accuracy of the scene condition representation.

[0047] In a second aspect, an embodiment of the present application provides an environmental control method for a central air conditioning system, wherein the central air conditioning system includes multiple sensors and a controller, and the method includes:

[0048] Monitor the environmental data of the current actual scene in real time through multiple sensors;

[0049] The controller obtains virtual sensor data for compensating environmental data based on environmental data and building topology data; the virtual sensor data is used to compensate for the environmental data, and the current actual scene is dynamically controlled based on the compensated environmental data;

[0050] Among them, virtual sensing data is obtained based on environmental data and building topology data, including:

[0051] A physical field model is established based on environmental data and building topology data, and a behavioral occupancy field model is established based on user behavior data in the environment; a multi-objective demand mapping operator is generated based on the physical field model, and a scene condition representation is obtained based on the multi-objective demand mapping operator and the behavioral occupancy field model; virtual sensor data for compensating environmental data is generated based on the scene condition representation.

[0052] In this application, by establishing a physical field model and a behavioral occupancy field model, multi-target requirements are integrated with the behavioral occupancy field model and the physical field model to obtain a scene condition representation, and virtual sensor data is generated based on the scene condition representation. By integrating the scene model with multi-target requirements, the generated virtual sensor data takes into account multi-target requirements, providing a reliable and detailed environmental data supplement for the central air-conditioning management system, improving the accuracy of the virtual sensor data and its adaptability in real scenarios, improving the efficiency of environmental control, and enhancing the precise control of the central air-conditioning system on the building space environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Shows a structural schematic diagram of the central air-conditioning system provided by this application;

[0054] Figure 2 It shows the structural diagram of the floor AHU provided by this application;

[0055] Figure 3 A schematic diagram of a central air-conditioning system provided by this application is shown;

[0056] Figure 4 A schematic diagram showing a flow chart of an environmental control method for a central air-conditioning system provided in the present application is shown;

[0057] Figure 5 A schematic diagram of the process of establishing a physical field model provided by the present application is shown;

[0058] Figure 6 A schematic diagram of the process of generating a multi-objective demand mapping operator provided by the present application is shown;

[0059] Figure 7 A schematic diagram of the process of generating scene condition representation provided by this application is shown;

[0060] Figure 8 The process of generating virtual sensor data provided by this application is shown as follows Figure 1 ;

[0061] Figure 9 The process of generating virtual sensor data provided by this application is shown as follows Figure 2 ;

[0062] Figure 10 A structural schematic diagram of the environment control device of the central air-conditioning system provided in this application is shown. DETAILED DESCRIPTION

[0063] In order to make the purpose and implementation of this application clearer, the exemplary implementation of this application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of this application. Obviously, the described exemplary embodiments are only part of the embodiments of this application, not all of the embodiments.

[0064] It should be noted that the brief descriptions of terms in this application are only for the purpose of facilitating the understanding of the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their ordinary and usual meanings.

[0065] In the specification and claims of this application and the accompanying drawings, the terms "first," "second," "third," etc. are used to distinguish similar or similar objects or entities, and are not necessarily intended to limit a particular order or sequence, unless otherwise noted. It should be understood that the terms used in this manner are interchangeable under appropriate circumstances.

[0066] The terms "comprise," "include," and "have," and any variations thereof, are intended to cover but not exclude inclusion; for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

[0067] Central air-conditioning systems are currently the core facilities for building environmental control. They are widely used in commercial buildings, data centers, hospitals, hotels, transportation hubs and other scenarios to provide efficient and stable temperature, humidity and air quality for large-scale space environments.

[0068] Figure 1 The structural diagram of the central air-conditioning system provided for this application is as follows: Figure 1 As shown, the specific central air-conditioning system of this application includes a driving mechanism, a cloud platform, a controller and multiple sensors. The driving mechanism includes an outdoor unit, a floor AHU (Air Handling Unit), and a VAV box (Variable Air Volume Box) in each room.

[0069] The outdoor unit is the core equipment of the refrigeration cycle, which includes a compressor, condenser and throttling device, and is responsible for the compression, condensation and heat dissipation of the refrigerant. The outdoor unit realizes the transportation of refrigerant (such as chilled water) between the outdoor unit and the floor AHU through the pipe shaft. Among them, the pipe shaft is a vertical channel in the building, which can be a liquid pipe or a gas pipe, used for laying refrigerant pipes, chilled water pipes and cables. For example, refrigerant is transported through the laid refrigerant pipes (such as seamless copper pipes) and cables (such as flame-retardant BV wires). The pipe shaft is connected to the outdoor unit and the floor AHU by welding or flanges.

[0070] The floor AHU (Air Handling Unit) is an air handling center that is used to complete the steps of fresh air introduction, filtration, cooling / heating, and humidity control. The input end of the floor AHU is connected to the pipe shaft, and the output end is connected to the air duct. The air duct includes the main air duct, branch air duct, VAV box interface, and muffler or fire damper. The air duct is used to distribute the conditioned air to each terminal (i.e., the VAV box in each room). In one embodiment of the present application, Figure 2 As shown, the floor AHU includes a fresh air valve, a filter, a surface cooler, a heater, a humidifier, and a fan. The floor AHU receives chilled water transmitted from the pipe shaft, passes the chilled water through the fresh air valve, the filter, the surface cooler, the heater, the humidifier, and the fan to generate air with a target temperature and target humidity, and then transports the treated air to the VAV box (Variable Air Volume Box) in each room through air ducts (e.g., galvanized iron air ducts).

[0071] Each room's VAV box (Variable Air Volume Box) serves as a terminal device, precisely adjusting the air supply volume based on room needs. The process works as follows: Each room's VAV box receives a thermostat signal, adjusts the damper opening (0-90°) based on the thermostat signal, and sets 30% air volume as the minimum air volume. The input of each room's VAV box is connected to the air duct, and the output faces the corresponding room. For example, each room's VAV box delivers air to the corresponding room through a linear diffuser.

[0072] Through the organic coordination of the above modules, the central air-conditioning system realizes full-link intelligent control from the cold source to the terminal.

[0073] Multiple sensors are deployed in the rooms, with the number and type of sensors varying between rooms. For example, sensor types may include, but are not limited to, temperature sensors (with a temperature measurement range of 0-50°C) and humidity sensors (with a humidity measurement range of 0-100%). For rooms in the office area, one sensor is used for every 50 square meters of environmental data, while for rooms in the computer room area, one sensor is used for every 20 square meters of environmental data.

[0074] The controller is the core of the central air conditioning system's intelligent decision-making. Its hardware configuration may include, but is not limited to, a processor (e.g., an ARM Cortex-M7) and memory (e.g., 1GB of DDR3). Its algorithms and strategies may include, but are not limited to, PID control (e.g., accuracy of ±0.5°C) and load forecasting (e.g., an LSTM neural network).

[0075] The cloud platform is used to analyze the energy efficiency of the central air-conditioning system and perform fault diagnosis on the operating status of the central air-conditioning system.

[0076] In one embodiment of the present application, Figure 3 As shown in Figure 1, the sensor transmits real-time monitoring data (e.g., temperature and humidity data) to the controller. When the controller determines that the sensor fails (e.g., data is missing or abnormal) based on the real-time monitoring data, it automatically switches to the process of generating virtual sensor data, determines the control strategy (e.g., water pump speed instruction or air valve opening) based on the virtual sensor data, and sends the control strategy to the actuator (i.e., Figure 1 As shown), at the same time, the controller uploads the working status of the equipment / modules and sensors in the actuator, as well as the aggregated data generated based on the virtual sensor data to the cloud platform, performs event recording through the cloud platform, and outputs subsequent response measures.

[0077] For example, in the event of a temporary overcrowding in a conference room, if the CO2 concentration measured by the CO2 sensor exceeds 1000 ppm (the CO2 concentration threshold), the controller generates a control strategy and controls the actuator to increase the VAV box air volume in the current room to its maximum overclocked value, triggering the fresh air unit to increase air volume by 30%, and uploading the current event to the cloud platform. Upon receiving the event, the cloud platform records it in the energy log and pushes an alert and evacuation recommendations to the property management app.

[0078] based on Figure 3 In the scenario shown, when the central air-conditioning system provides efficient and stable temperature, humidity and air quality to the spatial environment, the existing technology usually collects existing measured data such as temperature and humidity, personnel flow, and energy consumption, and then uses linear interpolation, time series extrapolation, etc. to supplement the environmental data points in the unmonitored time and space to obtain a set of aggregated data that can cover the entire space or a complete time period, and then performs data analysis based on the aggregated data to regulate the current environment.

[0079] However, when using methods based on data inference or simple interpolation to supplement environmental data points, it is necessary to rely on sufficiently dense and stable original observation data. If the sensor data in certain spatial areas are severely missing, or the indoor and outdoor working conditions fluctuate significantly over time, the aggregated data supplemented by existing linear interpolation or extrapolation methods will be difficult to accurately reflect the actual environmental conditions. Relying on historical data to fill the gaps in spatial and temporal dimensions will also make it difficult to adapt to the dynamic changes in building usage patterns, thereby reducing the accuracy of environmental control.

[0080] Therefore, at present, the existing technology proposes to aggregate the sensor data, energy load and occupancy information in the building space into a large model, and use neural networks to automatically learn the distribution of environmental indicators under different conditions, so as to generate corresponding predictions or simulation values ​​when there is a lack of monitoring at new times or in new areas.

[0081] However, in complex dynamic scenarios with multiple indicator requirements, the generation efficiency of existing technologies is low, and the accuracy of the virtual sensor data obtained is low, making it difficult to adapt to multiple requirements in real scenarios, thereby reducing the accuracy of the central air-conditioning system's control of the building space environment.

[0082] The environmental control method for a central air-conditioning system provided in the present application determines building topology data through a controller according to the connectivity relationship of each room in the current actual scenario, generates a physical field model according to the building topology data and environmental data of the environment monitored in real time by multiple sensors, generates a behavior occupancy field model according to user behavior data in the current actual scenario, determines sensor attention weight data based on the physical field model, determines a multi-target demand mapping operator corresponding to the multi-target demand based on the sensor attention weight data, fuses and aligns the behavior occupancy field model with the physical field model to obtain a fusion field model, determines a scene fusion vector corresponding to the multi-target demand based on the fusion field model and the multi-target demand mapping operator, and determines a scene condition representation according to the target scene characteristics. According to the scene condition representation, physical constraints and noise weighting strategies are added to the conditional diffusion model to generate a conditional diffusion model with noise correction and trajectory constraints, so as to generate preliminary virtual sensor data through the updated conditional diffusion model. According to the scene conditions corresponding to the current actual scene, the preliminary virtual sensor data is subjected to multi-perspective feature extraction and fusion processing, as well as multi-perspective feature optimization processing, to finally obtain the optimized virtual sensor data under the scene conditions. Thereafter, the current virtual sensor data is corrected to obtain the corrected virtual sensor data. Finally, the corrected virtual sensor data is subjected to multi-target fusion processing to generate virtual sensor data that meets multi-target requirements for compensating environmental data, providing reliable and precise environmental data supplement for the central air-conditioning management system and cloud platform, ensuring data integrity when sensor coverage is incomplete or sensor deployment is sparse, improving the accuracy of virtual sensor data and its adaptability in real scenarios, improving environmental control efficiency, and enhancing the precise control of the central air-conditioning system on the building space environment.

[0083] The following uses the controller of the central air-conditioning system as an example to illustrate how the central air-conditioning system dynamically controls the environment.

[0084] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0085] Figure 4 The flow chart of the environmental control method of the central air-conditioning system provided in this application is as follows: Figure 4 As shown, the method includes:

[0086] S401: Acquire environmental data of the current actual scene monitored in real time by multiple sensors.

[0087] In some embodiments, the types of the multiple sensors include temperature sensors, humidity sensors, and carbon dioxide sensors.

[0088] S402: Obtain virtual sensing data for compensating the environmental data according to the environmental data and the building topology data.

[0089] More specifically, virtual sensing data is obtained based on environmental data and building topology data, including: establishing a physical field model based on the environmental data and building topology data, and establishing a behavioral occupancy field model based on user behavior data in the current actual scenario; generating a multi-objective demand mapping operator based on the physical field model and the behavioral occupancy field model, and obtaining a scene condition representation based on the multi-objective demand mapping operator; and generating virtual sensing data for compensating for environmental data based on the scene condition representation.

[0090] like Figure 5 As shown in the figure, a physical field model is established based on environmental data and building topology data, including:

[0091] S501: Based on the building topology data in the current actual scenario, determine the dynamic distribution information of the temperature sensor, the humidity sensor, and the carbon dioxide concentration sensor in the building topology data.

[0092] More specifically, the building topology data is determined based on the connectivity relationship between rooms in the current actual scenario. Exemplarily, the connectivity relationship includes the vertical connectivity relationship between floors and the horizontal adjacency relationship between rooms on the same floor.

[0093] For example, the current actual scene includes multiple buildings, each building includes multiple rooms, and according to the floor height, location coordinates and room segmentation boundary information of each building in the current actual scene, the building space coordinate set {(x m ,y m ,z m )} and the room index set {r m Obtain the vertical connectivity relationship between floors and the horizontal adjacency relationship between rooms on the same floor, and generate building topology data (i.e. matrix A) based on the obtained connectivity relationship. mn Indicates room r m With room r n The spatial connectivity between the rooms is m With room r n When there is no connectivity between them, determine the corresponding element A mn =0; when room rm With room r n When there is a common boundary or open space between them, determine the corresponding element A mn = 1. Determine the actual deployment of sensors under the above building topology data, where the types of sensors include but are not limited to temperature sensors (identified by the abbreviation T), humidity sensors (identified by the abbreviation H), and carbon dioxide sensors (identified by the abbreviation C) and other sensors that measure physical indicators.

[0094] S502 : Generate a sensor distribution tensor according to the dynamic distribution information; and determine sensor confidence mapping data according to the sensor distribution tensor.

[0095] More specifically, for the same region, the sensor confidence map data is positively correlated with the observation density under the region.

[0096] For example, a sensor distribution tensor is generated based on dynamic distribution information. Where M represents the number of rooms in the target building, N represents the number of sensor types, and K represents the discrete number of time dimensions. The elements in the sensor distribution tensor D are D m,n,k If the room r m There is a type n sensor (e.g., temperature sensor) collecting data in the kth time slice, then D m,n,k is the current valid observation value of sensor type n; otherwise, D m,n,k The corresponding value is 0. Within each room indicated by the sensor distribution tensor, if the sensor distribution per unit area in that room is below a first threshold, or if the ratio of effective observation time to total observation time in each room, as determined based on the values ​​of the elements in the sensor distribution tensor, is below a preset threshold, the room is considered a sparse area. Otherwise, it is considered a dense area, and the density level of the area is marked in the corresponding room area. A matching confidence level is assigned to the room based on the density level. Higher sensor coverage in the room indicates a higher confidence level, while lower sensor coverage indicates a lower confidence level.

[0097] S503: Generate a physical field model based on the environmental data and the sensor confidence mapping data.

[0098] For example, for temperature sensors, humidity sensors, and carbon dioxide sensors, the multi-parameter evolution function in time and space is U(x,y,z,t)=[T(x,y,z,t),H(x,y,z,t),C(x,y,z,t)] T , a coupled evolution equation is established based on the real-time monitored temperature data, humidity data and carbon dioxide concentration data, namely, Among them, α is the diffusion coefficient, which is used to characterize the natural diffusion effect of heat, humidity and carbon dioxide gas between rooms and different areas in the same room. Q is the exogenous term vector, which is used to represent the amount of cold, heat, water vapor and fresh air brought in by the interaction between the room and the outdoor environment. β is the field suppression coefficient, which is used to control the intensity of central air conditioning regulation on the environmental field. X = [x, y, z] is the three-dimensional coordinate position of the room. F(U) is a function of forced convection or mechanical ventilation. The above function is used to describe the active intervention amount of the air supply and exhaust system on the physical field distribution. This application can characterize the dynamic evolution law of three types of key environmental parameters (temperature, humidity and carbon dioxide concentration) at the physical level based on the above-mentioned coupled evolution equations. The dynamic evolution law of the three types of key environmental parameters (temperature, humidity and carbon dioxide concentration) is integrated with the sensor confidence mapping data under the building topology data, that is, the temperature, humidity and carbon dioxide concentration fields solved in U(x, y, z, t) are aligned with the real sensor data to obtain a physical field model. W is the fusion operation symbol for point-by-point multiplication by position, sen (x, y, z) is the sensor confidence mapping data W sen The value corresponding to the coordinate point X in the building topology data A, A(x, y, z) represents the room at the coordinate point X in the building topology data A, and t is the discrete index of the physical field model in the time dimension.

[0099] In this application, building topology data is generated according to the connectivity relationship between rooms / floors, and based on the building topology data, the dynamic distribution information of different types of sensors is determined to generate a sensor distribution tensor, and the sensor confidence mapping data under the building topology data is determined according to the sensor distribution tensor. After aligning the environmental data monitored by different types of sensors with the confidence mapping data, a physical field model related to the dynamic distribution of sensors is obtained, which improves the fit of the physical field model with the current actual scene and helps to improve the accuracy of virtual sensor data.

[0100] like Figure 6 As shown in Figure 2, a multi-objective demand mapping operator is generated based on the physical field model, specifically including:

[0101] S601: Determine sensor attention weight data based on a physical field model, where the sensor attention weight data of any region is inversely correlated with corresponding sensor confidence mapping data.

[0102] More specifically, based on the physical field model Θ(X, t), the sensor confidence mapping data in the current actual scenario is determined. After obtaining the sensor confidence mapping data, the attention weight data is determined based on the numerical value in the sensor confidence mapping data. The larger the numerical value in the sensor confidence mapping data (i.e., the sensors in the room are densely deployed), the smaller the corresponding attention weight data; the smaller the numerical value in the sensor confidence mapping data (i.e., the sensors in the room are sparsely deployed), the larger the corresponding attention weight data.

[0103] S602: Determine a multi-target requirement mapping operator corresponding to the multi-target requirements according to the sensor attention weight data.

[0104] More specifically, the multi-target requirements are determined according to the sensor attention weights, and the target requirement operator Ω is constructed by measuring the degree of coupling between each target requirement and the physical field model in multidimensional space (e.g., three-dimensional space) and time domain. k =∫∫∫ Ω×[0,T] Γ k (Θ(X,t))dXdt, where k is the kth target demand, Γ k is the mapping kernel function of the kth objective requirement, which is used to perform weighted aggregation on the characteristic components corresponding to the multi-objective requirements under the physical field model.

[0105] Optionally, by changing the mapping kernel function Γ k With the time scale [0, T], it is possible to focus the target demand on environmental indicators at different angles such as local temperature gradient and air flow rate.

[0106] Optionally, if there are k target requirements in the current actual scenario, the integrated multi-target requirements in the current actual scenario are mapped to Ω total (When there are k target requirements in total, the corresponding multi-target requirement mapping can be Ω(k)), and Among them, λ k Represents the weight coefficient of the kth target demand.

[0107] In this application, the multi-target demand mapping operator corresponding to the multi-target demand is determined by the sensor attention weight data in the current actual scenario. This not only avoids generating virtual sensor data only when the environmental data is missing or abnormal, but also can adaptively generate matching virtual sensor data according to actual needs to supplement the sensor data monitoring, which helps to achieve precise regulation of environmental data between buildings.

[0108] like Figure 7 As shown in the figure, the scene condition representation is obtained based on the multi-objective demand mapping operator and the behavior occupancy field model, which specifically includes the following steps:

[0109] S701: Obtain a fused field model by fusing and aligning the behavior occupancy field model with the physical field model.

[0110] In some embodiments, to compensate for the limitations of a simple physical field model in describing actual scenarios, a behavioral occupancy field model is established based on user behavior data within the environment. User behavior data includes, but is not limited to, data on users entering and exiting rooms and room lighting data. The behavioral occupancy field model is used to quantify the additional impact of human activity density and device usage intensity on the local spatial environment. For example, the equation for the behavioral occupancy field model O(x, y, z, t) is: Among them, D o is the occupied diffusion coefficient, S ext Including security system entrance and exit logs and lighting control system dynamic introduction of source data, according to the security system entrance and exit logs and lighting control system dynamic introduction of source data to determine the additional occupancy intensity caused by users entering and leaving the room or turning on high-power equipment during a certain period of time, μ is the management control coefficient, G(O(x,y,z,t)) represents the intensity of the central air-conditioning system to control the environment of over-occupied areas. In order to describe the interactive effects of personnel flow, energy consumption and heat and moisture transfer, after obtaining the behavioral occupancy field model, based on The behavioral occupancy model is coupled with the physical field model, where B ′ (x, y, z, t) is the occupancy intensity corresponding to the security system, lighting system and building traffic (i.e., the occupancy intensity in time and space), and λ1 is the weight coefficient of the smoothness of the equilibrium behavior occupancy field and the consistency of the data.

[0111] After the behavioral occupancy field model completes data assimilation with the temperature field T, humidity field H, and carbon dioxide concentration field C in the physical field model, the fusion field model U is obtained. * =[T,H,C,O] T .

[0112] S702: Determine a scene fusion vector corresponding to the multi-target requirement based on the fusion field model and the multi-target requirement mapping operator.

[0113] In some embodiments, based on Φ(U * (x,y,z,t),Ω(k))=W k ⊙U * (x,y,z,t) vs. U * Reweighting is performed to differentiate the dynamic impact of occupancy behavior and physical quantities such as temperature, humidity, and carbon dioxide concentration under multi-objective requirements to form a scene fusion vector F corresponding to the multi-objective requirements. k (x,y,z,t). Among them, Ω(k) is the multi-objective demand mapping, Φ is the fusion operator, W kis the weighted vector set on the physical quantity and occupancy according to the target demand index k.

[0114] Optionally, for the target demand k that is more sensitive to temperature, the T in W can be increased. k The weight in .

[0115] S703: Determine a scene condition representation based on the scene fusion vector.

[0116] In some embodiments, in order to improve the matching degree between the environmental control of the central air-conditioning system and the target demand, a mutual information gain operator is established based on the multi-field mutual information gain criterion. Where p(·) represents the probability distribution. Mutual information gain operator I(F k ; O) used for scene fusion vector F k The coupling effect of the occupancy field and other physical quantities in (x, y, z, t) is quantitatively calculated. If the mutual information component between the behavioral occupancy field model and the scene fusion vector is high, it indicates that the personnel activity has a strong influence on the target demand k. In order to further strengthen the mutual information effect, the scene fusion vector and the mutual information gain operator are nonlinearly weighted to obtain the scene condition representation Z under the dynamic coupling of the behavioral occupancy field model. k (x,y,z,t).

[0117] For example, Z k (x,y,z,t)={F k (x,y,z,t)} α ·exp[β·I(F k ; O)], where α and β are adjustable hyperparameters used to enhance the influence of high mutual information components and suppress the influence of low mutual information components under multi-objective requirements.

[0118] In this application, by fusing the behavioral occupancy field model and the multi-objective demand mapping under the physical field model, a scene fusion vector is obtained. By determining the mutual information component between the behavioral occupancy field model and the scene fusion vector, the impact of human activities on the environmental state and multi-objective demands is further highlighted, thereby improving the authenticity and accuracy of the scene condition representation.

[0119] Currently, the forward diffusion probability distribution of the conditional diffusion model is Among them, α t and σ t are the scaling factor and noise intensity that change with time steps, respectively, and are used to gradually inject the original features into the random noise space in the forward diffusion stage. trepresents the random variable of the diffusion process at time step t. Existing technologies typically incorporate a small amount of prior conditions based on the aforementioned conditional diffusion model to generate virtual sensor data. However, existing technologies struggle to fully understand the requirements of complex scenarios, which can lead to poor adaptability to multi-target requirements in the reverse sampling phase during the later stages of diffusion, resulting in reduced accuracy of the generated virtual sensor data.

[0120] like Figure 8 As shown, virtual sensing data for compensating environmental data is generated based on scene condition representation, specifically including:

[0121] S801. Based on the scene condition representation, physical constraints for correcting the diffusion trajectory are added in the reverse sampling of the conditional diffusion model, and the noise corresponding to the multi-objective requirements is dynamically corrected using a noise weighting strategy to generate an adaptively updated conditional diffusion model.

[0122] In some embodiments, based on the characteristics corresponding to the physical field model in the current actual scene in the scene condition representation and the characteristics corresponding to the multi-objective demand mapping, the stochastic differential equation corresponding to the inverse sampling of the adaptive conditional diffusion model is constructed as follows: in, E(·) is the energy function in the traditional conditional diffusion model, λ pde is the PDE constraint adjustment coefficient, The step quantity is used to indicate the spatiotemporal distribution of temperature, humidity, and carbon dioxide concentration during the reverse sampling process. The step quantity involves a multi-target demand mapping operator, which can physically correct the traditional boundless diffusion trajectory, improving the matching of virtual sensor data with the multi-target requirements of the current real-world scenario.

[0123] In some embodiments, if there are k target requirements in the current actual scene, based on the features corresponding to the multi-target requirement mapping in the current actual scene in the scene condition representation, a noise weighting matrix W corresponding to the multi-target requirement mapping Ω(k) is established in the reverse sampling process of the conditional diffusion model. k (t), is used to distinguish the impact of different target requirements on the noise injection intensity. The noise weighting matrix formula is: ν k is the sensitivity coefficient corresponding to the target demand k, is the estimated demand satisfaction under the current diffusion state. The noise weighting matrix described above can be used to amplify the noise correction constraints imposed by demands that differ significantly from the actual target demand at different times. This allows the covariance term Σ(t) in the reverse sampling process to be adaptively adjusted based on the priorities of multiple target demands, helping the output data meet the multiple target demands in the current actual scenario.

[0124] In some embodiments, based on the above two embodiments of step S801, the reverse sampling process under the conditional diffusion model is adaptively updated to in, Adaptively aggregate the multi-objective requirements at the current moment and combine them with the noise vector dw after aggregation t Multiplying them together can achieve targeted noise injection tuning for each target requirement, further improving the accuracy of the output results.

[0125] S802 : Perform data generation processing based on the adaptively updated conditional diffusion model to obtain preliminary virtual sensing data.

[0126] S803: Generate virtual sensing data for compensating environmental data based on the preliminary virtual sensing data.

[0127] In this application, based on the existing traditional conditional diffusion model, physical constraints for correcting the diffusion trajectory are added in the reverse sampling of the conditional diffusion model, and the noise corresponding to the multi-target requirements is dynamically corrected by using a noise weighting strategy. The reverse sampling process of the conditional diffusion model can be automatically adapted as the scene evolves, and the problem of the lack of physical consistency and multi-target balancing ability in the existing technology is solved, so that the generated virtual sensor data covers multiple requirements in the current actual scene, thereby improving the accuracy and reliability of the virtual sensor data.

[0128] like Figure 9 As shown, generating virtual sensing data for compensating environmental data based on preliminary virtual sensing data specifically includes the following steps:

[0129] S901: Establish multiple scene conditions and determine the scene condition corresponding to the current actual scene.

[0130] More specifically, in real-world applications, the central air-conditioning system's control of the environment may vary, for example, due to fluctuations in the number of people at different times, differences in energy consumption patterns in different functional areas, external weather, etc. Multiple scenario conditions are established based on the multiple possible scenarios. For example, based on indicators such as peak and valley traffic flow, specific functional zones, seasons / external temperatures, the operating state is divided into multiple scenarios {s1, s2, ..., s K}, where s1 indicates weekday daytime (high traffic + higher cooling load), s2 indicates weekday evening (sudden drop in traffic + lights turned off in some areas), and s3 indicates special events on weekends (surge in demand in some areas).

[0131] When central air conditioning systems regulate the inter-building environment, real-time environmental data monitored by sensors is presented as multi-dimensional attributes of space, events, and needs. Data from the same target object (e.g., a room, device, or time period) is represented by several independent yet complementary representations based on different feature channels and business needs. This is known as multi-perspective.

[0132] In some embodiments, the multi-perspective of the present application includes a spatial perspective (ν space ), time perspective (ν time ), demand perspective (ν demand ) and the physical a priori perspective (ν pde The spatial perspective focuses on building topology, sensor placement, and spatial region division information; the temporal perspective emphasizes the temporal evolution and periodicity of data, such as the diurnal variation of cooling plant load; the demand perspective focuses on business or functional requirements, target energy consumption, indoor comfort, and equipment control priorities; and the physical prior perspective is the physical constraints in step S801.

[0133] In some embodiments, in order to express scene information in a multi-view model, a scene condition is established. Used to illustrate the key attributes of scene s (such as high load, low traffic, etc.). When the actual scene corresponding to the sample data is determined to be scene s, the scene condition E corresponding to the scene s is added to its multi-view feature input s , realizing scenario-based parameter adjustment of multi-view models.

[0134] S902. Based on the scene conditions, the preliminary virtual sensor data is subjected to multi-perspective feature extraction and fusion processing to obtain the multi-perspective collaborative features of the preliminary virtual sensor data under the scene conditions; the multi-perspective collaborative features are optimized using the cross-perspective coupling kernel function and the cross-perspective coupling divergence to generate optimized virtual sensor data under the scene conditions.

[0135] More specifically, the preliminary virtual sensing data includes multiple samples, different feature vectors are extracted from the same sample at different viewing angles, and the multi-view mapping operator is used to Obtain the implicit representation of the sample at each perspective.

[0136] In some embodiments, for a perspective index set If sample i is at the viewing angle ν v The corresponding eigenvector is Then through the multi-view mapping operator Obtaining implicit representation of perspective Right now in, Represents PDE auxiliary information (such as thermal and moisture field distribution gradient constraints).

[0137] In some embodiments, in order to explicitly capture the interactive information between perspectives, this application establishes a cross-perspective projection operator π with bilinear fusion and shared channel attention structure to extract the potential correlation between the two perspectives (e.g., the impact of spatial distribution on the target demand side). v ,ν w Implicit representation of The corresponding cross-view projection results are:

[0138] In some embodiments, the cross-view projection results of sample i in the preliminary virtual sensor data at different viewpoints are subjected to multi-view collaborative aggregation processing to generate a multi-view collaborative vector M of sample i i ,Right now Among them, the multi-view collaborative vector M i Contains learnable weights And the high-dimensional multi-view collaborative vector M of the sample i at all views i Serves as the master representation for subsequent contrastive learning.

[0139] In some embodiments, the present application establishes a cross-view coupling divergence and the corresponding cross-view coupling kernel function (κ CVC ), to measure the correlation between different samples (e.g., sample i and sample j) in the multi-view collaborative vector (e.g., the multi-view collaborative vector M of sample i). i and the multi-view collaborative vector M of sample j j ) on high-order interaction features (e.g., similarity and difference).

[0140] Optionally, the cross-view coupling divergence in, is a dynamic transformation operator that evolves with a continuous parameter t∈[0,1], aiming to expand the metric dimension and reflect the arrive high-order differences.

[0141] Optionally, the cross-view coupling kernel function κ CVC It is based on the cross-view coupling divergence Established, that is Among them, δ>0 is the scaling factor. When M i ,M j The greater the difference in the multi-view space, the greater the divergence The higher the cross-view coupling kernel function κ CVC tends to 0; if M i ,M j Similarity in multi-view space, divergence The lower the cross-view coupling kernel function value κ CVCClose to 1.

[0142] Optionally, multiple cross-view angles (e.g., ν v and ν w ) are accumulated to ensure that the feature differences of samples at different perspectives can be captured at all main perspective coupling levels.

[0143] In some embodiments, based on the above cross-view coupling divergence and the corresponding cross-view coupling kernel function (κ CVC ) Establish multi-view depth contrast loss In the process of multi-view model training, the positive sample pairs are brought closer and the negative sample pairs are pushed further away, so as to obtain a more discriminative multi-view representation of the preliminary virtual sensor data under multiple perspectives. Including pairs of the same object at different perspectives or times, as well as sample pairs with similar requirements / spatial attributes, negative sample pairs It refers to the sample pairs with significant spatial location differences when the demand-side conflict is serious (high energy consumption vs. low energy consumption targets).

[0144] For example, the formula of the two-stage contrast loss is as follows, Among them, l pos κ is used to encourage positive sample pairs CVC Taking a high value (i.e., similarity increases and divergence decreases), l neg Used to promote negative sample pairs Raise to avoid confusion in multiple viewing angles.

[0145] Optionally, the multi-view model is trained, and the training process specifically includes the following steps: based on the preliminary virtual sensor data, initializing Π,Ξ, Structure or parameters; in each round of training, positive and negative sample pairs are extracted Calculate M i With M j The divergence / kernel value of , thus accumulating the multi-view depth contrast loss And perform back propagation and gradient update on the parameters until the loss converges, then the multi-view collaborative vector M corresponding to sample i is obtained i or its dimensionality reduction variant The multi-view collaborative vector M corresponding to the sample i i or its dimensionality reduction variant The multi-view difference information is retained, and high-order features are formed under the zoom-in / pull-out mechanism of contrastive learning, that is, optimized virtual sensor data is obtained.

[0146] In some other embodiments, based on the scene condition E obtained in step S901 s , implicitly representing the perspective Updated to and Ω adopts a gating mechanism so that the representation within each perspective carries the adjustment preference of the current scene. For example, under scene condition s1 (weekday daytime), it emphasizes comfort matching; under scene condition s2 (nighttime), it emphasizes energy saving.

[0147] During the training process, after the positive sample pairs are brought closer and the negative sample pairs are pushed further away, the scene condition E obtained in step S901 is obtained. s , set up a scenario-based collaborative coordination unit to dynamically adjust the weight of each perspective in the cross-perspective coupling kernel function or cross-perspective coupling divergence. Among them, the scenario-based coordination unit f α is a learnable function (e.g., MLP). Input the scene condition E to the scene coordination unit s , you can output the viewing angle ν in scene s v The weight of the updated perspective implicit representation And the updated multi-view collaborative vector M i The expression of is used to generate virtual sensing data from multiple perspectives that match the current actual scene.

[0148] The positive and negative sample pairs are determined based on the scene conditions corresponding to the current actual scene. That is, in the same scene s, the sample pairs with similar perspective information are regarded as positive sample pairs; in different scenes, the sample pairs with dissimilar perspective information are regarded as negative sample pairs.

[0149] Update the multi-view depth contrast loss based on the scene conditions corresponding to the current actual scene Right now

[0150] Among them, β s(i),s(j) is the factor corresponding to the scene condition, and Using scenario-based kernel weights and scenario embedding mechanisms for calculation, β s(i),s(j) Provide different factors in cross-scenario and same-scenario situations.

[0151] Optionally, in the calculation κ CVC (M i ,M j ), if samples i and j are in scenes s(i) and s(j) respectively, we can use A scenario-based coordination unit with dual scene weighting. It can also be used directly after unifying the scene. s.

[0152] Optionally, based on normalization rules (e.g., ), which can achieve flexible allocation of attention among different perspectives.

[0153] In this application, by performing feature separation on the preliminary virtual sensor data generated by the conditional diffusion model under multiple perspectives, the technical problem of feature confusion under multiple perspectives in the existing technology is solved, the accuracy of the optimized virtual sensor data is enhanced, and the feature weights under multiple perspectives are redistributed through the scene conditions corresponding to the current actual scene, thereby achieving adaptive updating of the scene during the multi-perspective comparative learning process, improving the adaptability of the multi-perspective model, and enhancing the accuracy of the optimized virtual sensor data and its matching degree with the current actual scene.

[0154] In some embodiments, whether the current actual scene switches is monitored in real time; when the current scene switches, the scene conditions corresponding to the switched scene are determined, multi-perspective scene embedding processing is performed according to the scene conditions, and virtual sensor data optimized under the switched scene conditions is generated.

[0155] For example, since the scene s changes with time periods and event triggers during building operation, this application introduces a scene switching sequence in the multi-view model training stage. And monitor the emergence of new scenes in real time. When the scene switch is determined (for example, switching from s1 to s2 after get off work), the scene condition E is dynamically updated. s and multi-view depth contrast loss.

[0156] For example, when a scene switch is detected, the scene fusion vector is recalculated according to step S702. Then, according to step S703, a scene condition representation corresponding to the scene fusion vector after the scene switch is calculated. The virtual sensor data for the scene after the scene switch is further calculated based on the scene condition representation. The specific calculation method has been described in the above embodiment and will not be elaborated in detail in this embodiment.

[0157] In this application, by monitoring whether the current actual scene switches, the rapid generation of virtual sensor data in different scenes is achieved, the generation efficiency of virtual sensor data is improved, and the efficiency of environmental control is improved.

[0158] In some embodiments, scene matching detection is performed based on the generated virtual sensor data to obtain a matching difference value between the virtual sensor data and the current actual scene; when the matching difference value exceeds a preset threshold, the virtual sensor data is corrected through the correction network, and the corrected virtual sensor data is output.

[0159] For example, the present application introduces a dynamic correction unit Γ to correct the multi-view model. Or the parameters in Ω are lightly tuned, that is, Among them, Γ is built based on a meta-learning update module to enable the multi-view model to quickly adapt to new scenes without destroying its performance in the original scene.

[0160] In this application, the accuracy of the virtual sensor data is further improved by determining the matching difference value between the current virtual sensor data and the current actual scene, and triggering the correction network to correct the virtual sensor data when the matching difference value exceeds a preset threshold.

[0161] In some embodiments, after the virtual sensor data is corrected, multi-target fusion processing is performed on the corrected virtual sensor data to generate virtual sensor data that is used to compensate for the environmental data and meets multi-target requirements.

[0162] Exemplarily, based on the overall optimization or priority among multiple objective requirements, the corrected virtual sensor data is dynamically weighted to achieve a demand balance among the multiple objective requirements, so that the fused virtual sensor data can take into account the multiple objective requirements.

[0163] In some embodiments, in step S902, the implicit representation of the viewing angle is obtained. As the conditional input of the conditional diffusion model, noise is gradually introduced through the forward diffusion process, and the noise is removed using the reverse diffusion process to finally generate preliminary virtual sensing data The state update formula of the diffusion process at each time t is: D t ~q(D t ∣D t-1 ), where D t In step S802, the initial virtual sensing data is obtained. Afterwards, based on Determine the matching difference between the preliminary virtual sensor data and the current actual scene. If the matching difference exceeds the preset threshold δs, the subsequent correction is triggered. During the correction process, a correction network is constructed.

[0164] And input to the correction network Output corrected virtual sensor data in, The correction network dynamically adjusts the weight of each perspective according to the current actual scene to ensure that the corrected virtual sensor data meets the target requirements in different scenes. The obtained virtual sensor data Weighted fusion is performed with multi-objective requirements to generate a method for compensating environmental data. Among them, W s This application is used to weight different target requirements to further ensure that the refined control requirements of the central air-conditioning system are met.

[0165] Optionally, after the scene switches, new data is generated Continue to iterate and correct to ensure rapid adaptation to new scenarios. Finally, output the corrected and fused virtual sensor data. It will serve as input data for the central air-conditioning system to provide support for subsequent energy efficiency optimization and comfort control.

[0166] In this application, the corrected virtual sensor data is further adjusted so that the adjusted virtual sensor data can achieve a demand balance between multiple target requirements, improve the adaptability of the virtual sensor data to the current actual scene, and enhance the accuracy of environmental control.

[0167] S403: Use the virtual sensor data to compensate for the environmental data, and dynamically control the environment based on the compensated environmental data.

[0168] The environmental control method for a central air-conditioning system provided in an embodiment of the present application generates a physical field model based on building topology data and environmental data through a controller, generates a behavioral occupancy field model based on user behavior data in the current environment, fuses multi-target requirements with the behavioral occupancy field model and the physical field model to obtain a scene condition representation, and generates virtual sensor data based on the scene condition representation. Thus, by fusing the scene model with multi-target requirements, the generated virtual sensor data takes into account multi-target requirements, provides a reliable and detailed environmental data supplement for the central air-conditioning management system, improves the accuracy of the virtual sensor data and its adaptability in real scenarios, improves the environmental control efficiency, and enhances the precise control of the building space environment by the central air-conditioning system.

[0169] Figure 10 The schematic diagram of the structure of the environmental control device of the central air-conditioning system provided in this application is as follows: Figure 10 As shown, the environment control device 100 of the central air-conditioning system provided in this embodiment includes:

[0170] Acquisition module 1001, used to monitor the environmental data of the current actual scene in real time through multiple sensors;

[0171] Processing module 1002 is configured to obtain virtual sensor data for compensating the environmental data based on the environmental data and the building topology data through a controller; compensate the environmental data using the virtual sensor data; and dynamically control the current actual scene based on the compensated environmental data;

[0172] Among them, virtual sensing data is obtained based on environmental data and building topology data, including:

[0173] A physical field model is established based on environmental data and building topology data, and a behavioral occupancy field model is established based on user behavior data in the environment; a multi-objective demand mapping operator is generated based on the physical field model, and a scene condition representation is obtained based on the multi-objective demand mapping operator and the behavioral occupancy field model; virtual sensor data for compensating environmental data is generated based on the scene condition representation.

[0174] In some embodiments of the present application, the processing module 1002 is further configured to, when generating virtual sensor data for compensating environmental data based on the scene condition representation, generate an adaptively updated conditional diffusion model by adding physical constraints for correcting diffusion trajectories in reverse sampling of the conditional diffusion model based on the scene condition representation, and dynamically correcting noise corresponding to multiple objective requirements using a noise weighting strategy;

[0175] Data generation and processing are performed based on the adaptively updated conditional diffusion model to obtain preliminary virtual sensing data;

[0176] Virtual sensing data for compensating the environmental data is generated according to the preliminary virtual sensing data.

[0177] In some embodiments of the present application, the processing module 1002 is further configured to establish a plurality of scene conditions and determine a scene condition corresponding to a current actual scene when generating virtual sensor data for compensating for environmental data based on preliminary virtual sensor data;

[0178] Based on the scene conditions, the preliminary virtual sensor data is subjected to multi-perspective feature extraction and fusion processing to obtain the multi-perspective collaborative features of the preliminary virtual sensor data under the scene conditions. The multi-perspectives include spatial perspective, temporal perspective, demand perspective and physical prior perspective.

[0179] The multi-view collaborative features are optimized using the cross-view coupling kernel function and cross-view coupling divergence to generate optimized virtual sensing data under scene conditions.

[0180] In some embodiments of the present application, the processing module 1002 is further configured to perform scene matching detection based on the currently obtained virtual sensor data to obtain a matching difference value between the virtual sensor data and the current actual scene;

[0181] When the matching difference value exceeds a preset threshold, the virtual sensor data is corrected through the correction network and the corrected virtual sensor data is output.

[0182] In some embodiments of the present application, the processing module 1002 is further used to perform multi-target fusion processing on the corrected virtual sensor data after correcting the virtual sensor data, so as to generate virtual sensor data that is used to compensate for the environmental data and meet multi-target requirements.

[0183] In some embodiments of the present application, the processing module 1002 is further configured to monitor in real time whether a current actual scene is switched;

[0184] When the current actual scene is switched, the scene conditions corresponding to the switched scene are determined, and based on the switched scene conditions, optimized virtual sensing data under the switched scene conditions are generated.

[0185] In some embodiments of the present application, the processing module 1002 is further configured to determine, when establishing a physical field model based on the environmental data and the building topology data, dynamic distribution information of the temperature sensor, the humidity sensor, and the carbon dioxide concentration sensor in the building topology data based on the building topology data in the current actual scenario, wherein the building topology data is determined based on the connectivity relationship between the rooms in the current actual scenario;

[0186] Generate a sensor distribution tensor based on the dynamic distribution information; determine sensor confidence mapping data based on the sensor distribution tensor. For the same area, the sensor confidence mapping data is positively correlated with the observation density in the area;

[0187] Generate a physics model based on environmental data and sensor confidence map data.

[0188] In some embodiments of the present application, the processing module 1002 is further configured to determine sensor attention weight data based on the physical field model when generating a multi-objective demand mapping operator based on the physical field model, wherein the sensor attention weight data of any region is inversely correlated with the corresponding sensor confidence mapping data;

[0189] The multi-target demand mapping operator corresponding to the multi-target demand is determined according to the sensor attention weight data.

[0190] In some embodiments of the present application, the processing module 1002 is further configured to obtain a fused field model by fusing and aligning the behavior occupancy field model with the physical field model when obtaining a scene condition representation based on the multi-objective demand mapping operator and the behavior occupancy field model;

[0191] Determine the scene fusion vector corresponding to the multi-target requirements based on the fusion field model and the multi-target requirement mapping operator;

[0192] A scene condition representation is determined based on the scene fusion vector.

[0193] The environmental control device for the central air-conditioning system provided in this embodiment can execute the environmental control method for the central air-conditioning system provided in the above method embodiment. Its implementation principle and technical effects are similar, and are not described in detail in this embodiment.

[0194] An embodiment of the present application further provides an electronic device comprising at least one processor and a memory. Optionally, the device further comprises a communication component. The processor, the memory, and the communication component are connected via a bus.

[0195] In a specific implementation process, at least one processor executes computer-executable instructions stored in a memory, so that the at least one processor performs the above method.

[0196] The specific implementation process of the processor can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0197] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0198] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk memory.

[0199] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified into address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0200] The present application also provides a computer program product, including a computer program, which implements the above-mentioned environmental control method for the central air-conditioning system when executed by a processor.

[0201] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above-mentioned environmental control method of the central air-conditioning system is implemented.

[0202] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0203] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0204] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, either through an interface, electrical, mechanical, or other means.

[0205] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0206] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0207] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0208] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

[0210] For ease of explanation, the above description has been presented in conjunction with specific embodiments. However, the above exemplary discussion is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Based on the above teachings, various modifications and variations are possible. The above embodiments have been selected and described to better explain the principles and practical applications, thereby enabling those skilled in the art to better utilize the embodiments and various variations of the embodiments suitable for specific use considerations.

Claims

1. A central air conditioning system, characterized in that: include: Multiple sensors are used to monitor the environmental data of the current actual scene in real time; and, a controller for obtaining virtual sensing data for compensating the environmental data based on the environmental data and the building topology data; Using the virtual sensor data to compensate for the environmental data, and dynamically regulating the current actual scene based on the compensated environmental data; Wherein, obtaining virtual sensing data according to the environmental data and the building topology data includes: A physical field model is established based on the environmental data and the building topology data, and a behavioral occupancy field model is established based on the user behavior data in the environment; a multi-objective demand mapping operator is generated based on the physical field model, and a scene condition representation is obtained based on the multi-objective demand mapping operator and the behavioral occupancy field model; and the virtual sensor data for compensating the environmental data is generated based on the scene condition representation.

2. The central air conditioning system according to claim 1, characterized in that: When the controller generates the virtual sensor data for compensating the environmental data based on the scene condition representation, the controller is specifically configured to: Based on the scene condition representation, physical constraints for correcting diffusion trajectories are added to the inverse sampling of the conditional diffusion model, and the noise corresponding to the multi-objective requirements is dynamically corrected using a noise weighting strategy to generate an adaptively updated conditional diffusion model. Performing data generation processing based on the adaptively updated conditional diffusion model to obtain preliminary virtual sensing data; Virtual sensing data for compensating the environmental data is generated according to the preliminary virtual sensing data.

3. The central air conditioning system according to claim 2, characterized in that: When the controller generates virtual sensor data for compensating the environmental data according to the preliminary virtual sensor data, the controller is specifically configured to: Establish multiple scene conditions and determine the scene conditions corresponding to the current actual scene; Performing feature extraction and fusion processing on the preliminary virtual sensor data under multiple perspectives based on the scene conditions to obtain multi-perspective collaborative features of the preliminary virtual sensor data under the scene conditions, wherein the multiple perspectives include a spatial perspective, a temporal perspective, a demand perspective, and a physical prior perspective; The multi-view collaborative features are optimized using a cross-view coupling kernel function and a cross-view coupling divergence to generate optimized virtual sensing data under the scene conditions.

4. The central air conditioning system according to claim 2 or 3, characterized in that: The controller is further configured to: Performing scene matching detection based on the currently obtained virtual sensor data to obtain a matching difference value between the virtual sensor data and the current actual scene; When the matching difference value exceeds a preset threshold, the virtual sensor data is corrected through a correction network, and the corrected virtual sensor data is output.

5. The central air conditioning system according to claim 4, characterized in that: After correcting the virtual sensor data, the controller is further configured to: The corrected virtual sensor data is subjected to multi-target fusion processing to generate virtual sensor data that is used to compensate for environmental data and meet multi-target requirements.

6. The central air conditioning system according to claim 3, characterized in that: The controller is further configured to: Real-time monitoring of whether the current actual scene is switched; When the current actual scene is switched, the scene conditions corresponding to the switched scene are determined, and based on the switched scene conditions, optimized virtual sensing data under the switched scene conditions are generated.

7. The central air conditioning system according to claim 1, characterized in that: When the controller establishes the physical field model according to the environmental data and the building topology data, the controller is specifically configured to: Determining, based on building topology data in a current actual scenario, dynamic distribution information of temperature sensors, humidity sensors, and carbon dioxide concentration sensors in the building topology data, wherein the building topology data is determined based on connectivity relationships between rooms in the current actual scenario; Generating a sensor distribution tensor according to the dynamic distribution information; determining sensor confidence mapping data according to the sensor distribution tensor, wherein for a same area, the sensor confidence mapping data is positively correlated with the observation density in the area; The physical field model is generated based on the environmental data and the sensor confidence map data.

8. The central air conditioning system according to claim 1, characterized in that: When the controller generates a multi-objective demand mapping operator based on the physical field model, the controller is specifically configured to: Determine sensor attention weight data based on the physical field model, where the sensor attention weight data of any region is inversely correlated with the corresponding sensor confidence mapping data; The multi-target demand mapping operator corresponding to the multi-target demand is determined according to the sensor attention weight data.

9. The central air conditioning system according to claim 1, characterized in that: When the controller obtains the scene condition representation according to the multi-objective demand mapping operator and the behavior occupancy field model, the controller is specifically configured to: Obtaining a fused field model by fusing and aligning the behavioral occupancy field model with the physical field model; Determining a scene fusion vector corresponding to the multi-objective requirement based on the fusion field model and the multi-objective requirement mapping operator; The scene condition representation is determined based on the scene fusion vector.

10. A method for controlling the environment of a central air-conditioning system, characterized in that: The central air conditioning system includes a plurality of sensors and a controller, and the method includes: Monitor the environmental data of the current actual scene in real time through multiple sensors; Obtaining, by a controller, virtual sensor data for compensating the environmental data based on the environmental data and the building topology data; compensating the environmental data using the virtual sensor data, and dynamically regulating the current actual scene based on the compensated environmental data; Wherein, obtaining virtual sensing data according to the environmental data and the building topology data includes: A physical field model is established based on the environmental data and the building topology data, and a behavioral occupancy field model is established based on the user behavior data in the environment; a multi-objective demand mapping operator is generated based on the physical field model, and a scene condition representation is obtained based on the multi-objective demand mapping operator and the behavioral occupancy field model; and the virtual sensor data for compensating the environmental data is generated based on the scene condition representation.

Images