Cloud collaborative power consumption dynamic perception and energy-saving optimization control method and system

Abstract

This invention provides a cloud-based collaborative method and system for dynamic sensing and energy-saving optimization control of power consumption, relating to the field of energy-saving control technology. The method includes: local deployment of heterogeneous sensor arrays to construct multiple edge intelligent sensing nodes; receiving multiple multi-source spatiotemporal data streams, performing load-person flow coupling prediction, and generating multiple initial control strategies; comparing multiple control feedback data and multiple initial control strategies to generate a regional energy consumption deviation matrix; correcting the regional energy consumption deviation matrix based on multiple thermal coupling coefficients to obtain multiple neighborhood compensation strategies; performing spatiotemporal collaborative optimization to generate multiple dynamic optimization strategies; and performing zoned autonomous closed-loop control. This invention solves the technical problem that existing technologies mainly rely on fixed patterns or timed control for regional power consumption control, which cannot dynamically adjust according to the actual usage of each area within a building, leading to excessively high or low energy consumption in some areas and reducing the overall system efficiency.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology, specifically to a cloud-based collaborative method and system for dynamic perception and energy-saving optimization control of power consumption. Background Technology

[0002] Energy conservation and emission reduction have become critical issues that urgently need to be addressed across various industries, especially the construction industry. The energy consumption of building equipment such as air conditioning, lighting, and ventilation systems accounts for a large portion of a building's total energy consumption. Traditional methods primarily rely on fixed patterns or timed control for regional power consumption control, typically lacking real-time monitoring and feedback capabilities. This lack of real-time feedback prevents dynamic adjustments based on the actual usage in different areas of the building, leading to excessively high or low energy consumption in some areas. This ultimately results in an unbalanced distribution of energy efficiency, with some areas consuming excessively much energy, causing energy waste, increasing electricity consumption, and reducing the overall system efficiency. Summary of the Invention

[0003] This application provides a cloud-based collaborative method and system for dynamic perception and energy-saving optimization control of power consumption. It aims to solve the technical problem that existing technologies mainly rely on fixed modes or timed control for regional power consumption control, which cannot dynamically adjust according to the actual usage of each area in the building, resulting in excessively high or low energy consumption in some areas and reducing the overall efficiency of the system.

[0004] The first aspect disclosed in this application provides a cloud-based collaborative method for dynamic sensing and energy-saving optimization control of power consumption. The method includes: deploying a heterogeneous sensor array locally based on the zonal topology of a central air conditioning system to construct multiple edge intelligent sensing nodes corresponding to multiple independent control areas. The multiple edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via an RS485 bus. The energy-saving optimization cloud receives multiple multi-source spatiotemporal data streams transmitted from the multiple edge intelligent sensing nodes, performs load-population coupling prediction, and generates multiple initial control strategies. A regional energy consumption deviation matrix is ​​generated by comparing multiple control feedback data and the multiple initial control strategies. The regional energy consumption deviation matrix is ​​corrected based on multiple thermal coupling coefficients of the multiple independent control areas to obtain multiple neighborhood compensation strategies. Spatiotemporal collaborative optimization of the multiple initial control strategies is performed based on the multiple neighborhood compensation strategies to generate multiple dynamic optimization strategies. The energy-saving optimization cloud distributes the multiple dynamic optimization strategies to the multiple edge intelligent sensing nodes to perform zonal autonomous closed-loop control of the multiple independent control areas.

[0005] The second aspect of this application discloses a cloud-collaborative dynamic sensing and energy-saving optimization control system for power consumption. This system is used in the aforementioned cloud-collaborative dynamic sensing and energy-saving optimization control method for power consumption. The system includes: a local deployment module for deploying heterogeneous sensor arrays locally according to the zonal topology of a central air conditioning system, constructing multiple edge intelligent sensing nodes corresponding to multiple independent control areas, wherein the multiple edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via an RS485 bus; and a coupling prediction module for receiving multiple multi-source spatiotemporal data streams returned by the multiple edge intelligent sensing nodes and performing load-person flow coupling prediction on the energy-saving optimization cloud. The system generates multiple initial control strategies; a deviation matrix generation module generates a regional energy consumption deviation matrix by comparing multiple control feedback data and the multiple initial control strategies; a deviation matrix correction module corrects the regional energy consumption deviation matrix according to multiple thermal coupling coefficients of the multiple independent control regions to obtain multiple neighborhood compensation strategies; a spatiotemporal collaborative optimization module optimizes the multiple initial control strategies based on the multiple neighborhood compensation strategies to generate multiple dynamic optimization strategies; and a closed-loop control module distributes the multiple dynamic optimization strategies from the energy-saving optimization cloud to the multiple edge intelligent sensing nodes to perform partitioned autonomous closed-loop control of the multiple independent control regions.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects:

[0007] By deploying multiple edge intelligent sensing nodes in each independent control area of ​​the central air conditioning system, the operating status and energy consumption of each area can be monitored in real time. These sensing nodes communicate bidirectionally with the energy-saving optimization cloud via an RS485 bus, ensuring real-time data transmission and feedback, and achieving highly accurate area perception and dynamic control. The energy-saving optimization cloud performs load-person flow coupling prediction based on the multi-source spatiotemporal data streams transmitted from the edge sensing nodes and generates multiple initial control strategies. These strategies are based on the relationship between person flow density and equipment load, enabling the system to efficiently respond to load changes when adjusting equipment such as air conditioners and fans, avoiding overcooling or overheating, thereby achieving energy savings. By comparing multiple control feedback data with multiple initial control strategies, a regional energy consumption deviation matrix is ​​generated. This matrix reflects the difference between actual and expected energy consumption in each area, effectively identifying energy efficiency problems in the system and making timely adjustments. Corrections are made based on the thermal coupling coefficients of multiple independent control areas. A regional energy consumption deviation matrix is ​​generated, leading to multiple neighborhood compensation strategies. The thermal coupling coefficient reflects the heat transfer and influence between different regions. The corrected regional energy consumption deviation matrix more accurately reflects the energy efficiency of each region and rationally adjusts the heat distribution between regions. Based on multiple neighborhood compensation strategies, spatiotemporal collaborative optimization of multiple initial control strategies is performed to generate multiple dynamic optimization strategies. These strategies not only consider the energy efficiency of a single region but also comprehensively consider the thermal coupling relationship and spatiotemporal dynamic changes between regions, achieving collaborative optimization of the overall system, avoiding excessive or insufficient energy consumption adjustment, and improving the energy efficiency and comfort of the entire building system. The energy-saving optimization cloud distributes the generated dynamic optimization strategies to various edge intelligent sensing nodes for zoned autonomous closed-loop control. Each independent control area automatically adjusts the working status of equipment such as air conditioners and fans according to cloud instructions, while feeding back new status information to the cloud, forming a closed-loop control that achieves precise energy efficiency management without manual intervention.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 A schematic diagram of the cloud-based collaborative dynamic perception and energy-saving optimization control method for power consumption provided in this application embodiment.

[0010] Figure 2 A schematic diagram of the cloud-based collaborative dynamic power consumption perception and energy-saving optimization control system provided in this application embodiment.

[0011] Figure labeling: Local deployment module 10, Coupled prediction module 20, Deviation matrix generation module 30, Deviation matrix correction module 40, Spatiotemporal collaborative optimization module 50, Closed-loop control module 60. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1 As shown in the embodiments of this application, a cloud-based collaborative method for dynamic sensing and energy-saving optimization control of power consumption is provided. The method includes:

[0014] Based on the zoning topology of the central air conditioning system, a heterogeneous sensor array is deployed locally to construct multiple edge intelligent sensing nodes corresponding to multiple independent control areas. The multiple edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via an RS485 bus.

[0015] The central air conditioning system's zoned topology is clearly defined. The control zones of the central air conditioning system are divided according to different areas within the building, such as different floors, rooms, and areas. Within each control zone, the types and number of sensors installed are selected based on the zone's needs, including temperature sensors, humidity sensors, airflow sensors, CO2 concentration sensors, and personnel flow sensors. Heterogeneous design here refers to the use of multiple different types of sensors within the same control zone. These sensors form an array to collect environmental data related to the operation of the central air conditioning system.

[0016] Sensors in each control area collect and process data through edge intelligent sensing nodes. These nodes are not merely data acquisition points but also possess processing capabilities, enabling preliminary data analysis, filtering, and preprocessing locally. These nodes are connected via an RS485 bus, facilitating bidirectional communication with the cloud. As an industrial communication standard, RS485 supports remote connections for multiple devices and exhibits strong anti-interference capabilities. The edge intelligent sensing nodes transmit and receive data with the energy-saving optimization cloud via the RS485 bus.

[0017] The energy-saving optimization cloud receives multiple multi-source spatiotemporal data streams transmitted back from multiple edge intelligent sensing nodes, performs load and population flow coupling prediction, and generates multiple initial control strategies.

[0018] The data transmitted back from each edge intelligent sensing node includes multi-dimensional spatiotemporal data streams, covering sensing information such as temperature, humidity, airflow, and CO2 concentration, as well as dynamic information such as personnel movement and entry / exit. This data includes not only the temporal dimension but also the spatial dimension, such as sensor data from different locations within each area.

[0019] Load refers to the energy consumption load of the air conditioning system, that is, the power consumption and operating demand of the air conditioner in different control areas. Pedestrian flow refers to the activity patterns of people, which is directly related to the air conditioning load. For example, in areas with a large number of people, the air conditioning load demand is higher. The energy-saving optimization cloud platform, based on the backed-out multi-source spatiotemporal data stream, uses machine learning models such as regression analysis, neural networks, and time series prediction to model and predict the coupling relationship between load and pedestrian flow, calculating the air conditioning load demand and pedestrian flow distribution in different time periods and areas.

[0020] Based on the forecast results, multiple initial control strategies are generated to adjust the operating status of the air conditioning system, such as temperature settings, fan speed adjustments, and zone cooling or heating strategies. These initial control strategies are generated based on coupled load and occupancy forecasts, aiming to optimize the energy efficiency of the air conditioning system and avoid over-operation or excessive energy waste.

[0021] A regional energy consumption deviation matrix is ​​generated by comparing multiple control feedback data and the multiple initial control strategies.

[0022] Control feedback data refers to data obtained from the actual operation of the air conditioning system, such as energy consumption, temperature changes, and changes in air quality within the area. This data reflects the performance of the air conditioning system after actual control measures are implemented and is fed back to the cloud system via sensors. The initial control strategy is compared with the actual control feedback data to analyze whether the initial control strategy achieved the expected energy efficiency optimization effect. For example, if the initial control strategy requires the temperature in a certain area to be maintained at 26℃, but the feedback data shows the temperature is 28℃, it indicates that the initial control strategy was not effectively implemented.

[0023] Based on the differences between the control feedback data and the initial control strategy, the energy consumption deviation for each region is calculated. For example, if the actual energy consumption of a region is 20% higher than expected, its energy consumption deviation is 20%. These deviation data constitute a regional energy consumption deviation matrix, which shows the gap between the energy consumption performance of different regions and the initial strategy. This matrix helps identify which regions' control strategies need adjustment and which regions have energy waste, thus providing a basis for optimizing air conditioning systems.

[0024] Based on the multiple thermal coupling coefficients of the multiple independent control regions, the regional energy consumption deviation matrix is ​​corrected to obtain multiple neighborhood compensation strategies.

[0025] The thermal coupling coefficient refers to the degree of mutual influence between different control zones due to factors such as building structure, airflow, and heat exchange. For example, if the air conditioning load in one zone is too high, it will affect the temperature and energy consumption of adjacent zones; conversely, a zone with a lower air conditioning load will compensate for the adjacent zones, reducing their burden. By analyzing the heat conduction paths of a building, such as walls, ceilings, and ventilation systems, the thermal coupling coefficient between each pair of control zones can be calculated, reflecting their mutual influence in air conditioning control.

[0026] The original regional energy consumption deviation matrix is ​​based on the difference between the actual and expected energy consumption of each control region. This matrix provides a reference for the energy efficiency deviation of each region, but it does not consider the mutual influence between regions. Introducing a thermodynamic coupling coefficient into this matrix corrects the energy consumption deviation between different regions. For example, if a region has a large energy consumption deviation and a strong thermodynamic coupling relationship with neighboring regions, the control strategy for that region can be appropriately adjusted to reduce the negative impact on the surrounding area. Based on the corrected energy consumption deviation matrix, algorithms such as linear optimization and optimization problem solving are used to generate a neighborhood compensation strategy for each region, thereby improving the overall system's energy efficiency.

[0027] Based on the multiple neighborhood compensation strategies, the multiple initial control strategies are spatiotemporally coordinated to generate multiple dynamic optimization strategies.

[0028] Spatiotemporal co-optimization refers to the joint optimization of strategies across two dimensions: time and space. Temporally, strategies are adjusted based on demand at different times of the day, such as the difference between day and night. Spatially, the allocation of air conditioning load is optimized based on energy efficiency requirements in different regions. Through spatiotemporal co-optimization, multiple dynamic optimization strategies are generated. Each strategy is based not only on current demand and energy efficiency forecasts but also on future demand changes, such as variations in air conditioning load and population movement. These strategies are dynamic because they adjust dynamically with changes in the environment, building usage, and the operating status of the air conditioning system.

[0029] The energy-saving optimization cloud sends multiple dynamic optimization strategies to multiple edge intelligent sensing nodes to perform partitioned autonomous closed-loop control of the multiple independent control areas.

[0030] Based on the control needs of each control area, the energy-saving optimization cloud sends the corresponding dynamic optimization strategies to each edge intelligent sensing node. After receiving the dynamic optimization strategies from the cloud, each edge intelligent sensing node automatically adjusts according to local sensing data and current environmental conditions. For example, it adjusts the air conditioner's temperature setting, fan speed, or cooling / heating mode based on data such as temperature, humidity, and personnel flow.

[0031] Zonal autonomy refers to the ability of each control zone to independently regulate its air conditioning system based on dynamic optimization strategies deployed from the cloud, without the need for continuous centralized intervention. Closed-loop control means that during zonal autonomous control, edge intelligent nodes in each zone continuously collect real-time data and feed it back to the cloud for subsequent optimization and adjustments. The cloud further optimizes its strategies based on the feedback data, ensuring the air conditioning system maintains optimal energy efficiency over the long term. This closed-loop control enables precise adjustment of each control zone, ensuring the overall energy-saving effect of the central air conditioning system while improving user comfort.

[0032] Furthermore, the energy-saving optimization cloud receives multiple multi-source spatiotemporal data streams transmitted back from the multiple edge intelligent sensing nodes, performs load-population coupling prediction, and generates multiple initial control strategies, including:

[0033] Using the first independent control area as the spatial boundary, a spatiotemporal fusion model of the first multi-source spatiotemporal data stream is performed to generate a first environmental state tensor. The first environmental state tensor is decoupled and extracted to obtain a first regional residence index, a first inter-regional migration intensity, a first air conditioning power change gradient, and a first thermal inertia coefficient. The first regional residence index, the first inter-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient are used as physical information decision factors to perform load-population coupling prediction and generate a first initial control strategy.

[0034] Define a first independent control area, such as a floor, room, or isolated zone, as the spatial boundary for analysis. Multi-source refers to multiple data streams from various types of sensors, including temperature, humidity, airflow, and pedestrian flow sensors. Spatiotemporal data streams refer to data that includes not only temporal but also spatial information; for example, temperature, humidity, and pedestrian flow data at a specific point in time within the area. Spatiotemporal fusion modeling combines these multi-source spatiotemporal data streams to generate a multi-dimensional state tensor. This tensor contains not only data at each time step but also data from various spatial locations. The generated first environmental state tensor represents the comprehensive environmental state of the area at a given moment and can serve as the basis for subsequent analysis and decision-making.

[0035] Spatiotemporal feature decoupling refers to separating different features in time and space from a multidimensional environmental state tensor. Specifically, the decoupling process involves extracting meaningful features from the time and space dimensions, which involve dynamic changes in different dimensions within the region, such as temperature and humidity changes.

[0036] The first zone dwell time index measures the time or density of people staying in a certain area. A high dwell time index means there are more people in that area. This index reflects the duration and density of people staying in the area, which is directly related to air conditioning demand. The first inter-zone migration intensity reflects the intensity of people migrating from one area to another. This migration affects the air conditioning load distribution. For example, when a large number of people move from one area to another, the air conditioning load of the latter needs to increase. This index can be derived by analyzing people flow data and combining it with spatial location information. The first air conditioning power change gradient reflects the rate or magnitude of air conditioning power change. Air conditioning power change is closely related to changes in ambient temperature, people flow in the area, and changes in heat load within the area. This feature helps predict how much power the air conditioning system needs to meet demand under different time and environmental conditions. The first thermal inertia coefficient refers to the responsiveness of buildings or rooms in an area to temperature changes. Areas with high thermal inertia usually have slower temperature changes, and the air conditioning system can adjust more slowly; while areas with low thermal inertia have faster temperature changes, and the air conditioning needs to respond more quickly. This coefficient affects how precisely the air conditioning system adjusts to ensure a comfortable temperature in a dynamic environment.

[0037] The first regional residence index, the first inter-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient are used as physical information decision factors to jointly determine the coupling relationship between air conditioning load and population flow. The coupling relationship between load and population flow refers to the impact of population flow on air conditioning load. Since population density, residence time, and inter-regional migration within a region affect air conditioning load demand, these factors need to be modeled and predicted. Machine learning algorithms, such as regression models, neural networks, and time series analysis, are used to predict air conditioning load demand under different conditions. Based on the load prediction results and combined with the coupling model, a first initial control strategy is generated, which includes a temperature setpoint sequence and a fan speed sequence.

[0038] Furthermore, the first regional residence index, the first inter-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient are used as physical information decision factors to perform load-population coupling prediction and generate a first initial control strategy, including:

[0039] A pre-constructed load-person flow coupling prediction model is established, comprising a temporal convolutional branch, a graph attention branch, and an attention fusion layer. The temporal convolutional branch and the graph attention branch are connected in parallel, and their outputs are fed into the attention fusion layer. The first air conditioning power change gradient and the first thermal inertia coefficient are input into the temporal convolutional branch, and the dynamic load response vector and thermal inertia time code are extracted via an expanded causal convolutional layer. The first regional residence index and the first cross-regional migration intensity are input into the graph attention branch for spatial neighborhood aggregation, outputting a spatial person flow coupling matrix and a regional thermal potential field. The temporal convolutional branch and the graph attention branch perform feature extraction operations in parallel. The dynamic load response vector, thermal inertia time code, spatial person flow coupling matrix, and regional thermal potential field are weighted and fused in the attention fusion layer to output a predicted load time series curve. Based on the predicted load time series curve, the first initial control strategy is optimized in the rolling time domain.

[0040] A pre-built load-person flow coupling prediction model is constructed, including a temporal convolutional branch, a graph attention branch, and an attention fusion layer. The temporal convolutional branch uses a temporal convolutional network to capture long-term dependencies in time-series data. Temporal convolutional networks are suitable for processing time-series data and can effectively extract patterns and trends from historical data to predict dynamic changes in air conditioning load. The input of the temporal convolutional branch is time-related data such as changes in air conditioning power and personnel flow. The graph attention branch uses a graph attention network to model the spatial correlation between regions. Since there may be mutual influences between different control regions, such as thermal coupling, the graph attention mechanism can learn the relative importance between different regions, helping the model focus on those regions closely related to load changes. The attention fusion layer is responsible for fusing the outputs of the temporal convolutional branch and the graph attention branch. The attention mechanism can dynamically assign different weights to different information sources to obtain the optimal feature fusion result. This layer integrates information from the temporal convolutional branch and the graph attention branch through weighted averaging or other fusion strategies, and finally outputs a comprehensive prediction result.

[0041] The parallel operation of the temporal convolutional branch and the graph attention branch means that these two networks work in parallel, extracting spatiotemporal features from different perspectives. This parallel structure helps to learn features independently from both temporal and spatial dimensions, avoiding over-reliance on information from one dimension. When the outputs of these two parts are fed into the attention fusion layer, the contribution of each part can be dynamically adjusted, adaptively fusing temporal information and spatial relationships according to the characteristics of the current data.

[0042] Dilated convolution is a technique that captures longer-term dependencies by expanding the receptive field of the convolutional kernel. It is very useful in time series forecasting because it can capture long-term temporal dependencies without increasing computational cost. Causal convolution ensures that when a model makes time series forecasts, the output value depends only on the current or previous inputs and does not leak future temporal information. This is crucial for time series forecasting tasks because it ensures that the model does not see future real data in advance.

[0043] The dynamic load response vector captures the time-varying pattern of air conditioning power, describing the fluctuations, adjustment trends, and correlations with other environmental factors of the air conditioning load. Changes in the thermal inertia coefficient over different time periods affect the air conditioning control strategy. By extracting the thermal inertia time code through causal convolutional layers, the model can be provided with the thermal response characteristics of each region, enabling it to consider the temperature regulation characteristics of different areas within the building when predicting air conditioning load.

[0044] Graph attention branching can effectively capture the interactions between different areas in a space. Through the graph attention mechanism, an attention weight can be assigned to each area to represent its importance in the overall building heat and population flow network. Spatial neighborhood aggregation refers to combining the residence index and inter-area migration intensity of each area with information from neighboring areas in the graph attention branching to form a graph structure that reflects the interactions and influences between different areas.

[0045] The spatial pedestrian flow coupling matrix reflects the relationship and intensity of pedestrian flow between different control areas. The elements in the matrix represent the intensity of pedestrian migration between different areas. This matrix can be used to infer the mutual influence of air conditioning loads between different areas. The regional thermal potential field refers to the distribution of heat energy within a region. Through graph attention mechanisms, the propagation and distribution of heat between different areas can be simulated, helping to understand the dynamic process of heat conduction in buildings. The regional thermal potential field reflects the relative positional relationship of heat in different areas, guiding how the air conditioning system adjusts temperature distribution.

[0046] The parallel execution of the temporal convolutional branch and the graph attention branch helps to extract key features from both temporal and spatial perspectives. Parallel computation can accelerate model training and improve efficiency. The temporal convolutional branch extracts dynamic features related to time, such as the trend of air conditioning load changes over time, while the graph attention branch focuses on spatial features such as pedestrian flow and heat coupling.

[0047] The attention fusion layer uses an attention mechanism to weightedly fuse the above inputs. This mechanism dynamically allocates weights based on the importance of each feature, highlighting the features most important for load forecasting. Weighted fusion helps the model automatically adjust the contribution of each feature based on current environmental conditions, population movement, and other data, thereby generating the optimal load forecast. After weighted energy fusion, the output is a predicted load time-series curve, which represents the expected changes in air conditioning load over a future period.

[0048] Rolling time-domain optimization refers to recalculating and optimizing the control strategy based on the latest load forecast results at each moment. This means that the air conditioning system will continuously update its control strategy over time to ensure that it always adapts to real-time environmental demands. During the rolling optimization process, load forecasts for a future period are considered, and the operating mode of the air conditioning is gradually adjusted, such as temperature setting, fan speed, and cooling / heating mode. Through rolling time-domain optimization, new control strategies are continuously generated, and these strategies are distributed to edge intelligent sensing nodes for execution.

[0049] Furthermore, taking the first independent control region as the spatial boundary, a spatiotemporal fusion model of the first multi-source spatiotemporal data stream is performed to generate a first environmental state tensor, including:

[0050] Extract the first pedestrian flow heatmap stream, the first equipment power time series stream, the first ambient temperature and humidity, and the first CO2 concentration data from the first multi-source spatiotemporal data stream; take the first independent control area as the spatial boundary, perform spatiotemporal point cloud rigid registration of the first pedestrian flow heatmap stream and the first equipment power time series stream to generate a first fused spatiotemporal grid; on the first fused spatiotemporal grid, stack and fuse the first ambient temperature and humidity and the first CO2 concentration data through tensor channels to construct the first environmental state tensor.

[0051] The first human flow heat map, generated by human flow sensors or thermal imaging equipment, reflects the distribution and flow of people within an area. The heat map represents the human density or activity intensity in different areas at a certain moment or time period. The first device power time series records the power consumption of the air conditioning system or other equipment, provided in the form of a time series, and can reveal the energy consumption trend of the air conditioning system. The first environmental temperature and humidity data are key factors in air conditioning control. Temperature affects the cooling / heating demand of the air conditioning, while humidity affects human comfort and the operating mode of the air conditioning. The first CO2 concentration data is an important indicator for measuring indoor air quality. High CO2 concentrations are related to high human density or poor ventilation. In air conditioning systems, CO2 concentration is related to ventilation regulation.

[0052] The first independent control area is used as the spatial boundary of the analysis, within which all sensor data are integrated. Spatiotemporal point clouds refer to multidimensional datasets with timestamps and spatial locations. For example, the first pedestrian flow heatmap stream provides the spatial distribution of pedestrian density at different time points, and the first equipment power time series stream provides the time series of air conditioner power. Rigid registration of spatiotemporal point clouds refers to aligning data from different data sources according to spatial and temporal dimensions. Rigid registration means that the spatial location and temporal order of the data remain unchanged during the registration process, achieved through registration algorithms such as the iterative nearest-point algorithm and mutual information methods.

[0053] Registration ensures complete spatial and temporal consistency between the first human flow heatmap and the first equipment power time series, eliminating alignment errors between different data sources. After registration, a first fused spatiotemporal mesh is generated, which integrates all spatiotemporally registered data for subsequent analysis and modeling.

[0054] A tensor is a multidimensional data structure, which can be viewed as a high-dimensional extension of a matrix. In this step, tensors are used to represent multidimensional data such as spatial, temporal, temperature, humidity, and CO2 concentration. Channel stacking refers to stacking different data along different dimensions of a tensor to form a new multidimensional tensor. Based on the spatiotemporal grid, these data are organized and fused in terms of time and space. Through stacking, temperature, humidity, and CO2 concentration data can be added as additional dimensions to the spatiotemporal grid, forming a multidimensional environmental state tensor. Each data point includes not only location and time information but also environmental variables such as temperature, humidity, and CO2 concentration. Through tensor channel stacking, the final first environmental state tensor contains all the key environmental information, and this information is aligned in both time and space.

[0055] Furthermore, by comparing multiple control feedback data and the multiple initial control strategies, a regional energy consumption deviation matrix is ​​generated, including:

[0056] After performing topological mapping of the multiple control feedback data and the multiple initial control strategies based on the multiple independent control regions, multiple energy consumption integral deviations are calculated; a zero matrix is ​​constructed based on the multiple independent control regions; the main diagonal elements of the zero matrix are filled using the multiple energy consumption integral deviations; after quantifying the neighborhood energy consumption interference intensity of the multiple independent control regions according to the partition topology, the off-diagonal elements of the zero matrix are filled to complete the construction of the regional energy consumption deviation matrix.

[0057] Topology mapping refers to mapping feedback data and initial control strategies from multiple independent control areas into a unified structure. This structure is a matrix or graph used to represent the relationships or influences between the various areas. Energy consumption integral deviation refers to the cumulative error between actual and expected energy consumption. This error is calculated by integrating the energy consumption differences over multiple time steps.

[0058] The zero matrix is ​​a matrix where all elements are zero. It is gradually filled in during subsequent steps and eventually transformed into an energy consumption deviation matrix. The dimension of the matrix is ​​the same as the number of independent control regions, meaning each matrix element corresponds to one control region. Assuming there are n control regions, the size of the matrix is ​​n×n.

[0059] In the matrix, the elements on the main diagonal typically represent the self-energy consumption deviation of each region. For the i-th control region, the elements M on the main diagonal... ii This reflects the energy efficiency differences within the region. The energy consumption integral deviation of each region will be filled onto the main diagonal of the zero matrix, assuming d i If the energy consumption integral deviation is in the i-th region, then the main diagonal element M of the matrix is... ii =d i This filling method allows for the initial establishment of energy consumption error information for each region. The filled matrix contains the deviation values ​​of each region's self-energy consumption, reflecting the operational deviations of the air conditioning systems in each region, thus providing a foundation for subsequent neighborhood interference quantification and matrix filling.

[0060] Neighborhood energy consumption interference intensity refers to the degree to which changes in energy efficiency in one area affect neighboring areas. This quantifiable value is related to factors such as thermal coupling coefficient, population flow, and building layout. The interference coefficient between each pair of areas can be obtained by calculating the interference intensity between any two areas, for example, through regression analysis based on physical models or historical data. Off-diagonal element M ij (where i≠j) represents the energy efficiency interference between region i and region j. These off-diagonal elements can be filled based on the quantification results of the neighborhood interference intensity.

[0061] By filling in the main diagonal and off-diagonal elements, a regional energy consumption deviation matrix is ​​finally constructed. This matrix not only reflects the energy efficiency deviation of each region, but also takes into account the mutual influence between regions, providing a necessary basis for subsequent optimization and regulation strategy generation.

[0062] Furthermore, by correcting the regional energy consumption deviation matrix based on multiple thermal coupling coefficients of the multiple independent control regions, multiple neighborhood compensation strategies are obtained, including:

[0063] Based on the partitioned topology, multiple sets of adjacent control regions of the multiple independent control regions are extracted; multiple sets of neighborhood interface thermal conductivity and multiple sets of shared interface area of ​​the multiple sets of adjacent control regions are extracted from the building BIM model; using the multiple sets of adjacent control regions as matrix element coordinates, a symmetric thermal coupling matrix is ​​constructed based on the multiple sets of neighborhood interface thermal conductivity and multiple sets of shared interface area; the symmetric thermal coupling matrix is ​​used to correct the regional energy consumption deviation matrix using the Hadamard product, generating a thermal coupling correction deviation matrix; the thermal coupling correction deviation matrix is ​​decomposed to extract multiple row vector coupling strengths of the multiple independent control regions; the multiple row vector coupling strengths are fitted with gain adjustment to output the multiple neighborhood compensation strategies.

[0064] A zoned topology represents the connectivity between control zones. Each control zone is adjacent to other zones, and these adjacent zones have thermodynamic interactions, such as heat transfer through walls, windows, or other structures. Adjacent control zones refer to those control zones that are spatially connected or physically linked; heat conduction and energy exchange between these zones affect each other's air conditioning control.

[0065] Building BIM model is a digital tool in building design and construction. It contains detailed information such as the geometry, structure, and materials of a building. In the building BIM model, the connection relationships of all building components and areas are clear, including walls, doors and windows, floors, roofs, etc.

[0066] By utilizing building BIM models, heat exchange information between various control zones within a building can be extracted, such as thermal conductivity and area. This information is crucial for modeling the thermal coupling matrix. The thermal conductivity of a neighborhood interface refers to the ability of a material or interface to transfer heat. Different building materials, such as walls and windows, have different thermal conductivity. Through the BIM model, the thermal conductivity of structures such as walls and floors connecting adjacent areas can be extracted. Shared interface area refers to the area of ​​the physical interface shared between adjacent areas, such as the area of ​​walls or floors. This area determines the efficiency of heat transfer; the larger the area, the stronger the heat transfer capacity.

[0067] The adjacency relationships of multiple control regions are represented as a matrix, where each matrix element M ij This represents the thermal coupling strength between region i and region j. Since heat transfer is bidirectional—meaning the influence of region i on region j is the same as the influence of region j on region i—this matrix is ​​symmetric. That is, M ij =M ji In a symmetric thermal coupling matrix, the thermal coupling strength between each control region and its adjacent regions is determined by their interfacial thermal conductivity and shared interfacial area. In the matrix, the elements M on the main diagonal... iiThe elements represent the thermal coupling strength between each region and itself. These elements do not involve physical meaning because there is no interface between the region and itself; however, they can be used to fill the thermal resistance of the region itself. The off-diagonal elements represent the thermal conduction effects between regions, ultimately constructing a complete thermal coupling matrix to describe the heat exchange relationships between the various control regions in the entire system.

[0068] The Hadamard product is the element-wise product of two matrices; that is, given two matrices A and B, their Hadamard product is... It is a matrix of the same size, whose elements are obtained by multiplying the corresponding elements of A and B: In this step, a Hadamard product operation is performed on the symmetric thermal coupling matrix and the regional energy consumption deviation matrix. The purpose is to incorporate thermal coupling information into the regional energy consumption deviation matrix. In this way, the thermal coupling matrix and the energy consumption matrix are combined to obtain the thermal coupling correction deviation matrix. The elements of the thermal coupling correction deviation matrix not only contain the original energy efficiency deviation information but also consider the influence of heat conduction between regions. Therefore, this matrix reflects the actual energy efficiency differences between regions more accurately than the original regional energy consumption deviation matrix.

[0069] Matrix decomposition is the process of representing a matrix as a product of multiple submatrices (or vectors). In this step, the thermal coupling correction deviation matrix is ​​decomposed to extract the row vector coupling strength of each control region. The row vector coupling strength represents the degree of thermal coupling and energy efficiency deviation between a certain region and all other regions. By decomposing the matrix, the influence of each region on other regions can be obtained, i.e., the thermal conduction coupling strength. In the decomposed matrix, each row vector represents the relationship between a control region and other regions, and each element in the row vector represents the thermal coupling strength between that region and other regions.

[0070] Gain adjustment refers to adjusting and optimizing the coupling strength extracted from the row vectors. This process is based on control theory or machine learning techniques, such as linear regression, least squares, or other optimization algorithms, to adjust the coupling strength of each region, making heat transfer between regions more balanced and optimizing the overall energy efficiency of the system. By adjusting the coupling strength, heat transfer between regions can be better addressed, ensuring that energy consumption is minimized while maintaining comfort. After gain adjustment fitting, the adjustment parameters for each region are the neighborhood compensation strategy. This strategy adjusts the influence of heat conduction between regions, enabling the system to dynamically optimize the air conditioning control strategy for each region based on the results of the thermal coupling correction deviation matrix.

[0071] Furthermore, the multiple row vector coupling strengths are fitted with gain adjustment to output the multiple neighborhood compensation strategies, including:

[0072] Multiple region gain coefficients are configured based on the heat capacity of the multiple independent control regions; the overall gain scaling of the coupling strength of the multiple row vectors is performed using the multiple region gain coefficients to output multiple gain adjustment row vectors; scalar aggregation is performed on the multiple gain adjustment row vectors to generate multiple compensation reference quantities; dual-thread decoupling transformation is performed on the multiple compensation reference quantities to output multiple temperature compensation strategies and multiple fan compensation strategies, which constitute the multiple neighborhood compensation strategies.

[0073] Heat capacity refers to the amount of heat stored in a region under a unit temperature change. It is determined by factors such as the building materials, air density, and volume of the region. Regions with higher heat capacity require more heat to cause a larger temperature change, thus their thermal response is slower; while regions with lower heat capacity experience faster temperature changes. In the regulation process of air conditioning systems, heat capacity affects the temperature response time and energy efficiency optimization of each region. The larger the heat capacity of a region, the longer the temperature regulation process takes, and therefore the corresponding regulation needs to be more precise.

[0074] Based on the heat capacity of each zone, a corresponding gain coefficient is assigned to each control zone. The gain coefficient is a scaling factor used to adjust the sensitivity of the zone control strategy. Zones with larger heat capacities are assigned lower gain coefficients because their temperatures change more slowly, requiring the air conditioning system to reduce its adjustment intensity in these zones. Zones with smaller heat capacities are assigned higher gain coefficients because their temperatures change more rapidly, requiring the air conditioning system to adjust more sensitively.

[0075] By applying multiple regional gain coefficients, the coupling strength of each row vector is scaled. Specifically, each element in the row vector (i.e., the heat transfer intensity between regions) is multiplicatively scaled according to the heat capacity gain coefficient of that region. For regions with larger gain coefficients, their corresponding row vectors are magnified, indicating that these regions are more sensitive to the thermal influence of other regions; for regions with smaller gain coefficients, their corresponding row vectors are shrunk, indicating that these regions have a slower thermal response. The scaled row vectors are called gain-adjusted row vectors, and they represent the regional coupling strength after correction by the thermal coupling correction bias matrix.

[0076] Scalar aggregation refers to merging elements from multiple gain adjustment row vectors to obtain a single, comprehensive compensation baseline for each region. These baselines are composite measures adjusted by multiple row vectors and used to describe the overall heat conduction effect in that region. The aggregation process can be accomplished in various ways, such as calculating the weighted average, maximum, and minimum values ​​of all elements in the row vectors. The compensation baseline for each region represents the overall thermal coupling effect of that region and can be understood as the compensation value that needs to be considered when adjusting the temperature in that region.

[0077] Dual-thread decoupling conversion refers to decomposing the compensation baseline into two different control strategies: a temperature compensation strategy and a fan compensation strategy. These two strategies target temperature regulation within a region and the operation regulation of the air conditioning fan, respectively. The temperature compensation strategy focuses on adjusting the temperature of the region based on the compensation baseline to maintain it within an ideal comfort range. The fan compensation strategy focuses on how to assist temperature regulation by adjusting the fan's operating speed or mode, ensuring air circulation and preventing the region from becoming too hot or too cold. Through these two compensation strategies, multiple neighborhood compensation strategies are ultimately output. These strategies are then distributed to various control areas to achieve dynamic temperature and fan regulation.

[0078] Furthermore, it also includes:

[0079] During the execution of the multiple edge intelligent sensing nodes and the multiple dynamic optimization strategies, the nodes simultaneously perform real-time dynamic operation status perception of the multiple independent control areas, and obtain multiple multi-source incremental data streams. Based on the abnormal fluctuation characteristics of the multiple multi-source incremental data streams, the nodes identify control deviations in the multiple independent control areas and locate P abnormal control areas. The nodes then perform power equipment fault location and maintenance on the P abnormal control areas.

[0080] Each edge intelligent sensing node controls system parameters such as air conditioning, fans, and temperature based on the previously generated dynamic optimization strategy, aiming to optimize energy efficiency and comfort. During execution, the edge intelligent sensing node in each control area collects environmental parameters, equipment status, and other key indicators in real time. Dynamic operational status sensing means that these sensing nodes not only collect data periodically but also dynamically track changes in the control area to ensure the system's real-time adaptability and flexibility. In each control area, the sensing node processes incremental data streams from multiple sensors. These data streams reflect real-time changes in monitoring information. Incremental data streams refer to continuously generated data, reflecting the changes in the sensing data at each moment compared to the data at the previous moment.

[0081] In multiple multi-source incremental data streams, real-time data fluctuations in each control area are analyzed to identify abnormal fluctuation characteristics. These abnormal fluctuations originate from equipment failures, environmental interference, system overload, and other factors. Abnormal fluctuation characteristics can be identified through methods such as setting thresholds, sliding window analysis, and time series analysis. For example, a sudden temperature change exceeding the expected range or abnormal energy consumption fluctuations are signs of equipment failure or the control system deviating from its normal operating state.

[0082] After identifying abnormal fluctuation characteristics, further analysis is conducted to determine whether these fluctuations deviate from the predetermined control strategy and objectives. If the system's control effect significantly deviates from expectations, it indicates a control deviation. Control deviation can be quantified by calculating the difference between the actual control state and the expected control state, such as the difference between the target temperature and the actual temperature, or the deviation between energy efficiency and the set target.

[0083] By analyzing the incremental data stream and deviation of each control area, areas with anomalies are identified, and P abnormal control areas are marked. These abnormal control areas are areas where there are problems in the system operation and require further troubleshooting and maintenance.

[0084] Power equipment fault location involves a detailed inspection of equipment within the system, such as air conditioners, electrically driven equipment, and fans, to identify the cause of the control deviation. Equipment faults include power supply problems, control signal loss, sensor damage, and hardware failure. By analyzing data from the abnormal control zone, historical operating logs, and equipment health monitoring data, the specific equipment or module causing the fault is located. For the located fault, automated or manual intervention solutions are provided, including equipment restart, replacement of damaged components, and adjustment of control parameters. Once the equipment fault is located and repaired, the abnormal control zone will return to normal operating status.

[0085] Furthermore, it also includes:

[0086] The crisis level is quantified based on the composite deterioration trend of the multiple multi-source spatiotemporal data streams, and multiple regional deterioration indices are output. K high-risk control areas are selected by traversing the multiple regional deterioration indices using a preset dynamic deterioration threshold. Short-term trend prediction is performed on the K multi-source spatiotemporal data streams to obtain K sets of extreme values ​​of operating states. K extreme control strategies are matched according to the K sets of extreme values ​​of operating states to carry out high-response priority and strong intervention control on the K high-risk control areas.

[0087] The compound deterioration trend refers to the overall problem caused by the superposition of different factors. For building control systems, these factors include: a sudden increase in pedestrian density, which may lead to changes in environmental parameters such as temperature and humidity when pedestrian density in certain areas increases sharply, and may even affect the load of air conditioning and ventilation equipment; abnormal temperature rise, which may affect comfort and energy efficiency when the temperature exceeds the normal range, and may even lead to equipment overload or failure; and excessive energy consumption, which indicates that the system load is too high when the energy consumption in certain areas exceeds the set limit, and there is a risk of over-operation.

[0088] Based on multiple multi-source spatiotemporal data streams, time-series analysis of data such as population density, temperature changes, and energy consumption is used to identify the overlapping trends of these three factors, i.e., whether their changes are synchronous or abnormal. A deterioration index quantifies the current crisis level of the controlled area by assigning weights to each factor and then calculating it based on their respective trends. A higher deterioration index indicates that the area is closer to an overload or failure state.

[0089] The dynamic deterioration threshold is defined through historical data analysis or preset rules. Dynamic means that the threshold of the deterioration index will change according to different system operating states. For example, as the system's operating time increases, or the fluctuation of parameters such as temperature and energy consumption increases, the threshold needs to be adjusted appropriately to adapt to the current environment and operating conditions.

[0090] Using a preset dynamic deterioration threshold, the deterioration index of multiple regions is compared with the threshold. Regions with a deterioration index exceeding the threshold are considered high-risk control regions. This method is used to select K high-risk control regions, i.e., those regions with high deterioration indices that require attention.

[0091] This method acquires K multi-source spatiotemporal data streams from K high-risk control areas. Using short-term time-domain prediction models, such as short-term time series forecasting models and LSTM neural networks, the multi-source spatiotemporal data streams of each high-risk area are analyzed to predict its short-term trends. During the prediction process, the model predicts the trend changes in the next few time steps based on the current incremental data streams, such as population density, temperature, and energy consumption. The extreme values ​​of the operating state for each area are calculated, including peak or minimum values ​​for temperature, energy consumption, and equipment load. This approach allows for the early identification of potential extreme operating states, such as excessively high temperatures or energy overload, thus providing proactive adjustment solutions.

[0092] Extreme control strategies are specific emergency control measures generated based on the predicted extreme operating conditions of each region, combined with the system's control capabilities and safety limits. These strategies may include: adjusting the operating parameters of air conditioners or fans to reduce temperature fluctuations; appropriately adjusting the system load to prevent energy consumption from exceeding safety limits; optimizing equipment operation sequences; or switching to standby equipment. The control strategy for each region is customized based on its predicted extreme conditions. For example, if the temperature in a region is predicted to exceed a set range, the air conditioning settings for that region will be automatically adjusted, or the fan output will be increased.

[0093] For high-risk control areas, high-priority control measures are applied, meaning these areas will be prioritized and subject to stricter control measures. Strong intervention controls include immediately adjusting the operating status of the air conditioning system, rapidly switching equipment, or transferring loads between areas. These measures aim to ensure that these areas do not experience further problems due to abnormal operating conditions, such as equipment failure or overheating. The ultimate goal is to ensure the stability and safety of the overall building system.

[0094] Example 2, based on the same inventive concept as the cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method in the aforementioned examples, such as... Figure 2 As shown in the figure, this application provides a cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control system, the system comprising:

[0095] The local deployment module 10 is used to deploy heterogeneous sensor arrays locally according to the zoning topology of the central air conditioning system, constructing multiple edge intelligent sensing nodes corresponding to multiple independent control areas. These edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via an RS485 bus. The coupling prediction module 20 is used by the energy-saving optimization cloud to receive multiple multi-source spatiotemporal data streams from the edge intelligent sensing nodes, perform load and population flow coupling prediction, and generate multiple initial control strategies. The deviation matrix generation module 30 is used to generate a regional energy consumption deviation matrix by comparing multiple control feedback data and the multiple initial control strategies. The deviation matrix correction module 40 is used to correct the regional energy consumption deviation matrix based on multiple thermal coupling coefficients of the multiple independent control areas, obtaining multiple neighborhood compensation strategies. The spatiotemporal collaborative optimization module 50 is used to perform spatiotemporal collaborative optimization of the multiple initial control strategies based on the multiple neighborhood compensation strategies, generating multiple dynamic optimization strategies. The closed-loop control module 60 is used by the energy-saving optimization cloud to distribute the multiple dynamic optimization strategies to the multiple edge intelligent sensing nodes, performing zoning autonomous closed-loop control of the multiple independent control areas.

[0096] Furthermore, the coupling prediction module 20 is used to perform the following operation steps:

[0097] Using the first independent control area as the spatial boundary, a spatiotemporal fusion model of the first multi-source spatiotemporal data stream is performed to generate a first environmental state tensor. The first environmental state tensor is decoupled and extracted to obtain a first regional residence index, a first inter-regional migration intensity, a first air conditioning power change gradient, and a first thermal inertia coefficient. The first regional residence index, the first inter-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient are used as physical information decision factors to perform load-population coupling prediction and generate a first initial control strategy.

[0098] Furthermore, the coupling prediction module 20 is used to perform the following operation steps:

[0099] A pre-constructed load-person flow coupling prediction model is established, comprising a temporal convolutional branch, a graph attention branch, and an attention fusion layer. The temporal convolutional branch and the graph attention branch are connected in parallel, and their outputs are fed into the attention fusion layer. The first air conditioning power change gradient and the first thermal inertia coefficient are input into the temporal convolutional branch, and the dynamic load response vector and thermal inertia time code are extracted via an expanded causal convolutional layer. The first regional residence index and the first cross-regional migration intensity are input into the graph attention branch for spatial neighborhood aggregation, outputting a spatial person flow coupling matrix and a regional thermal potential field. The temporal convolutional branch and the graph attention branch perform feature extraction operations in parallel. The dynamic load response vector, thermal inertia time code, spatial person flow coupling matrix, and regional thermal potential field are weighted and fused in the attention fusion layer to output a predicted load time series curve. Based on the predicted load time series curve, the first initial control strategy is optimized in the rolling time domain.

[0100] Furthermore, the coupling prediction module 20 is used to perform the following operation steps:

[0101] Extract the first pedestrian flow heatmap stream, the first equipment power time series stream, the first ambient temperature and humidity, and the first CO2 concentration data from the first multi-source spatiotemporal data stream; take the first independent control area as the spatial boundary, perform spatiotemporal point cloud rigid registration of the first pedestrian flow heatmap stream and the first equipment power time series stream to generate a first fused spatiotemporal grid; on the first fused spatiotemporal grid, stack and fuse the first ambient temperature and humidity and the first CO2 concentration data through tensor channels to construct the first environmental state tensor.

[0102] Furthermore, the deviation matrix generation module 30 is used to perform the following operation steps:

[0103] After performing topological mapping of the multiple control feedback data and the multiple initial control strategies based on the multiple independent control regions, multiple energy consumption integral deviations are calculated; a zero matrix is ​​constructed based on the multiple independent control regions; the main diagonal elements of the zero matrix are filled using the multiple energy consumption integral deviations; after quantifying the neighborhood energy consumption interference intensity of the multiple independent control regions according to the partition topology, the off-diagonal elements of the zero matrix are filled to complete the construction of the regional energy consumption deviation matrix.

[0104] Furthermore, the deviation matrix correction module 40 is used to perform the following operation steps:

[0105] Based on the partitioned topology, multiple sets of adjacent control regions of the multiple independent control regions are extracted; multiple sets of neighborhood interface thermal conductivity and multiple sets of shared interface area of ​​the multiple sets of adjacent control regions are extracted from the building BIM model; using the multiple sets of adjacent control regions as matrix element coordinates, a symmetric thermal coupling matrix is ​​constructed based on the multiple sets of neighborhood interface thermal conductivity and multiple sets of shared interface area; the symmetric thermal coupling matrix is ​​used to correct the regional energy consumption deviation matrix using the Hadamard product, generating a thermal coupling correction deviation matrix; the thermal coupling correction deviation matrix is ​​decomposed to extract multiple row vector coupling strengths of the multiple independent control regions; the multiple row vector coupling strengths are fitted with gain adjustment to output the multiple neighborhood compensation strategies.

[0106] Furthermore, the deviation matrix correction module 40 is used to perform the following operation steps:

[0107] Multiple region gain coefficients are configured based on the heat capacity of the multiple independent control regions; the overall gain scaling of the coupling strength of the multiple row vectors is performed using the multiple region gain coefficients to output multiple gain adjustment row vectors; scalar aggregation is performed on the multiple gain adjustment row vectors to generate multiple compensation reference quantities; dual-thread decoupling transformation is performed on the multiple compensation reference quantities to output multiple temperature compensation strategies and multiple fan compensation strategies, which constitute the multiple neighborhood compensation strategies.

[0108] Furthermore, the closed-loop control module 60 is used to perform the following operation steps:

[0109] During the execution of the multiple edge intelligent sensing nodes and the multiple dynamic optimization strategies, the nodes simultaneously perform real-time dynamic operation status perception of the multiple independent control areas, and obtain multiple multi-source incremental data streams. Based on the abnormal fluctuation characteristics of the multiple multi-source incremental data streams, the nodes identify control deviations in the multiple independent control areas and locate P abnormal control areas. The nodes then perform power equipment fault location and maintenance on the P abnormal control areas.

[0110] Furthermore, the coupling prediction module 20 is used to perform the following operation steps:

[0111] The crisis level is quantified based on the composite deterioration trend of the multiple multi-source spatiotemporal data streams, and multiple regional deterioration indices are output. K high-risk control areas are selected by traversing the multiple regional deterioration indices using a preset dynamic deterioration threshold. Short-term trend prediction is performed on the K multi-source spatiotemporal data streams to obtain K sets of extreme values ​​of operating states. K extreme control strategies are matched according to the K sets of extreme values ​​of operating states to carry out high-response priority and strong intervention control on the K high-risk control areas.

[0112] Through the foregoing detailed description of the cloud-based collaborative dynamic power consumption perception and energy-saving optimization control method, those skilled in the art can clearly understand the cloud-based collaborative dynamic power consumption perception and energy-saving optimization control system in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0113] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A cloud-based collaborative method for dynamic sensing and energy-saving optimization control of power consumption, characterized in that: The method includes: Based on the zoning topology of the central air conditioning system, a heterogeneous sensor array is deployed locally to construct multiple edge intelligent sensing nodes corresponding to multiple independent control areas. The multiple edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via an RS485 bus. The energy-saving optimization cloud receives multiple multi-source spatiotemporal data streams transmitted back from multiple edge intelligent sensing nodes, performs load and population flow coupling prediction, and generates multiple initial control strategies. By comparing multiple control feedback data and the multiple initial control strategies, a regional energy consumption deviation matrix is ​​generated. Based on the multiple thermal coupling coefficients of the multiple independent control regions, the regional energy consumption deviation matrix is ​​corrected to obtain multiple neighborhood compensation strategies. Based on the multiple neighborhood compensation strategies, the multiple initial control strategies are spatiotemporally coordinated to generate multiple dynamic optimization strategies. The energy-saving optimization cloud sends the multiple dynamic optimization strategies to the multiple edge intelligent sensing nodes to perform partitioned autonomous closed-loop control of the multiple independent control areas; By comparing multiple control feedback data and the multiple initial control strategies, a regional energy consumption deviation matrix is ​​generated, including: After performing topological mapping of the multiple control feedback data and the multiple initial control strategies based on the multiple independent control regions, multiple energy consumption integral deviations are calculated. A zero matrix is ​​constructed based on the multiple independent control regions; The main diagonal elements of the zero matrix are filled using the multiple energy consumption integral deviations. After quantifying the neighborhood energy consumption interference intensity of the multiple independent control regions according to the partition topology, the off-diagonal elements of the zero matrix are filled to complete the construction of the regional energy consumption deviation matrix. Based on the multiple thermal coupling coefficients of the multiple independent control regions, the regional energy consumption deviation matrix is ​​corrected to obtain multiple neighborhood compensation strategies, including: Based on the partition topology, multiple sets of adjacent control regions of the multiple independent control regions are extracted; Extract the thermal conductivity of multiple neighboring interfaces and the area of ​​multiple shared interfaces of the multiple sets of adjacent control areas from the building BIM model; Using the multiple sets of adjacent control regions as matrix element coordinates, a symmetric thermal coupling matrix is ​​constructed based on the multiple sets of neighborhood interface thermal conductivity and multiple sets of shared interface area. The symmetric thermal coupling matrix is ​​used to correct the Hadamard product of the regional energy consumption deviation matrix, thereby generating a thermal coupling correction deviation matrix; Decompose the thermal coupling correction deviation matrix to extract the coupling strength of multiple row vectors in the multiple independent control regions; Gain adjustment fitting is performed on the coupling strength of the multiple row vectors to output the multiple neighborhood compensation strategies; Gain adjustment fitting is performed on the coupling strength of the multiple row vectors to output the multiple neighborhood compensation strategies, including: Multiple region gain coefficients are configured based on the heat capacity of the multiple independent control regions; The overall gain scaling of the coupling strength of the multiple row vectors is performed using the multiple region gain coefficients, and multiple gain-adjusted row vectors are output. Scalar aggregation is performed on the multiple gain adjustment row vectors to generate multiple compensation reference quantities; The multiple compensation reference quantities are decoupled and transformed in a dual-thread manner to output multiple temperature compensation strategies and multiple fan compensation strategies, which constitute the multiple neighborhood compensation strategies.

2. The cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method as described in claim 1, characterized in that, The energy-saving optimization cloud receives multiple multi-source spatiotemporal data streams transmitted back from multiple edge intelligent sensing nodes, performs load-population coupling prediction, and generates multiple initial control strategies, including: Using the first independent control region as the spatial boundary, a spatiotemporal fusion model of the first multi-source spatiotemporal data stream is performed to generate the first environmental state tensor; Spatiotemporal features are decoupled and extracted from the first environmental state tensor to obtain the first region residence index, the first cross-region migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient. Using the first regional residence index, the first cross-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient as physical information decision factors, load-human flow coupling prediction is performed to generate the first initial control strategy.

3. The cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method as described in claim 2, characterized in that, Using the first regional residence index, the first inter-regional migration intensity, the first air conditioning power change gradient, and the first thermal inertia coefficient as physical information decision factors, load-population coupling prediction is performed to generate a first initial control strategy, including: A pre-constructed load-passenger flow coupling prediction model is provided, wherein the load-passenger flow coupling prediction model includes a temporal convolutional branch, a graph attention branch, and an attention fusion layer. The temporal convolutional branch and the graph attention branch are connected in parallel, and their outputs are fed into the attention fusion layer. The first air conditioner power change gradient and the first thermal inertia coefficient are input into the time convolution branch, and the dynamic load response vector and thermal inertia time code are extracted through the dilated causal convolution layer. The first region residence index and the first cross-region migration intensity are input into the graph attention branch for spatial neighborhood aggregation, and the spatial human flow coupling matrix and regional thermal potential field are output. The temporal convolution branch and the graph attention branch perform feature extraction operations in parallel. The dynamic load response vector, thermal inertia time encoding, spatial pedestrian coupling matrix, and regional thermal potential field are weighted and fused in the attention fusion layer to output the predicted load time series curve. Based on the predicted load time-series curve, the rolling time-domain optimization output of the first initial control strategy is performed.

4. The cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method as described in claim 2, characterized in that, Using the first independent control region as the spatial boundary, a spatiotemporal fusion model of the first multi-source spatiotemporal data stream is performed to generate a first environmental state tensor, including: Extract the first human flow heat map stream, the first equipment power time series stream, the first ambient temperature and humidity, and the first CO2 concentration data from the first multi-source spatiotemporal data stream; Using the first independent control area as the spatial boundary, perform spatiotemporal point cloud rigid registration of the first pedestrian flow heat map flow and the first equipment power time series flow to generate the first fused spatiotemporal grid; On the first fused spatiotemporal grid, the first environmental temperature and humidity data and the first CO2 concentration data are fused by stacking tensor channels to construct the first environmental state tensor.

5. The cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method as described in claim 1, characterized in that, Also includes: During the execution of the multiple edge intelligent sensing nodes, the multiple dynamic optimization strategies are executed simultaneously, and the real-time dynamic operating status of the multiple independent control areas is perceived, resulting in multiple multi-source incremental data streams. Based on the abnormal fluctuation characteristics of the multiple multi-source incremental data streams, control deviation identification is performed on the multiple independent control regions to locate P abnormal control regions; Perform power equipment fault location and maintenance on the P abnormal control zones.

6. The cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method as described in claim 1, characterized in that, Also includes: The crisis level is quantified based on the combined deterioration trend of the multiple multi-source spatiotemporal data streams, and multiple regional deterioration indices are output. K high-risk control areas are selected by iterating through the deterioration indices of the multiple regions using a preset dynamic deterioration threshold. Short-time-domain trend prediction is performed on K multi-source spatiotemporal data streams to obtain K sets of extreme values ​​of the operating state; Based on the K sets of extreme operating states, K extreme control strategies are matched to perform high-response priority and strong intervention control on the K high-risk control areas.

7. A cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control system, characterized in that, The system is used to implement the cloud-based collaborative dynamic power consumption sensing and energy-saving optimization control method according to any one of claims 1-6, the system comprising: The local deployment module is used to deploy heterogeneous sensor arrays locally according to the partition topology of the central air conditioning system, and to build multiple edge intelligent sensing nodes corresponding to multiple independent control areas. The multiple edge intelligent sensing nodes and the energy-saving optimization cloud are bidirectionally connected via RS485 bus. The coupling prediction module is used to receive multiple multi-source spatiotemporal data streams transmitted back from multiple edge intelligent sensing nodes in the energy-saving optimization cloud, perform load and population flow coupling prediction, and generate multiple initial control strategies. The deviation matrix generation module is used to generate a regional energy consumption deviation matrix by comparing multiple control feedback data and the multiple initial control strategies. The deviation matrix correction module is used to correct the regional energy consumption deviation matrix based on the multiple thermal coupling coefficients of the multiple independent control regions, thereby obtaining multiple neighborhood compensation strategies. The spatiotemporal collaborative optimization module is used to perform spatiotemporal collaborative optimization of the multiple initial control strategies based on the multiple neighborhood compensation strategies, and generate multiple dynamic optimization strategies. The closed-loop control module is used to send the multiple dynamic optimization strategies from the energy-saving optimization cloud to the multiple edge intelligent sensing nodes, and to perform partitioned autonomous closed-loop control of the multiple independent control areas.

Images