Intelligent building air conditioning group control energy-saving system and method based on multi-source data fusion

Abstract

The present invention relates to the field of air-conditioning control technology, and discloses an intelligent building air-conditioning group control energy-saving system and method based on multi-source data fusion. The method comprises collecting heterogeneous trajectories of personnel activities in a building, constructing a spatiotemporal energy-demand coupling map after cross-modal coupling and abnormal outlier suppression processing; then, based on the map, an optimized air-delay compensation factor β' is obtained to divide energy-demand-significant areas into energy-demand-insignificant areas; finally, a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated based on the regional division results and the optimized air-delay compensation factor; the method can accurately control the air-conditioning, significantly improve the energy utilization efficiency of the intelligent building air-conditioning system, and reduce energy consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning control, and more particularly to an intelligent building air conditioning group control energy-saving system and method based on multi-source data fusion. Background Art

[0002] With the development of science and technology, the air-conditioning system in smart buildings can improve indoor comfort and achieve energy-saving goals. Under the current technical environment, although the existing air-conditioning group control technology has made certain progress, there are still many areas that need to be improved.

[0003] Among the existing related technologies, Chinese patent application publication number CN118654372A discloses a method and system for intelligent group control of central air conditioning. This prior art collects comprehensive indoor and outdoor environmental information, building and user information in real time, analyzes the impact of crowd density, regional characteristics, and temperature differences in the target area on comfort, and assesses user comfort to adjust the air conditioning. It also analyzes the impact of user age range and indoor environment on cooling capacity demand to calculate actual demand and adjust the air conditioning. However, this method primarily relies on comprehensive information and simple analysis of influencing factors to control the air conditioning, and lacks comprehensive and in-depth consideration of complex and changing human activity patterns. In actual smart building scenarios, human activity patterns vary significantly across different areas. For example, activity patterns in office areas and conference rooms are distinct. This prior art fails to fully explore the complex spatiotemporal patterns underlying these differences, making it difficult to accurately adapt to the actual needs of different areas, potentially leading to energy waste and poor comfort. For example, when a meeting is added to a conference room and people gather quickly, using existing analysis methods alone may not be able to adjust the air conditioning capacity in a timely and accurate manner, affecting human comfort and causing unreasonable energy consumption.

[0004] Chinese patent application publication number CN118499910A discloses a method for controlling an intelligent building air conditioning system. This method proposes providing HVAC control presets to a calibrated digital building model based on a preset control algorithm. This method then simulates and feeds back temperature and humidity data based on local weather information, modifying the HVAC setpoint schedule according to thermal comfort standards to control air conditioning operation. However, this method also has drawbacks. When faced with complex occupant activity patterns, it primarily relies on the building digital model and weather information, and lacks integration of real-time dynamic data on occupant activity. In intelligent buildings, occupant activity is a key factor affecting air conditioning energy consumption and comfort. This existing technology lacks the integration and utilization of occupant activity data from multiple sources, such as access control systems and conference room reservation systems. This makes it difficult to accurately grasp real-time changes in occupant activity across different areas, hindering precise air conditioning control to meet actual needs. For example, if office workers leave work early or late, or if meeting times in conference rooms are temporarily changed, the system may be unable to make timely and appropriate air conditioning adjustments, resulting in energy waste and reduced indoor comfort.

[0005] When dealing with the complex and diverse human activity patterns in smart buildings, existing technologies have problems such as insufficient in-depth mining of human activity data and insufficient fusion of multi-source data. This makes it difficult to accurately adapt to the air-conditioning needs of different areas, and it is impossible to achieve efficient energy saving while ensuring comfort. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent building air conditioning group control energy-saving system and method based on multi-source data fusion. By collecting the heterogeneous trajectories of people's activities in the building, a spatiotemporal energy-demand coupling map is constructed through multi-step processing. Based on the map, key factors are obtained and regions are divided. Finally, a dynamic energy-saving control strategy is formulated to accurately grasp the relationship between people's activities in the building and energy demand, reasonably allocate air conditioning resources, and improve energy utilization efficiency and indoor comfort.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The energy-saving method of intelligent building air conditioning group control based on multi-source data fusion includes:

[0009] Heterogeneous trajectories reflecting people's activities in the building are collected, and cross-modal coupling and abnormal outlier suppression are performed on the heterogeneous trajectories to obtain the heterogeneous trajectories after outlier suppression and the observation value weight matrix. Based on the heterogeneous trajectories after outlier suppression and the observation value weight matrix, a spatiotemporal energy demand coupling map is constructed.

[0010] Based on the spatiotemporal energy demand coupling map, the optimized airlock compensation factor β' is obtained; the spatiotemporal energy demand coupling map is clustered at multiple scales to extract energy demand-significant regions and non-energy demand-significant regions; the energy demand-significant regions are divided into steady-state cores and transient cores, and the non-energy demand-significant regions are marked as dormant cores;

[0011] According to the division results of steady-state core, transient core and dormant core, combined with the optimization of the air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated.

[0012] Furthermore, the collection of heterogeneous trajectories reflecting the activities of people in the building includes:

[0013] Collect the dynamic time series of personnel entering and leaving key nodes to generate time series displacement trajectories;

[0014] Scan the target area with near-infrared wavelengths to capture the spatial distribution characteristics of human radiation energy levels, construct a thermodynamic energy gradient field that reflects the degree of human gathering, and use a gradient clustering algorithm to calibrate the hot spots where people gather to generate thermal distribution trajectories.

[0015] Extract the agenda multi-dimensional attribute parameters of the agenda event and generate the agenda association trajectory;

[0016] The temporal displacement trajectory, thermal distribution trajectory and agenda association trajectory are combined to form heterogeneous trajectories.

[0017] Furthermore, the cross-modal coupling and abnormal outlier suppression processing of heterogeneous trajectories includes:

[0018] Apply sliding window Fourier decomposition to the time series displacement trajectory to extract the periodic fundamental frequency component and non-stationary residual component;

[0019] Map the thermal distribution trajectory to a three-dimensional gridded energy field and calculate the local thermal entropy growth rate and thermal core diffusion coefficient;

[0020] Using the non-stationary residual component as the observation sequence, identifying the mutation points in the observation sequence, and generating the smoothed and corrected non-stationary residual component and observation value weight matrix;

[0021] The periodic fundamental frequency component, the non-stationary residual component after smoothing correction, the local thermal entropy growth rate and the thermal core diffusion coefficient are combined to form a heterogeneous trajectory after outlier suppression.

[0022] Furthermore, identifying the mutation point in the observation sequence includes:

[0023] Calculate the change rate of the population density at each moment in the observation sequence and generate a change rate sequence;

[0024] Normalizing the change rate sequence to obtain a normalized change rate sequence;

[0025] A mutation point identification threshold ξ is set, and the moment when the normalized change rate exceeds ξ is marked as a mutation point to generate a mutation point sequence.

[0026] Furthermore, the identifying mutation points in the observation sequence also includes: using the agenda association trajectory as a latent variable, constructing a causal reasoning network according to the observation sequence and the latent variable; and using the causal reasoning network to identify the mutation points in the observation sequence.

[0027] Furthermore, generating the smoothed and corrected non-stationary residual component includes:

[0028] Correlation analysis is performed based on the agenda correlation trajectory to determine whether the mutation point is a normal mutation point or an abnormal mutation point; the corresponding value of the abnormal mutation point in the observation sequence is marked as an outlier; and the data smoothing method is used to obtain the non-stationary residual component after smoothing correction.

[0029] Furthermore, determining whether the mutation point is a normal mutation point or an abnormal mutation point includes:

[0030] Extract the timestamp of each mutation point in the mutation point sequence , and retrieve the timestamp in the agenda associated track Agenda events within the nearby Δt' time range;

[0031] If there is an agenda event within the range of Δt', then the correlation analysis is performed between the agenda multidimensional attribute parameters of the agenda event and the personnel density change rate of the mutation point, and the correlation coefficient ρ' is calculated;

[0032] If the correlation coefficient ρ' exceeds the preset correlation threshold ρ0, the mutation point is marked as a normal mutation point, otherwise it is marked as an abnormal mutation point.

[0033] Furthermore, generating the observation value weight matrix includes: calculating the observation value weight coefficient ω according to the correlation coefficient ρ' of the normal mutation point, and constructing the observation value weight matrix.

[0034] Furthermore, constructing the spatiotemporal energy demand coupling map includes:

[0035] The heterogeneous trajectories that have undergone outlier suppression processing are subjected to a tensor product operation to generate a space-time energy demand coupling tensor; the space-time energy demand coupling tensor is subjected to dimensionality reduction projection to form a two-dimensional space-time energy demand coupling map that matches the building floor plan; and the pixels in the two-dimensional space-time energy demand coupling map are assigned credibility attributes according to the observation value weight matrix to obtain the space-time energy demand coupling map.

[0036] Furthermore, obtaining the optimized hovering compensation factor β' includes:

[0037] Identifying spatiotemporal energy requires that the credibility attribute in the coupled map is lower than the preset confidence threshold The low-confidence areas of the spatiotemporal energy-demand coupling map are marked; the air conditioning operating parameter sequences of the same type and historical period are extracted to calculate the ideal temperature control response curve; the deviation between the measured temperature response in the current low-confidence area and the ideal temperature control response curve is compared and quantified as the temperature control hysteresis ΔT; the adversarial generative network (ACGAN) is constructed based on the temperature control hysteresis ΔT, and the generator G in the adversarial generative network (ACGAN) learns the probability distribution of the air stasis compensation factor β; and the optimized air stasis compensation factor β' is obtained by sampling from the probability distribution of β.

[0038] Furthermore, the extraction of energy-demand-significant regions and energy-insignificant regions includes:

[0039] Local binary pattern encoding is applied to the spatiotemporal energy-demand coupling map to extract the energy-demand gradient features and generate an LBP feature map. Morphological filtering is performed on the LBP feature map. Using the morphologically filtered LBP feature map as input, an adaptive spectral clustering algorithm is used to extract energy-demand-significant regions and non-energy-demand-significant regions.

[0040] Furthermore, dividing the energy demand significance region into a steady-state core and a transient core includes:

[0041] Extract the energy demand time series of the significant energy demand area;

[0042] Based on the periodic fundamental frequency component, the energy demand time series in the energy demand significant area is analyzed to obtain the energy demand autocorrelation coefficient. and decay time constant τ; where, Indicates that the time series can be The value of Indicates that the time series can be The value of It represents the autocorrelation coefficient of the energy time series at time t and time t+Δt, where Δt represents the time lag interval;

[0043] Setting the autocorrelation threshold , decay time threshold And the required intensity threshold E0, according to ,τ, and , the energy demand significance region is divided into steady-state core and transient core.

[0044] An intelligent building air conditioning group control energy-saving system based on multi-source data fusion is used to implement the above-mentioned intelligent building air conditioning group control energy-saving method based on multi-source data fusion. The system includes:

[0045] Energy-demand coupling map construction module: This module is used to collect heterogeneous trajectories reflecting people's activities in the building, perform cross-modal coupling and abnormal outlier suppression on the heterogeneous trajectories, and obtain the heterogeneous trajectories and observation weight matrix after outlier suppression. Based on the heterogeneous trajectories and observation weight matrix after outlier suppression, a spatiotemporal energy-demand coupling map is constructed.

[0046] Region division module: Based on the spatiotemporal energy demand coupling map, the optimized airlock compensation factor β' is obtained; the spatiotemporal energy demand coupling map is clustered at multiple scales to extract energy demand-significant regions and non-energy demand-significant regions; the energy demand-significant regions are divided into steady-state cores and transient cores, and the non-energy demand-significant regions are marked as dormant cores;

[0047] Energy-saving control strategy formulation module: Based on the division results of steady-state cores, transient cores and dormant cores, combined with the optimization of the air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention collects heterogeneous trajectories of personnel activities, constructs a spatiotemporal energy demand coupling map through multi-step processing, obtains key factors based on the map and divides the regions, and finally formulates a dynamic energy-saving control strategy. This overall solution can accurately grasp the relationship between personnel activities and energy demand in the building, and rationally allocate air-conditioning resources based on the energy demand characteristics of different regions, such as steady-state core, transient core, and dormant core areas, to avoid energy waste. At the same time, the introduction of the optimized air stagnation compensation factor β' can effectively compensate for the lag problem of air-conditioning temperature control, allowing the air-conditioning system to respond to changes in energy demand more quickly and accurately. Overall, it significantly improves the energy utilization efficiency of the intelligent building air-conditioning system, reduces energy consumption, and provides an effective way to achieve building energy conservation and improve indoor comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 The figure is a flow chart showing the principle of the energy-saving method for group-controlled air conditioning in intelligent buildings based on multi-source data fusion in the present invention;

[0052] Figure 2 This is a flow chart of a method for collecting heterogeneous trajectories in the intelligent building air conditioning group control energy-saving method based on multi-source data fusion of the present invention;

[0053] Figure 3 Schematic diagram of the principle of a specific embodiment of the present invention Figure 1 ;

[0054] Figure 4 Schematic diagram of the principle of a specific embodiment of the present invention Figure 2 ;

[0055] Figure 5 This is a flow chart of a method for performing cross-modal coupling and abnormal outlier suppression processing on heterogeneous trajectories in the intelligent building air conditioning group control energy-saving method based on multi-source data fusion of the present invention;

[0056] Figure 6 Flowchart of a method for obtaining an optimized dead-end compensation factor β' in the intelligent building air conditioning group control energy-saving method based on multi-source data fusion of the present invention;

[0057] Figure 7 A flowchart of a method for extracting significant energy demand areas and non-significant energy demand areas, and extracting energy demand time series of significant energy demand areas in the intelligent building air conditioning group control energy-saving method based on multi-source data fusion of the present invention;

[0058] Figure 8 This is a functional module diagram of the intelligent building air conditioning group control energy-saving system based on multi-source data fusion in the present invention.

[0059] Figure numerals: 1. conference room; 2. air conditioning; 3. personnel. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] Example 1

[0062] See also Figure 1 As shown, this embodiment provides an intelligent building air conditioning group control energy-saving method based on multi-source data fusion, including:

[0063] Step S1000: Collect heterogeneous trajectories, perform cross-modal coupling and outlier suppression on the heterogeneous trajectories, and obtain the heterogeneous trajectories and observation weight matrices after outlier suppression. Construct a spatiotemporal energy-demand coupling map based on the heterogeneous trajectories and observation weight matrices after outlier suppression.

[0064] Furthermore, step S1000 includes:

[0065] Step S1100, collecting heterogeneous trajectories;

[0066] Furthermore, if Figure 2 As shown, step S1100 includes:

[0067] Step S1110 , collecting the dynamic time series chain of personnel entering and leaving key nodes to generate a time series displacement trajectory;

[0068] Step S1120: Scan the target area with near-infrared bands to capture the spatial distribution characteristics of human radiation energy levels, construct a thermodynamic energy gradient field reflecting the degree of human gathering, and calibrate the hot spots of human gathering through a gradient clustering algorithm to generate a thermal distribution trajectory;

[0069] Step S1130 , extracting agenda multi-dimensional attribute parameters of the agenda event and generating an agenda association trajectory;

[0070] In step S1140 , the temporal displacement trajectory, the thermal distribution trajectory, and the agenda association trajectory are combined to form a heterogeneous trajectory.

[0071] Specifically, step S1110 collects dynamic time-series chains of personnel entering and exiting key nodes, generating time-series displacement trajectories. In smart buildings, key nodes generally refer to locations such as entrances and exits to each floor and entrances to major functional areas (such as conference rooms, offices, and rest areas). These locations are important channels for personnel flow. The building's access control system is activated. This system typically consists of an access controller, card reader, and electronic door locks. Embedded encrypted beacons play a key role in this process. These beacons, devices that emit specific encrypted signals, accurately record the time information of personnel entering and exiting key nodes, thereby forming a dynamic time-series chain. For example, in a large office building, during rush hour, employees swipe their cards to enter the lobby on the first floor. The encrypted beacons in the access control system record the employee's identity information and entry time. This information, aggregated from multiple employees, forms a dynamic time-series chain. The time-varying curves generated from these dynamic time-series chains, or time-series displacement trajectories, can reflect the distribution of personnel flow within the building in real time. This provides a precise basis for subsequently determining the air conditioning energy demand of different areas. Since the fluctuation of population density in different areas will directly affect the air conditioning energy demand of the area, such as Figure 3-Figure 4 As shown, in crowded conference rooms, people dissipate more heat, requiring greater cooling capacity from the air conditioner. In less crowded conference rooms, people dissipate less heat, requiring less cooling capacity from the air conditioner. By obtaining time-series displacement trajectories, we can predict energy demand changes in each area based on the flow of people, allowing us to adjust the operating parameters of the air conditioning system appropriately to avoid energy waste. For example, if we predict that a conference room on a certain floor will be crowded over the coming period, we can increase the cooling capacity of the air conditioner in that area in advance to ensure indoor comfort. This also avoids excessive energy consumption caused by adjusting the air conditioner after people arrive, thus achieving energy savings.

[0072] Step S1120 scans the target area in the near-infrared band to capture the spatial distribution characteristics of human radiation energy levels, construct a thermodynamic energy gradient field reflecting the degree of human gathering, and use a gradient clustering algorithm to identify cluster hotspots, generating a thermal distribution trajectory. The infrared topological imaging unit deployed within the building area is a device with infrared detection and imaging capabilities, capable of scanning the target area in the near-infrared band. The human body continuously radiates infrared radiation, the radiation energy level of which is related to factors such as body temperature and activity level. When scanning the target area, the infrared topological imaging unit captures the infrared radiation emitted by the human body and converts it into electrical or digital signals for processing. By analyzing these signals, a thermodynamic energy gradient field reflecting the degree of human gathering can be constructed. In this gradient field, areas with larger energy gradients generally indicate higher concentrations of human beings. The gradient clustering algorithm is a data processing algorithm that can classify data into clusters based on the similarity between data points. In this step, this algorithm can be used to identify cluster hotspots, i.e., areas with high concentrations of human beings. The generated thermal distribution trajectory can reflect the intensity and aggregation patterns of human activity in real time. This has the added benefit of providing a decision-making basis for the subsequent generation of personalized air conditioning control strategies. This is because different activity intensities and gathering patterns create different air conditioning requirements. For example, in a gym, where activity is intense, people dissipate heat quickly, requiring stronger cooling. In contrast, in a library, where people gather in quieter, more concentrated areas, cooling requirements are relatively low. Based on the heat distribution trajectory, personalized air conditioning control strategies can be developed for the specific occupants in different areas, improving the energy efficiency and comfort of the air conditioning system.

[0073] Step S1130 extracts the multidimensional attribute parameters of the agenda event and generates an agenda-related trajectory. Semantic parsing algorithms in deep learning use neural networks to understand and analyze text data. In conference room agenda management systems, there are numerous structured agenda tags, such as meeting start time, end time, number of participants, and agenda type (urgent / routine). By mining these agenda tags with semantic parsing algorithms, multidimensional attribute parameters such as duration, number of participants, and agenda type can be extracted. For example, in a company's conference room management system, a meeting's agenda tag is recorded as "9:00-11:00, department meeting, expected number of participants: 30." Using semantic parsing algorithms, it is possible to extract information such as the meeting duration of 2 hours and the number of participants. These parameters are used to generate a preload vector for the energy needs of attendees, known as the agenda-related trajectory. The agenda-related trajectory uses agenda information to predict the gathering behavior of participants over a period of time. This advantageously enables the air conditioning system to proactively detect changes in personnel activity, improving air conditioning response speed and energy efficiency. For example, if a company's daily meetings are known through agenda-linked tracking to be a meeting of 30 people in a certain conference room in half an hour, the air conditioning system can be activated in advance and adjusted to the appropriate temperature and air speed. This allows everyone to immediately enjoy a comfortable environment upon entering the room, avoiding energy waste and reduced comfort caused by temporary air conditioning adjustments. Furthermore, by adjusting air conditioning settings in advance based on agenda information before the meeting ends, unnecessary energy consumption is reduced and energy efficiency is improved.

[0074] In step S1140, the temporal displacement trajectory, thermal distribution trajectory, and agenda-related trajectory are combined to form heterogeneous trajectories. These three trajectories reflect the activities of people within the building from different perspectives. The temporal displacement trajectory focuses on the flow paths and temporal sequence of people, the thermal distribution trajectory focuses on the degree of gathering and activity intensity of people, and the agenda-related trajectory uses meeting agenda information to predict people's gathering behavior. They complement each other and together form heterogeneous trajectories. For example, in a comprehensive office building, the temporal displacement trajectory shows that people on a certain floor gradually gather in a certain area, the thermal distribution trajectory indicates that the degree of gathering and activity intensity in this area are high, and the agenda-related trajectory indicates that a large meeting is about to be held in this area. The combination of these three trajectories can more comprehensively and accurately reflect the activities of people in this area. The beneficial effect is that it provides a rich data foundation for the subsequent cross-modal coupling and outlier suppression processing of heterogeneous trajectories, thereby improving the accuracy and reliability of the energy-saving method of intelligent building air conditioning group control. Through comprehensive analysis of these three trajectories, we can more accurately judge the air-conditioning energy demand in different areas, provide stronger support for constructing a spatiotemporal energy-demand coupling map and formulating dynamic energy-saving control strategies, and ultimately achieve more efficient air-conditioning group control energy saving, reduce building energy consumption, and improve the level of intelligent building management.

[0075] Step S1200 , performing cross-modal coupling and abnormal outlier suppression processing on the heterogeneous trajectories to obtain heterogeneous trajectories and observation value weight matrices after outlier suppression processing;

[0076] Furthermore, if Figure 5 As shown, step S1200 includes:

[0077] Step S1210, applying sliding window Fourier decomposition to the time series displacement trajectory to extract the periodic fundamental frequency component and the non-stationary residual component;

[0078] Specifically, sliding window Fourier decomposition is a signal processing technique. A sliding window is a continuous data segment of a fixed length selected from time series data, which is then moved point by point along the time series displacement trajectory at a fixed step size. For example, suppose the time series displacement trajectory is a series of occupancy density data at a certain time point. A sliding window of 10 time points is selected, starting from the first time point and sliding backwards one time point at a time. Fourier decomposition, based on the Fourier transform principle, decomposes a time domain signal into a superposition of sine and cosine functions of different frequencies. By performing sliding window Fourier decomposition on the time series displacement trajectory, periodic fundamental frequency components and non-stationary residual components can be extracted. Periodic fundamental frequency components reflect long-term trends in occupancy density. This is because there are periodic patterns in the flow of people, such as the patterns of people entering and leaving a building during similar time periods on weekdays. These patterns manifest as periodic variations in the time series displacement trajectory, and the periodic fundamental frequency components extracted through Fourier decomposition can capture these patterns. For example, in an office building on weekdays, there's a peak in people entering around 9:00 AM and a peak in people leaving around 5:00 PM. The periodic fundamental frequency component can reflect this recurring daily trend in population density. The non-stationary residual component reflects short-term fluctuations in population density. Unexpected events can occur within a building, such as impromptu meetings that lead to short gatherings of people. These changes in population density manifest as short-term fluctuations in the time series displacement trajectory, reflected by the non-stationary residual component.

[0079] This step has significant benefits. From the perspective of energy conservation in air conditioning cluster control, the extracted periodic fundamental frequency component and non-stationary residual component provide more targeted input features for subsequent cross-modal fusion. Understanding the long-term trends and short-term fluctuations in occupancy density is crucial for air conditioning system control. The periodic fundamental frequency component enables the approximate changes in occupancy density over a period of time to be predicted, allowing for pre-emptive adjustments to the air conditioning system's operating mode. For example, before the daily peak in occupancy, the cooling or heating power of the air conditioner can be increased to ensure indoor comfort while avoiding unnecessary energy consumption. The non-stationary residual component helps capture sudden changes in occupancy density, enabling the air conditioning system to respond quickly. When a short-term increase in occupancy density is detected, the cooling capacity of the air conditioner is rapidly increased to meet occupant comfort needs, avoiding energy waste and a decrease in comfort. The extraction of these two components enhances understanding and control of occupancy density fluctuations, providing strong support for the subsequent development of more precise air conditioning control strategies, and contributing to achieving energy conservation goals for intelligent building air conditioning cluster control.

[0080] Step S1220 , mapping the thermal distribution trajectory to a three-dimensional gridded energy field, and calculating the local thermal entropy growth rate and thermal core diffusion coefficient;

[0081] Specifically, step S1220 mainly maps the thermal distribution trajectory to a three-dimensional gridded energy field and calculates the local thermal entropy growth rate and thermal nuclear diffusion coefficient. Its purpose is to more accurately describe the impact of personnel gathering on the temperature of the local area and provide more detailed constraints for subsequent energy demand analysis. First, the thermal distribution trajectory is mapped to a three-dimensional spatial grid. Three-dimensional gridding is to divide a continuous space into small cubic units (i.e., grids), just like dividing a large space into many small grids. The thermal distribution trajectory contains the spatial distribution information of the radiation energy level of the personnel. By mapping it to the three-dimensional grid, each grid can correspond to a specific spatial location, thereby constructing a digital twin of the energy distribution that matches the physical structure of the building. This digital twin is a digital simulation of the energy distribution of a real building space. It can more intuitively show the relationship between personnel activities and energy distribution.

[0082] Then calculate the local thermal entropy growth rate and heat diffusion coefficient of each grid cell. Thermal entropy is a thermodynamic concept that represents the degree of disorder in a system. In this scenario, the local thermal entropy growth rate reflects the speed of change in the degree of disorder in the heat distribution within the grid cell due to factors such as the gathering of people. When calculating the local thermal entropy growth rate, according to relevant thermodynamic theories, for a grid cell, the change in its thermal entropy is related to factors such as the inflow and outflow of heat and temperature. Assuming that in a certain period of time, the amount of heat absorbed by the grid cell is Q and the temperature is T, the change in thermal entropy ΔS can be approximately expressed as ΔS= (This is based on simple thermodynamic principles; actual calculations may be more complex and require consideration of additional factors.) The local thermal entropy growth rate is simply the ratio of the change in thermal entropy to time. The thermal core diffusion coefficient describes the ability of heat to diffuse between grid cells. It is related to factors such as the thermal conductivity of the material and the spatial structure. In a building environment, the thermal conductivity of materials varies in different areas, and the concentration of people also affects heat diffusion. By calculating the thermal core diffusion coefficient, this heat diffusion capacity can be quantified.

[0083] The beneficial effects of this step are reflected in multiple aspects. From the perspective of energy demand analysis, the local thermal entropy growth rate and the thermal nuclear diffusion coefficient, as key physical quantities that describe the impact of human concentration on local area temperature, provide spatially fine-grained constraints for subsequent energy demand analysis. When determining the cooling or heating needs of the air-conditioning system, it is no longer necessary to rely solely on macroscopic occupant density information. Instead, these physical quantities for each grid cell can be used to more accurately calculate the energy needs of different areas. For example, in areas with dense populations and high local thermal entropy growth rates and large thermal nuclear diffusion coefficients, heat accumulates and diffuses quickly, requiring the air-conditioning system to require greater cooling power to maintain a comfortable temperature. Through these precise calculations, energy waste can be avoided in areas that do not require large amounts of cooling or heating, thereby improving energy efficiency. From the perspective of air-conditioning group control strategy formulation, these physical quantities can help more rationally allocate air-conditioning resources. According to the energy demand characteristics of different areas, the operating parameters of the air-conditioning can be adjusted in a targeted manner to achieve more precise temperature control and improve the comfort of indoor occupants.

[0084] Step S1230 uses the non-stationary residual component as the observation sequence, identifies the mutation point in the observation sequence, and generates a smoothed and corrected non-stationary residual component and an observation value weight matrix;

[0085] Furthermore, step S1230 includes:

[0086] Step S1231, using the non-stationary residual component as the observation sequence, identifying the mutation point in the observation sequence;

[0087] Furthermore, step S1231 includes:

[0088] Step S12311, calculating the change rate of the personnel density at each moment in the observation sequence to generate a change rate sequence;

[0089] Step S12312, normalizing the change rate sequence to obtain a normalized change rate sequence;

[0090] Step S12313: Set a mutation point identification threshold ξ, mark the moment when the normalized change rate exceeds ξ as a mutation point, and generate a mutation point sequence.

[0091] Specifically, in step S1231, the primary goal is to identify the mutation points in the observation sequence (i.e., the non-stationary residual component). This provides basic data for subsequent determination of the nature of these mutation points (whether they are normal or abnormal), thereby improving the accuracy of the judgment of changes in occupant density and making the air conditioning control decisions based on this more reasonable and energy-efficient. The observation sequence reflects the short-term fluctuations in occupant density.

[0092] In step S12311, the rate of change of the population density at each moment in the observation sequence is calculated to generate a rate of change sequence. The observation sequence reflects the short-term fluctuation of the population density. The rate of change of the population density is a key indicator to measure how fast the population density changes over time. When calculating the rate of change of the population density, for a certain moment in the observation sequence, , assuming that the population density at this moment is , the population density at the previous moment is , then the population density change rate The calculation formula is (when To avoid division by zero, other reasonable calculation methods can be used, such as using a very small positive number replace For example, at a certain moment According to the collected data, the population density in a certain area is 50 people. The population density is 40 people. According to the formula, the population density change rate at this moment is . By calculating each moment in the observation sequence, the rate of change sequence can be obtained. The beneficial effect of this step is that the rate of change sequence can highlight the trend and amplitude of changes in population density. Compared with the original population density data, it can more keenly capture changes in population density. The original population density data may not be easy to intuitively find its changing trend due to its large values and relatively slow changes. The rate of change sequence quantifies the changing trend, making the changes in population density clearer and more discernible. For example, in an office area on a weekday, the population density may gradually increase over a long period of time, but the growth trend of the original data is not obvious. By calculating the rate of change sequence, you can clearly see the speed of its changes, which helps to promptly detect abnormal fluctuations in population density and provide a more sensitive data basis for subsequent abnormal judgments.

[0093] In step S12312, the rate of change sequence is normalized to obtain a normalized rate of change sequence. Normalization is a common data processing method whose purpose is to map data with different dimensions and value ranges to a unified interval, making the data comparable. For the rate of change sequence, the minimum-maximum normalization method is typically used to map all values in the rate of change sequence to the interval [0, 1] to obtain a normalized rate of change sequence. The beneficial effect of this step is that after normalization, the population density change rates at different times and amplitudes are on the same dimensional scale, facilitating the subsequent setting of a unified threshold for mutation point identification. In actual applications, the population density change rates in different regions and time periods can vary significantly. Without normalization, it is difficult to establish a universal standard for mutation point identification. However, after normalization, analysis can be performed based on a unified standard, improving the accuracy and reliability of mutation point identification and avoiding misjudgments due to dimensional differences. For example, in conference rooms on different floors, due to the different sizes and usage frequencies of the conference rooms, the numerical ranges of the population density change rates vary greatly. Through normalization processing, this difference can be eliminated, providing a reliable basis for accurately judging the mutation point.

[0094] In step S12313, a mutation point identification threshold ξ is set. Moments where the normalized rate of change exceeds ξ are marked as mutation points, generating a mutation point sequence. The mutation point identification threshold ξ is a key parameter for distinguishing normal from abnormal changes in occupant density. It is determined based on extensive historical data and practical experience. In practice, it is necessary to conduct long-term monitoring and analysis of occupant density changes in different areas of a building over different time periods. The fluctuation range of the rate of change under normal circumstances must be observed, and an appropriate threshold value must be determined by comprehensively considering factors such as occupant activity patterns and building functional characteristics. For example, after long-term monitoring and analysis of occupant density data for an office building, it was found that the normalized rate of change of occupant density rarely exceeds 0.3 under normal circumstances. Therefore, the mutation point identification threshold ξ is set to 0.3. When the normalized rate of change at a given moment reaches or exceeds 0.3, that moment is marked as a mutation point. All marked mutation points constitute a mutation point sequence. The beneficial effect of this step is that by setting the mutation point identification threshold, moments of drastic changes in occupant density can be quickly identified, providing key clues for subsequently determining whether these changes are caused by normal events or unexpected events. This helps promptly detect potential abnormal occupant movement, which is crucial for precise air conditioning system control and optimal energy allocation. Failure to accurately identify sudden changes in occupant density can lead to missed critical changes in occupant density that require adjustments to the air conditioning system, resulting in decreased indoor comfort and wasted energy. For example, at the start and end of a conference room meeting, occupant density can fluctuate significantly. Failure to accurately identify these sudden changes can prevent the air conditioning system from adjusting cooling or heating power in a timely manner, impacting indoor comfort and causing inappropriate energy consumption.

[0095] Step S1232: performing correlation analysis based on the agenda correlation track to determine whether the mutation point is a normal mutation point or an abnormal mutation point;

[0096] Furthermore, step S1232 includes:

[0097] Step S12321, extract the timestamp of each mutation point in the mutation point sequence , and retrieve the timestamp in the agenda associated track Agenda events within the nearby Δt' time range;

[0098] Step S12322: If there is an agenda event within the range of Δt', then a correlation analysis is performed between the agenda multi-dimensional attribute parameters of the agenda event and the personnel density change rate at the mutation point to calculate the correlation coefficient ρ';

[0099] Step S12323: If the correlation coefficient ρ' exceeds the preset correlation threshold ρ0, the mutation point is marked as a normal mutation point, otherwise it is marked as an abnormal mutation point.

[0100] Specifically, in step S1232, the main purpose is to perform correlation analysis based on the agenda-related trajectory to determine whether the mutation point is a normal mutation point or an abnormal mutation point, so as to more accurately judge the cause of the change in personnel density, provide a reliable basis for the subsequent precise control of the air-conditioning system, and improve the accuracy and reliability of the air-conditioning group control energy-saving method.

[0101] Step S12321 extracts the timestamp of each mutation point in the mutation point sequence , and retrieve the timestamp in the agenda associated track Agenda events within the nearby Δt' time range; timestamp The precise moment of the mutation point is recorded, while Δt' is a time range set based on actual circumstances, used to identify potentially related agenda events near the mutation point. For example, if the mutation point occurs at 10:00 AM and Δt' is set to 30 minutes before and after, then the agenda association track is searched for agenda events between 9:30 AM and 10:30 AM. This step aims to identify agenda information potentially related to the mutation point, providing a basis for subsequent determination of the nature of the mutation point. Its beneficial effect is that by establishing a temporal association between the mutation point and nearby agenda events, the scope of determination can be narrowed, allowing for more targeted analysis of the cause of the mutation point. Without this step, it would be difficult to directly identify information related to the mutation point from the numerous agenda events, resulting in inefficient and error-prone determination. For example, in a large office building, numerous meetings and personnel activities occur daily. Without associating agenda events by timestamp, it would be difficult to determine which agenda event a particular population density mutation point is associated with. This step, however, allows for rapid identification of potentially related agenda items, improving the accuracy and efficiency of determination.

[0102] In step S12322, correlation analysis is a statistical method used to measure the closeness of the linear relationship between two variables. In this step, a correlation analysis is performed between the multidimensional attribute parameters of the agenda event (such as the number of participants and meeting duration) and the rate of change in the crowd density at the mutation point. The Pearson correlation coefficient is a commonly used calculation method. This step is beneficial in that by calculating the correlation coefficient, the degree of association between the agenda event and the change in crowd density can be accurately measured, providing a quantitative basis for determining whether the mutation point is normal or abnormal. Compared to purely qualitative judgments, quantitative results are more objective and accurate, reducing the subjectivity and uncertainty of human judgment. For example, when determining whether a crowd density mutation point is caused by a meeting, calculating the correlation coefficient can more accurately determine the relationship between the two, avoiding incorrect adjustments to the air conditioning control strategy due to subjective misjudgment.

[0103] In step S12323, the preset correlation threshold ρ0 is determined based on actual conditions and experience to distinguish between normal and abnormal mutation points. When the correlation coefficient ρ' exceeds ρ0, it indicates a strong correlation between the agenda event and the rate of change in occupant density, and the mutation point is likely caused by a normal agenda event. Otherwise, it is considered an abnormal mutation point caused by an unexpected event (such as an unscheduled meeting or overstay). For example, after extensive data analysis and practical verification, the correlation threshold ρ0 was determined to be 0.6. When the calculated correlation coefficient ρ' is 0.7, the mutation point is marked as a normal mutation point; if ρ' is 0.4, it is marked as an abnormal mutation point. The beneficial effect of this step is that it classifies mutation points through clear judgment criteria, making the determination of the cause of occupant density changes clearer and more accurate. This is of great significance for adjusting the control strategy of the air conditioning system. Areas corresponding to normal mutation points can be air-conditioned according to the conventional control strategy, while areas corresponding to abnormal mutation points require further analysis and special control measures to ensure indoor comfort and efficient energy utilization. For example, during normal meetings, changes in air conditioning demand caused by changes in crowd density can be adjusted according to preset strategies. However, for unexpected gatherings of people, such as temporary training activities, the air conditioning power needs to be adjusted in a timely manner to meet the needs of indoor people while avoiding energy waste.

[0104] Step S1233: Mark the value corresponding to the abnormal mutation point in the observation sequence as an outlier; use a data smoothing method to obtain a smoothed and corrected non-stationary residual component;

[0105] The method of adopting the data smoothing method to obtain the smoothed and corrected non-stationary residual component includes: replacing the outliers with the weighted average of adjacent normal values.

[0106] Specifically, the values corresponding to abnormal mutation points in the observation sequence (i.e., non-stationary residual components) are first labeled as outliers. An abnormal mutation point indicates a dramatic change in population density that does not conform to normal patterns. For example, in the population density data for a certain office area, if the population density changes relatively slowly under normal circumstances, it suddenly increases significantly at a certain moment, and this change is determined to be unrelated to known agenda events, this point is identified as an abnormal mutation point, and its value in the observation sequence is an outlier. Then, using data smoothing methods, the outliers are replaced with the weighted average of adjacent normal values to obtain the smoothed and corrected non-stationary residual component. The weighting coefficient is typically determined based on the distance between the outlier and adjacent normal values or other relevant factors, with normal values closer to the outlier receiving a larger weighting coefficient. By processing the values corresponding to all abnormal mutation points, the smoothed and corrected non-stationary residual component can be obtained.

[0107] This step has significant benefits. From a data quality perspective, it eliminates the impact of abnormal events on non-stationary residual components, allowing the data to more accurately reflect actual changes in occupancy density. Outliers can interfere with subsequent data analysis and model calculations, leading to biased results. For example, when constructing a spatiotemporal energy-demand coupled map, using non-stationary residual components containing outliers can lead to inaccurate representation of energy demand intensity in certain areas of the map, affecting the assessment of actual energy demand within the building. Smoothing, however, makes the data more stable and reliable, improving the accuracy of subsequent analysis results. From the perspective of developing air conditioning group control strategies, analyzing the smoothed and corrected non-stationary residual components can more accurately determine the changing trends in occupancy density, thereby providing more appropriate control signals for the air conditioning system. For example, when predicting future changes in occupancy density, using corrected data can improve prediction accuracy, allowing the air conditioning system to make appropriate adjustments in advance, avoiding energy waste and improving indoor comfort.

[0108] Step S1234, calculate the observation value weight coefficient ω according to the correlation coefficient ρ' of the normal mutation point, and construct the observation value weight matrix:

[0109]

[0110] Here, λ is the weight decay factor, which can be set empirically. The larger ρ' is, the stronger the correlation between the mutation point and the agenda event, and the closer the weight coefficient ω is to 1. Conversely, the smaller ρ' is, the weaker the correlation is, and the closer the weight coefficient ω is to 0.

[0111] Specifically, the purpose of step S1234 is to quantify the credibility of observations at different times, and provide a weighted basis for the subsequent heterogeneous trajectory fusion, thereby improving the accuracy of the fusion result. First, according to the formula Calculate observation weight coefficients ,in is the weight attenuation factor, which can be set based on experience. It is obtained by performing a correlation analysis on the agenda multi-dimensional attribute parameters of the agenda event and the personnel density change rate of the mutation point in step S1232. The larger the value, the stronger the correlation between the mutation point and the agenda event. For example, in a building, the correlation coefficient between the change in the number of participants in a meeting and the change rate of the mutation point of the personnel density is When it is high, it indicates that the mutation point is likely to be a normal change caused by this meeting. Used to adjust the weight coefficient Follow The speed of change. When the value is large, Follow More sensitive to changes in When the value is small, The change of is relatively gentle. By calculating the weight coefficient ω for each normal mutation point and arranging these weight coefficients in a certain order, the observation value weight matrix is constructed.

[0112] The beneficial effects of this step are reflected in several aspects. The observation weight matrix plays a key role in the heterogeneous trajectory fusion process. Observations at different times have varying degrees of reliability due to the influence of various factors. Using the observation weight matrix, highly reliable observations (i.e., those with weight coefficients close to 1) are assigned greater weight during the fusion process, while less reliable observations (those with weight coefficients close to 0) are given smaller weights. This prevents unreliable observations from significantly influencing the fusion results, ensuring that the fused results more accurately reflect the actual situation. For example, when constructing a spatiotemporal energy-demand coupling map, weighting different data points based on the observation weight matrix can more accurately reflect the impact of factors such as occupancy density and activity intensity on energy demand, providing more reliable input for intelligent air conditioning optimization control. From the perspective of energy conservation in air conditioning cluster control, control strategies based on more accurate fusion results can more rationally allocate air conditioning resources, improve energy efficiency, meet indoor occupant comfort requirements, and achieve energy conservation goals for intelligent building air conditioning cluster control.

[0113] In step S1240 , the periodic fundamental frequency component, the smoothed and corrected non-stationary residual component, the local thermal entropy growth rate, and the thermal core diffusion coefficient are combined to form a heterogeneous trajectory that has undergone outlier suppression processing.

[0114] Specifically, the purpose of step S1240 is to provide a more accurate and representative data foundation for the subsequent construction of a spatiotemporal energy demand coupling map, thereby improving the accuracy and reliability of the energy-saving method for group air conditioning control in intelligent buildings. The periodic fundamental frequency component reflects the long-term trend of occupant density. As mentioned above, the daily entry and exit patterns of people in office spaces on weekdays exhibit a certain periodicity, which is reflected in the periodic fundamental frequency component. The smoothed non-stationary residual component removes the influence of abnormal mutation points and more accurately reflects short-term fluctuations in occupant density. For example, after data smoothing, the data is freed from interference from abnormal fluctuations in occupant density caused by unexpected events, making short-term fluctuations more representative of actual changes in occupant mobility. The local thermal entropy growth rate and the thermal nuclear diffusion coefficient describe the impact of occupant concentration on local temperature from a thermodynamic perspective. In densely populated areas, the local thermal entropy growth rate is higher, and the thermal nuclear diffusion coefficient can also exhibit specific values depending on factors such as occupant activity and building structure. For example, in a crowded conference room, heat accumulates rapidly, resulting in a high local thermal entropy growth rate. Simultaneously, heat diffuses into the surrounding area, and the thermal nuclear diffusion coefficient reflects this diffusion capacity. These different types of features are combined to form heterogeneous trajectories that have undergone outlier suppression because they provide information about the relationship between occupant activity and energy demand from different perspectives. The periodic fundamental frequency component and the non-stationary residual component focus on changes in occupant density, while the local thermal entropy growth rate and the thermal core diffusion coefficient focus on the impact of occupant aggregation on temperature. These complementary pieces of information provide a more comprehensive description of the relationship between occupant activity and energy demand within a building.

[0115] From the perspective of constructing a spatiotemporal energy demand coupling map, using heterogeneous trajectories processed with outlier suppression as input allows the constructed map to more accurately reflect the energy demand intensity of different areas within a building at different times. By removing the interference of abnormal data and integrating multiple aspects of information, the energy demand intensity in the map is more consistent with actual conditions. For example, the map more accurately shows which areas require more air conditioning energy due to long-term gatherings and high activity intensity, and which areas experience fluctuations in energy demand due to short-term gatherings. From the perspective of formulating air conditioning group control strategies, air conditioning control strategies based on a more accurate spatiotemporal energy demand coupling map can more rationally allocate air conditioning resources. Air conditioning parameters such as cooling or heating power and wind speed can be precisely adjusted based on the energy demand characteristics of different areas, avoiding energy waste and improving energy efficiency while also meeting the comfort requirements of indoor occupants, achieving the dual goals of energy conservation and comfort in intelligent building air conditioning group control.

[0116] Step S1300: construct a spatiotemporal energy demand coupling map based on the heterogeneous trajectories and observation weight matrix that have undergone outlier suppression processing; identify low-confidence areas of the spatiotemporal energy demand coupling map and mark the low-confidence areas of the spatiotemporal energy demand coupling map.

[0117] Furthermore, step S1300 includes:

[0118] Step S1310 , performing a tensor product operation on the heterogeneous trajectories that have undergone outlier suppression processing to generate a spatiotemporal energy demand coupling tensor;

[0119] Specifically, tensor product is a mathematical operation used to combine multiple vectors or matrices into a tensor of higher dimension to comprehensively express complex data information. In this embodiment, the heterogeneous trajectories processed with outlier suppression contain a variety of information such as periodic fundamental frequency components, non-stationary residual components after smooth correction, local thermal entropy growth rate and thermal core diffusion coefficient. These information reflect the relationship between personnel activities and energy demand in buildings from different perspectives. For example, the periodic fundamental frequency component reflects the long-term trend of personnel density, such as the pattern of people entering and leaving the building at fixed time periods every day on weekdays; the non-stationary residual components after smooth correction reflect short-term fluctuations in personnel density, such as the gathering or evacuation of people caused by sudden meetings; the local thermal entropy growth rate and thermal core diffusion coefficient describe the degree of influence of personnel gathering on the temperature of the local area and the heat diffusion. Assume that the heterogeneous trajectories processed with outlier suppression can be represented as vectors respectively. (Represents information related to the periodic fundamental frequency component) (represents information related to the non-stationary residual component after smoothing correction), (represents information related to the local thermal entropy growth rate), (represents information related to the thermal nuclear diffusion coefficient). The tensor product operation combines these vectors according to specific mathematical rules to generate a space-time energy coupling tensor Tensor Each element in the energy demand intensity corresponds to the energy demand intensity of a specific physical location in the building at a specific moment. It incorporates multiple influencing factors, such as occupant density, activity intensity, gathering patterns, and agenda attributes. For example, in a large office building, a tensor product operation integrates information about occupant activity across different floors and areas at different points in time. The resulting tensor element accurately reflects the energy demand intensity at that location and time. For example, at 10:00 a.m. on a weekday, the tensor element for the conference room area on a specific floor would comprehensively consider factors such as the occupant concentration during that period (reflected by the periodic fundamental frequency component and the non-stationary residual component), the heat generated by occupant activity, and its diffusion (reflected by the local thermal entropy growth rate and the thermal nuclear diffusion coefficient).

[0120] This step aims to integrate multi-source, heterogeneous data into a unified tensor structure that comprehensively reflects building energy demand. Its beneficial effects are evident in multiple ways. From a data processing perspective, the tensor product operation organically combines previously dispersed and independent information, forming a more comprehensive and systematic data set that facilitates subsequent unified analysis and processing. In building air conditioning cluster control scenarios, this integrated tensor data can more accurately reflect actual energy demand, providing a more comprehensive and accurate data foundation for the subsequent construction of spatiotemporal energy demand coupling maps. Compared to analyzing each information component separately, data in tensor form can more completely reflect the interactions and combined impacts of different factors, helping to improve understanding and grasp of building energy demand. From the perspective of air conditioning control strategy formulation, subsequent analysis based on this tensor can more accurately determine energy demand in different areas and at different times, providing strong support for the development of more reasonable and efficient air conditioning control strategies, and avoiding energy waste and reduced indoor comfort caused by incomplete or inaccurate information.

[0121] Step S1320, performing dimensionality reduction projection on the spatiotemporal energy demand coupling tensor to form a two-dimensional spatiotemporal energy demand coupling map that matches the building plan layout;

[0122] Specifically, the spatiotemporal energy demand coupling tensor is a high-dimensional data structure that contains rich energy demand information for each location within a building at different times. However, direct use of this high-dimensional data for analysis and decision-making presents challenges. Through dimensionality reduction and projection, it is converted into a two-dimensional map that matches the building's floor plan, making the data more intuitive. For example, for a multi-story commercial building, the three-dimensional spatiotemporal energy demand coupling tensor is projected onto a two-dimensional plane. Each pixel in the map corresponds to a specific area on the building plane (such as a store or public area), and the color of the pixel intuitively corresponds to its energy demand intensity, with darker colors indicating higher energy demand at a given moment. Techniques such as principal component analysis (PCA) are often used to implement dimensionality reduction and projection. PCA performs eigenvalue decomposition on the tensor data to identify the principal components (i.e., directions with the greatest variance). These principal components represent the key features of the data. The tensor data is then projected into a low-dimensional space formed by these principal components. In practice, the covariance matrix of the spatiotemporal energy demand coupling tensor is first calculated. The covariance matrix reflects the correlations between the various dimensions. The covariance matrix is then subjected to eigenvalue decomposition to obtain eigenvalues and eigenvectors. The first few major eigenvectors (those with larger eigenvalues, representing the main direction of data variation) are selected, and the tensor data is projected into the two-dimensional space spanned by these eigenvectors, ultimately forming a two-dimensional space-time energy-demand coupling map.

[0123] The primary purpose of this step is to transform complex, high-dimensional energy demand data into an intuitive, easy-to-understand two-dimensional map, with significant benefits. From a visualization perspective, the two-dimensional map significantly improves data readability and comprehensibility. Managers or control systems can quickly extract key information from the map, such as clearly identifying areas within the building with high and low energy demand, as well as the spatial distribution trend of energy demand. This helps quickly locate energy consumption hotspots and facilitates targeted adjustments to the air conditioning system. From a control strategy development perspective, the two-dimensional map provides a more concise data structure for subsequent data analysis and processing. Map-based operations such as cluster analysis and trend prediction are more efficient, enabling more accurate control strategies for air conditioning fleets. For example, based on the energy demand intensity of different areas in the map, cooling or heating power for air conditioners can be rationally allocated, optimizing energy distribution, reducing energy waste, and improving the building's overall energy management.

[0124] Step S1330, assigning credibility attributes to pixels in the two-dimensional spatiotemporal energy demand coupling map according to the observation value weight matrix to obtain the spatiotemporal energy demand coupling map;

[0125] Specifically, the observation weight matrix is obtained in the previous step by calculating the correlation coefficient of normal mutation points. Its function is to quantify the credibility of observations at different times. In the two-dimensional spatiotemporal energy demand coupling map, each pixel represents the energy demand intensity of a certain area in the building at a certain moment. However, due to the accuracy of personnel density prediction and the degree of impact of abnormal events, the credibility of the energy demand intensity represented by these pixels varies. For example, in some areas with stable personnel flow patterns and little impact from abnormal events, such as regular office areas, the energy demand intensity reflected by the pixels is more credible; while in some areas with complex personnel flow and easy to be affected by emergencies, such as near large conference rooms or event venues, the credibility of the pixels is relatively low.

[0126] Based on the observation weight matrix, each pixel is assigned a corresponding weight coefficient as a credibility attribute. Assuming a weight coefficient in the observation weight matrix is ω, and the corresponding pixel energy demand intensity is I, the credibility attribute for that pixel can be represented as a two-tuple (I, ω) consisting of the energy demand intensity and the weight coefficient. Pixels with high credibility (i.e., weight coefficients ω close to 1) indicate that the occupancy density prediction for that area is relatively accurate and less affected by abnormal events. Their energy demand intensity can be directly used to guide air conditioning control decisions. For example, in a typical office floor, after long-term data monitoring and analysis, the weight coefficients of the pixels in that area are close to 1. When formulating air conditioning control strategies, the energy demand intensity of that pixel can be directly used to adjust air conditioning operating parameters, such as temperature settings and fan speed adjustments. However, for pixels with low credibility (weight coefficients ω close to 0), their energy demand intensity requires further correction before it can be used in control decisions. For example, this can be corrected by incorporating more real-time data, such as current occupancy monitoring data, changes in outdoor ambient temperature, or by using other more accurate prediction models to ensure accurate air conditioning control.

[0127] The primary purpose of this step is to improve the accuracy and reliability of air conditioning control decisions. By assigning a credibility attribute to pixels, the control system can more effectively utilize the atlas information. This beneficial effect is reflected in multiple aspects. Regarding the accuracy of air conditioning control decisions, for areas with high credibility, precise air conditioning control can be performed directly based on their energy demand intensity, improving control accuracy and timeliness, and avoiding energy waste and reduced indoor comfort caused by inaccurate data. For example, in areas with high credibility, air conditioning power can be precisely adjusted to meet occupant comfort needs while minimizing energy consumption. For areas with low credibility, control errors caused by blindly using inaccurate data are avoided, reducing the risk of energy waste and reduced indoor comfort. From the perspective of system stability and reliability, this credibility-based approach enhances the adaptability of the entire air conditioning group control system to complex environments and uncertainties. Even in situations with complex occupant movement or abnormal events, the system can maintain stable operation by rationally processing low-credibility data, ensuring indoor comfort while achieving energy savings.

[0128] Step S1340: Identify that the credibility attribute in the spatiotemporal energy demand coupling map is lower than a preset confidence threshold. The low confidence areas of the space-time energy demand coupling map are marked.

[0129] Specifically, low confidence areas correspond to areas with drastic fluctuations in population density and are greatly affected by abnormal events. The energy demand intensity in these areas needs to be further corrected before being used for air conditioning control decisions. The identification of low confidence areas is mainly based on the credibility attribute assigned to the pixels by the observation weight matrix. Generally speaking, when the credibility attribute of a pixel is lower than a pre-set confidence threshold (set as ), the area where the pixel is located can be identified as a low confidence area. This threshold The setting is based on extensive historical data and practical experience, and is determined through a comprehensive analysis of factors such as changes in occupancy density, the frequency of abnormal events, and the effectiveness of air conditioning control in different areas under different circumstances. For example, after long-term monitoring and analysis of data from a certain building, it was found that when the weight coefficient is less than 0.5, the energy demand intensity of the area represented by the pixel is significantly affected by fluctuations in occupancy density and abnormal events. Therefore, 0.5 is used as the threshold for identifying low-confidence areas.

[0130] After the low confidence area is determined, it needs to be marked. There are many ways to mark it, such as highlighting the low confidence area with a specific color (such as red) on the map, or adding special identification fields to the pixels in the low confidence area at the data level. For example, on the visualized spatiotemporal energy demand coupling map, the pixels in the low confidence area are filled with red, so that when viewing the map, relevant personnel can intuitively distinguish which areas have low energy demand intensity confidence. During data storage and processing, add "low confidence "Identification fields make it easier for subsequent data analysis and processing programs to quickly identify these areas.

[0131] The purpose of this step is to identify areas within the map where energy demand data exhibits high uncertainty, enabling appropriate action to be taken during the subsequent development of air conditioning control strategies. Its benefits are primarily reflected in the following aspects. From the perspective of air conditioning control accuracy, marking low-confidence areas prevents the direct use of inaccurate energy demand data in these areas for control decisions, thereby improving control precision. For example, directly implementing air conditioning control based on raw energy demand data in low-confidence areas may result in over-cooling or over-heating, resulting in energy waste and poor indoor comfort. However, marking these areas allows for more precise monitoring methods or more complex predictive models to be used to correct energy demand data, enabling more precise control. From an energy management perspective, accurately identifying low-confidence areas helps optimize energy allocation. Air conditioning operation strategies can be adjusted based on actual conditions in these areas. For example, when low-confidence areas are less frequent, air conditioning power can be appropriately reduced to avoid unnecessary energy consumption and improve energy efficiency. From the perspective of overall system stability, effectively addressing low-confidence areas enhances the stability and reliability of the air conditioning group control system. Through special treatment of these areas with higher uncertainty, system fluctuations and failures caused by inaccurate data are reduced, the stable operation of the air-conditioning system of the entire building is guaranteed, and the comfort experience of indoor occupants is improved.

[0132] Step S2000: Based on the spatiotemporal energy demand coupling map, an optimized airlock compensation factor β' is obtained; multi-scale clustering is performed on the spatiotemporal energy demand coupling map to extract significant energy demand regions and insignificant energy demand regions, and energy demand time series of the significant energy demand regions are extracted; the significant energy demand regions are divided into steady-state cores and transient cores, and the insignificant energy demand regions are marked as dormant cores;

[0133] Furthermore, step S2000 includes:

[0134] Step S2100: activating the air hysteresis calibration loop in the low confidence region of the spatiotemporal energy demand coupling map, learning the probability distribution of the air hysteresis compensation factor β through the adversarial generative network, and sampling to obtain the optimized air hysteresis compensation factor β';

[0135] Further, if Figure 6 As shown, step S2100 includes:

[0136] Step S2110: extract the air conditioning operating parameter sequence of the same type of area in the same period in history and calculate the ideal temperature control response curve;

[0137] Specifically, historically co-cyclical, co-typed areas refer to building areas that are in the same time period as the current moment (e.g., the same time of day, the same weekday, etc.) and have similar functions and occupant activity patterns. For example, in a large office building, the office areas on each floor from 9:00 AM to 11:00 AM every Monday would constitute historically co-cyclical, co-typed areas. These areas share similarities in occupant activity patterns and air conditioning usage habits, and their air conditioning operating parameter sequences can reflect the operating characteristics of the air conditioning system under normal operating conditions. These air conditioning operating parameter sequences encompass a variety of parameters, such as temperature setpoints, compressor frequency, and fan speed. The changes in these parameters over time reflect the operating status of the air conditioning system at different moments. Using big data mining techniques, it is possible to filter out the air conditioning operating parameter sequences for these areas from massive amounts of historical data.

[0138] The ideal temperature control response curve is calculated based on the extracted historical data through data analysis and processing methods. It reflects the standard response law of the air conditioning system to the given energy demand change under normal working conditions. Assume that the temperature setting values of a certain area at multiple times in the same period of history are , the corresponding time point is ,in, is the temperature setting value at the nth moment, for At the corresponding time point, by performing curve fitting on these data (such as using a fitting algorithm such as the least squares method), a functional relationship that can describe the change of the temperature setting value over time is obtained. The curve corresponding to this function is the ideal temperature control response curve.

[0139] The purpose of this step is to obtain a reference curve that represents the standard response of the air conditioning system under normal operating conditions, providing a benchmark for subsequent evaluation of the actual temperature control response in the current low-confidence zone. This beneficial effect is reflected in multiple aspects. From the perspective of air conditioning system performance evaluation, the ideal temperature control response curve provides a reference for determining whether the current air conditioning system is operating normally. If the actual temperature control response in the current area deviates significantly from this curve, it indicates that the air conditioning system may be operating abnormally or the control strategy is unreasonable. From the perspective of energy-saving control, it provides a basis for optimizing the air conditioning control strategy. By comparing the actual response with the ideal curve, shortcomings in the current control strategy can be identified, allowing targeted adjustments to improve energy efficiency. For example, if the actual temperature rise rate is found to be slower than the ideal curve, it may mean that the cooling or heating power of the air conditioning system is insufficient, and the power setting should be appropriately increased to avoid wasting energy in ineffective temperature adjustment.

[0140] Step S2120 , comparing the deviation between the actual temperature response in the current low confidence zone and the ideal temperature control response curve, and quantifying the deviation as a temperature control hysteresis ΔT;

[0141] Specifically, the measured temperature response in the current low-confidence zone is collected using an array of temperature sensors installed throughout the building. These temperature sensors are distributed throughout the building and can monitor temperature changes in the area in real time. By taking a point-by-point difference between the measured temperature response and the ideal temperature control response curve obtained in step S2110, a quantitative indicator of the lead or lag between the cold and heat source output and the terminal temperature change can be calculated, namely the temperature control lag ΔT. The magnitude of the temperature control lag reflects the air conditioning system's response speed to energy demand fluctuations. The larger the lag, the slower the air conditioning system's response speed to energy demand fluctuations. For example, if at a certain moment, the ideal temperature control response curve indicates that the temperature should reach 25°C, but the measured temperature is only 22°C, the temperature control lag is -3°C, indicating that the air conditioning system has failed to adjust the temperature to the ideal state in a timely manner and has a slow response speed.

[0142] The primary purpose of this step is to quantify the temperature control response hysteresis of the air conditioning system in the current low-confidence zone, providing a quantitative basis for subsequent control measures. This has significant benefits. From the perspective of energy conservation potential, accurately quantifying temperature control hysteresis can reveal potential energy savings in the air conditioning system. A large temperature control hysteresis indicates a significant delay in the air conditioning system's response to changes in energy demand, potentially wasting energy while waiting for temperature adjustment. By focusing on and optimizing these areas, energy efficiency can be improved. For example, in areas with significant temperature control hysteresis, air conditioning operating parameters can be adjusted in advance to enable faster response to energy demand changes and reduce energy waste. From the perspective of improving indoor comfort, understanding temperature control hysteresis facilitates timely adjustments to air conditioning control strategies, ensuring that indoor temperatures reach their setpoints quickly and accurately, enhancing occupant comfort. If temperature control hysteresis persists, indoor temperatures may remain outside the comfortable range for extended periods, impacting occupants' work and daily lives.

[0143] Step S2130: constructing a generative adversarial network (ACGAN) based on the temperature control hysteresis ΔT, and learning the probability distribution of the hovering compensation factor β through the generator G in the generative adversarial network (ACGAN);

[0144] Specifically, the ACGAN is a deep learning model consisting of a generator G and a discriminator D. In this step, the generator G receives random noise z and the temperature control hysteresis ΔT as input. The random noise z is a randomly generated vector within a certain range. Its function is to provide diverse inputs to the generator, enabling it to produce different results and preventing the model from falling into a local optimal solution.

[0145] Generator Mapping to The probability distribution of : .in, is the Sigmoid activation function, which can map the input value to Intervals are often used to convert numerical values into probability form; is the entropy weight of the spatiotemporal energy demand coupling map pixel. The entropy weight reflects the importance of the pixel in the energy demand analysis. It is obtained by calculating the entropy value of the information contained in the pixel. The larger the entropy value, the higher the uncertainty of the pixel, and its weight is Adjust accordingly to highlight or weaken the role of the pixel in the calculation; The sub-region temperature gradient describes the speed of temperature change between different sub-regions and is obtained by calculating the ratio of the temperature difference to the distance between adjacent sub-regions at the same time. is the hyperbolic tangent function, It is the air pressure change rate, which reflects the change of air pressure over time and is obtained by differential calculation of air pressure values at different times.

[0146] The discriminator D in the generative adversarial network (ACGAN) compares the β values generated by the generator G with the distribution of real data, guiding the generator G to learn the prior distribution of β by minimizing the JS divergence. JS divergence measures the degree of difference between two probability distributions. By continuously adjusting the parameters of the generator G, the distribution of the generated β values is made as close as possible to the distribution of real data. For example, during the actual training process, the discriminator D will judge whether the β values generated by the generator G match the characteristics of the real data. If not, feedback is provided to the generator G, instructing it to adjust its parameters and regenerate until the generated β values can deceive the discriminator D, making it difficult for the discriminator D to distinguish between the generated β values and the real data.

[0147] Introducing temperature gradients and pressure change rates establishes a quantitative correlation between the environmental physical field and the intensity of human activity, increasing the confidence level in the generated compensation factor. Because human activity intensity can affect temperature and pressure changes in a local area, incorporating these physical quantities into the calculation more accurately reflects actual conditions, making the generated hovering compensation factor β more in line with practical needs.

[0148] The purpose of this step is to leverage the powerful learning capabilities of the generative adversarial network to learn a reasonable probability distribution for the air lag compensation factor β based on the temperature control hysteresis and related environmental physical quantities. This has beneficial effects in several ways. From the perspective of air conditioning control strategy optimization, accurately learning the probability distribution of β provides a more reliable basis for subsequent sampling to optimize the air lag compensation factor β'.

[0149] Step S2140: sampling from the probability distribution of β to obtain the optimized hovering compensation factor β'.

[0150] Specifically, in step S2130, the probability distribution of the airlock compensation factor β is obtained through learning using the Generative Adversarial Network (ACGAN). Sampling is the process of randomly selecting a value from this probability distribution as the optimized airlock compensation factor β'. For example, assuming the probability distribution of β conforms to the normal distribution N(μ, σ²) (μ is the mean, σ² is the variance), a random number generator can be used to generate a random number based on the probability density function of this normal distribution. This random number is the sampled optimized airlock compensation factor β'. The sampled β' value can best match the current energy demand response lag, providing an adaptive basis for the subsequent dynamic configuration of air conditioning control parameters. A larger β' value indicates a greater power compensation is required to overcome the lag effect. For example, in a conference room, the rapid gathering of people at the start of the meeting causes the indoor temperature to rise rapidly, but the air conditioning system is responding with a lag. A larger β' value sampled in this step indicates that in subsequent air conditioning control, the cooling power of the air conditioner needs to be increased to quickly reduce the indoor temperature and meet the comfort requirements of the occupants.

[0151] The goal of this step is to obtain an optimized deadweight compensation factor β' that can be adaptively adjusted based on current conditions, providing a key parameter for precise control of the air conditioning system. Its beneficial effects are reflected in multiple aspects. From the perspective of adaptive control of the air conditioning system, β' can dynamically adjust based on the energy demand response lag in different areas, enabling the air conditioning system to better adapt to complex and changing indoor environments. Whether it is rapid temperature changes caused by sudden gatherings of people or temperature control lags caused by other factors, the air conditioning system can promptly adjust control parameters based on β', improving system response speed and stability. From the perspective of energy conservation and comfort assurance, a reasonable β' value ensures that the air conditioning system meets indoor comfort requirements while avoiding unnecessary energy consumption. Through precise power compensation, indoor temperature can be adjusted quickly without wasting energy due to overcompensation, achieving dual optimization of energy conservation and comfort. For example, β' can be adjusted according to actual conditions in different seasons, time periods, and occupant activity patterns, ensuring that the air conditioning system operates efficiently at all times, improving the overall energy management level of the building and the comfort experience of the occupants.

[0152] Step S2200 , performing multi-scale clustering on the spatiotemporal energy demand coupling map, extracting energy demand significant regions and non-energy demand significant regions, and extracting energy demand time series of the energy demand significant regions;

[0153] Furthermore, if Figure 7 As shown, step S2200 includes:

[0154] Step S2210, applying local binary pattern coding to the spatiotemporal energy demand coupling map, extracting energy demand gradient features, and generating an LBP feature map;

[0155] Specifically, the Local Binary Pattern (LBP) algorithm is an algorithm used to describe local texture features in an image. In this embodiment, it is applied to a spatiotemporal energy demand coupling map to extract the spatial variation of energy demand. For each pixel in the spatiotemporal energy demand coupling map, LBP generates a binary code by comparing the magnitude of the central pixel with the surrounding pixels. Assuming the central pixel value is Ic, and the surrounding pixel values are I1, I2, ..., I8 (using a common 3×3 neighborhood as an example), the LBP code is calculated as follows: starting from I1, compare Ii with Ic (i = 1, 2, ..., 8) in a clockwise direction. If Ii ≥ Ic, the corresponding bit is marked as 1; otherwise, it is marked as 0, resulting in an 8-bit binary number. For example, if the central pixel value is 50 and the surrounding pixel values are 45, 55, 52, 48, 50, 53, 47, and 51, the corresponding LBP code is 01101100. By calculating each pixel in the map, an LBP feature map can be generated. The LBP feature map reflects areas of sudden changes in energy demand in the spatial dimension. These sudden changes often correspond to sharp increases in energy demand caused by the gathering of people. Because the gathering of people causes a rapid increase in energy demand in that area, it manifests as a significant change in pixel values in the spatiotemporal energy demand coupling map. LBP coding can keenly capture such changes. For example, when a meeting is about to begin in a conference room and a large number of people arrive, the pixel values in the spatiotemporal energy demand coupling map for that area will rise rapidly. Through LBP coding, these changes are reflected in the LBP feature map as specific binary patterns, thereby highlighting the spatial variation characteristics of energy demand.

[0156] The purpose of this step is to convert the energy demand information in the spatiotemporal energy demand coupling map into a more easily analyzable texture feature form. Its beneficial effect is that the LBP feature map can enhance the characteristic expression of energy demand variation areas, making the subsequent extraction of significant energy demand regions more accurate. Compared with the original spatiotemporal energy demand coupling map, the LBP feature map can more clearly demonstrate the spatial variation trends and local characteristics of energy demand, providing a more effective data foundation for subsequent cluster analysis.

[0157] Step S2220, performing morphological filtering on the LBP feature map;

[0158] Specifically, morphological filtering is an image processing technique based on mathematical morphology, primarily used for image denoising, feature extraction, and shape analysis. In this step, the LBP feature map is sequentially eroded and dilated using opening and closing operations, effectively suppressing noise points caused by sensor errors and fusing small energy islands. The opening operation first performs an erosion operation on the LBP feature map. Erosion is a basic morphological operation that slides a structuring element (usually a small geometric shape such as a circle or square) across the image, replacing the center pixel value with the minimum pixel value within the structuring element's coverage. For example, for an LBP feature map eroded with a 3×3 square structuring element, if the minimum pixel value within the area covered by the structuring element is 0 and the center pixel value is originally 1, the erosion will cause the center pixel value to become 0. Erosion can remove isolated noise points and small bumps in the image, as the structuring element replaces these noise points and small bumps with the surrounding background values. Next, the closing operation is performed, which is a dilation operation applied to the erosion. Dilation is the opposite of erosion. It replaces the center pixel value with the maximum pixel value within the area covered by the structuring element. For example, if the maximum pixel value within the area covered by the structuring element is 1 and the center pixel value is originally 0, then after dilation, the center pixel value becomes 1. Dilation can fill small holes and narrow discontinuities in an image, blending fragmented energy islands to make the shape of the energy area more continuous and complete.

[0159] By performing morphological filtering on the LBP feature map, high-frequency noise and isolated pixels are eliminated, improving the image quality and stability. The purpose of this step is to provide more reliable data for the subsequent accurate extraction of significant energy demand areas. Its beneficial effects are reflected in multiple aspects. From the perspective of data accuracy, removing noise points and fusing fragmented areas can avoid misjudgments caused by sensor errors or data fluctuations, making the subsequently extracted significant energy demand areas more consistent with actual conditions. For example, in certain areas within a building, sensors may generate some abnormal energy demand data points due to interference. These noise points are removed after morphological filtering, thereby improving the accuracy of energy demand analysis. From the perspective of subsequent processing efficiency, the filtered image data is smoother and more continuous, which is conducive to the subsequent use of clustering algorithms for region division, reducing the complexity of the algorithm processing and improving processing efficiency. For example, when using the adaptive spectral clustering algorithm, the filtered LBP feature map can enable the algorithm to converge to accurate clustering results more quickly, saving computing resources and time.

[0160] Step S2230, using the LBP feature map after morphological filtering as input, and using the adaptive spectral clustering algorithm to extract the energy-demand salient regions and the energy-non-salient regions;

[0161] Specifically, the spectral clustering algorithm is a clustering algorithm based on graph theory. It calculates the similarity matrix between data points and performs spectral decomposition on it to obtain a clustering result that can maximize the similarity within the cluster and minimize the similarity between clusters. In this embodiment, for the LBP feature map after morphological filtering, each pixel is regarded as a data point. First, the similarity matrix S between the data points is calculated. The calculation of similarity is usually based on the distance metric between the data points, such as the Euclidean distance. After obtaining the similarity matrix S, it is spectrally decomposed. Spectral decomposition is the process of decomposing a matrix into eigenvalues and eigenvectors, that is, S=UΛU T , where Λ is a diagonal matrix whose diagonal elements are eigenvalues, U is the eigenvector matrix, U T is the transpose of the eigenvector matrix. By selecting appropriate eigenvectors, the data points can be divided into different clusters.

[0162] The adaptive mechanism plays a key role in the spectral clustering process. It automatically determines the optimal number of clusters based on changes in modularity during the clustering process. Modularity is a metric used to evaluate clustering quality. During the clustering process, different numbers of clusters are repeatedly tried, the corresponding modularity values are calculated, and the number of clusters that maximizes the modularity value is selected as the optimal number of clusters. Mapping the clustering results back to the original spatiotemporal map yields a set of areas with prominent energy demands and dense personnel density—known as energy-demand-significant areas—while the remaining areas are classified as non-energy-demand-significant areas. For example, in the spatiotemporal energy demand coupling map analysis of a large office building, the adaptive spectral clustering algorithm can accurately identify high-energy-demand areas such as densely populated office areas and conference rooms as energy-significant areas, while areas with relatively low energy demands, such as corridors and restrooms, are identified as non-energy-demand-significant areas.

[0163] The goal of this step is to accurately delineate energy-demand-significant and energy-insignificant areas from the processed LBP feature map, providing a foundation for developing tailored air conditioning control strategies for these areas. This benefit is reflected in several aspects. From an energy management perspective, accurately demarcating energy-demand areas facilitates the rational allocation of air conditioning resources. For energy-demand-significant areas, air conditioning operating parameters can be prioritized and optimized to improve cooling or heating efficiency and meet occupant comfort requirements. For energy-insignificant areas, air conditioning power can be appropriately reduced to avoid energy waste. From the perspective of air conditioning system control accuracy, clarifying the energy demand characteristics of different areas enables more precise air conditioning operation and improves indoor comfort. For example, in energy-demand-significant areas, the air conditioning temperature and wind speed can be adjusted promptly based on changes in occupancy density and activity intensity to provide a more comfortable indoor environment.

[0164] Step S2240: extract the energy demand intensity value of each significant energy demand area at each moment in the time dimension to form an energy demand time series of the significant energy demand area.

[0165] Specifically, the energy demand intensity value represents the energy demand intensity for each pixel in the spatiotemporal energy demand coupling map. It reflects the degree of air conditioning energy demand in that area at a given moment. For each significant energy demand area, the energy demand intensity values at different moments are sequentially extracted along the time dimension. For example, in the fresh produce section of a large supermarket, which is a significant energy demand area, the energy demand intensity values for this area are recorded at regular intervals (e.g., 10 minutes) during business hours. These recorded values form a time series of energy demand for that area. This energy demand time series reflects the temporal changes in energy demand in significant energy demand areas. By analyzing this energy demand time series, we can understand the dynamic patterns of energy demand in that area, providing important insights for subsequent energy demand trend forecasting and the development of appropriate air conditioning control strategies. For example, by observing the energy demand time series, we can find that the energy demand intensity in a conference room gradually increases before a meeting, remains high during the meeting, and then rapidly decreases after the meeting. Based on this pattern, the air conditioning operation mode can be adjusted in advance, increasing the cooling or heating power before the meeting and reducing it promptly after the meeting, thereby achieving energy savings while maintaining indoor comfort.

[0166] The purpose of this step is to obtain dynamic energy demand data for areas with significant energy demand, providing information support in the temporal dimension for subsequent analysis and decision-making. Its beneficial effects are reflected in multiple aspects. From the perspective of dynamic control of air-conditioning systems, energy demand time series provide a basis for real-time adjustment of air-conditioning control strategies. Based on real-time changes in energy demand, timely adjustments are made to air-conditioning parameters such as cooling or heating power and wind speed, enabling the air-conditioning system to better adapt to the actual needs of the building and improve energy efficiency. From an energy planning perspective, long-term analysis of energy demand time series can predict energy demand in different time periods, provide reference for energy supply and allocation, and optimize the allocation of energy resources. For example, in commercial buildings, based on the characteristics of energy demand time series in different areas, energy supply plans can be rationally arranged to reduce energy costs.

[0167] In step S2300 , the energy demand significant region is divided into steady-state cores and transient cores, and the non-energy demand significant region is marked as dormant cores.

[0168] The main purpose of step S2300 is to divide the energy demand-significant area into steady-state cores and transient cores according to the energy demand time series characteristics, and mark the non-energy demand-significant area as a dormant core, so that differentiated air-conditioning control strategies can be formulated for different areas in the future to achieve more accurate energy allocation and efficient air-conditioning operation management.

[0169] Furthermore, step S2300 includes:

[0170] Step S2310: Based on the periodic fundamental frequency component, perform autocorrelation analysis on the energy demand time series of the energy demand significant area to obtain the energy demand autocorrelation coefficient and decay time constant τ; where, Indicates that the time series can be The value of Indicates that the time series can be The value of It represents the autocorrelation coefficient of the energy time series at time t and time t+Δt, and Δt represents the time lag interval in the autocorrelation analysis;

[0171] Specifically, the energy demand time series of the energy demand significant area records the energy demand intensity value of the energy demand significant area at each moment, reflecting the change of energy demand in the time dimension. Autocorrelation analysis is a method used to measure the correlation between time series data at different time points. , calculate the autocorrelation coefficient under different time delays Δt , and its calculation formula is:

[0172]

[0173] Among them, Cov is the covariance, which is used to measure the overall error of two variables. Here, it measures the degree of coordinated change of the energy demand time series at time t and t+Δt; Var is the variance, which is used to describe the degree of discreteness of the variable, that is, the fluctuation of the value of the energy demand time series at each time.

[0174] Intuitively, if the energy demand time series remains stable in the long term, it means that the human activities and energy demand in the region are relatively stable, then the larger Under low Δt conditions, the correlation between energy demand values is strong, and ρ will remain close to 1. For example, in an office area with a continuously occupied office and stable occupancy density, the energy demand time series may still have a high autocorrelation coefficient even with a time delay of several hours. This is because in such a stable environment, occupant activity patterns and demand for air conditioning are relatively fixed and do not fluctuate significantly, resulting in a high degree of similarity between energy demand values at different times. Conversely, if the series fluctuates rapidly, indicating frequent changes in occupant activity and energy demand, ρ will drop sharply to 0 at a small Δt. For example, in a conference room, people gather and evacuate quickly at the beginning and end of a meeting, resulting in rapid changes in energy demand intensity. In this case, even with a small time delay, the energy demand values at different times vary greatly, and the autocorrelation coefficient will quickly drop to near 0.

[0175] decay time constant Represents the autocorrelation coefficient Decays from an initial value (usually 1) to The desired number of time steps. The decay of follows the exponential decay law: ; The larger the The slower the decay, the stronger the long-term correlation and trend stability of the series. For example, in the stable office area example above, since the energy demand time series changes slowly, the number of time steps required for the autocorrelation coefficient to decay to 0.368 will be greater, that is, the τ value will be larger. However, in areas with large energy demand fluctuations, such as conference rooms, the autocorrelation coefficient decays quickly to a lower level, and the τ value will be smaller.

[0176] The purpose of this step is to quantify the stability and correlation characteristics of the energy demand time series through autocorrelation analysis, providing a data basis for the subsequent classification of steady-state and transient cores. This beneficial effect is reflected in multiple aspects. From an energy management perspective, analyzing the energy demand autocorrelation coefficient and decay time constant can more accurately understand the changing patterns of energy demand in different regions. For areas with stable energy demand, more stable air conditioning control strategies can be adopted to avoid energy waste caused by frequent equipment adjustments. For areas with large energy demand fluctuations, flexible control schemes can be developed to improve energy efficiency. From the perspective of air conditioning system operational stability, clarifying the characteristics of energy demand time series helps optimize the operating parameters of air conditioning equipment, reduce frequent equipment starts and stops and over-adjustments, extend equipment life, and reduce maintenance costs. For example, in areas with large τ values, the air conditioning system can maintain relatively stable output power for a longer period of time, reducing unnecessary energy consumption and equipment wear. In areas with small τ values, the air conditioning system can adjust power in a timely manner according to rapid changes in energy demand, meeting indoor comfort requirements while avoiding excessive energy consumption.

[0177] Step S2320: Setting the autocorrelation threshold , decay time threshold And the required intensity threshold E0, according to ,τ, and , the energy demand significance region is divided into steady-state core and transient core.

[0178] like and , it is classified as a steady-state nucleus, otherwise it is classified as a transient nucleus.

[0179] Specifically, the autocorrelation threshold , decay time threshold The energy demand intensity threshold E0 is determined based on extensive historical data and actual operational experience. In practice, long-term monitoring and analysis of energy demand across different building areas over varying time periods is necessary, taking into account multiple factors. For example, different functional areas of a building have distinct occupant activity patterns and energy demand characteristics. Energy demand stability and variability vary across office areas, shopping malls, and conference rooms, necessitating different threshold settings. Appropriate thresholds are determined through statistical analysis of extensive time series data on energy demand across these areas, combined with actual air conditioning system performance and energy-saving requirements.

[0180] When the conditions are met and When , the area with significant energy demand is designated as a steady-state core. This means that energy demand in this area remains highly stable over a long period of time, and human activity and energy demand are relatively constant. For the steady-state core area, a constant loading strategy is more appropriate. Because its energy demand trend is highly predictable, calculating the constant loading power by predicting the future mean energy demand (as described in step S3100) enables the air conditioning system to maintain a constant output power in this area, reducing energy loss. This is because constant output power prevents the air conditioner from frequently adjusting the cooling or heating intensity, reduces the number of equipment starts and stops, and reduces the additional energy consumption caused by frequent switching of equipment operating states. It also helps extend the service life of the air conditioning equipment.

[0181] If the above conditions are not met, the area is designated as a transient core. Transient core areas exhibit rapid energy demand fluctuations, with frequent gatherings and dispersals of people, making long-term energy demand prediction difficult. This step aims to rationally delineate areas of significant energy demand using clear quantitative criteria, providing a basis for developing differentiated air conditioning control strategies. The benefits are significant. From an energy-saving perspective, employing different loading strategies for different types of areas can more accurately match air conditioning system output with actual energy demand, avoiding excessive energy consumption in unneeded areas and improving overall energy efficiency. For example, a constant loading strategy in steady-state core areas avoids unnecessary power adjustments, while a flexible transient loading strategy in transient core areas adjusts power based on real-time energy demand changes, reducing energy waste. From the perspective of ensuring indoor comfort, appropriate loading strategies ensure that indoor temperatures in different areas are adjusted promptly based on occupant activity and energy demand fluctuations, providing a more comfortable indoor environment. For example, in transient core areas with dense occupancy and volatile energy demand, timely adjustments to air conditioning power can prevent indoor overheating or overcooling. From an air conditioning system management perspective, this zoning strategy helps optimize overall system operation and improve system stability and reliability. Air conditioning equipment in different areas can be controlled and managed according to their specific characteristics, reducing interference between devices and improving overall system efficiency.

[0182] Non-energy-demanding areas typically refer to those with low human activity and relatively low energy demand, such as corridors and small storage rooms within a building. These areas are often occupied by people for short periods of time or are almost completely empty, resulting in low energy demand for air conditioning. This contrasts sharply with the energy demand characteristics of energy-demanding areas, such as crowded meeting rooms and office areas. Marking these areas as dormant cores provides a unique identifier for air conditioning group control strategies. This marking can be performed at the data level, for example, by adding a specific "dormant core" tag to the energy demand data for these areas in the relevant database or control system. This allows the system to quickly identify and take appropriate measures for these areas during subsequent control strategy formulation and execution. From an energy management perspective, explicitly marking non-energy-demanding areas as dormant cores helps avoid over-allocation of energy to these areas. In previous air conditioning control systems, failure to make this distinction could result in unnecessarily high power output being continuously provided to low-energy-demand areas, resulting in energy waste. By marking these areas as dormant cores, the system can reduce the energy supply to these areas for most of the time, maintaining a lower background load, thereby achieving energy savings. From the perspective of air conditioning system operational stability, setting a background load power for dormant core areas ensures that the system can quickly respond to sudden increases in human activity and maintain a stable indoor temperature. This avoids frequent starts and stops of the air conditioning system due to sudden changes in energy demand, reduces equipment wear and failure, and extends its service life.

[0183] Step S3000 : Based on the division results of steady-state cores, transient cores, and dormant cores, and in combination with the optimized air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air conditioning group is formulated.

[0184] The core of step S3000 is to develop a dynamic energy-saving control strategy for the intelligent AC cluster based on the previously determined steady-state, transient, and dormant cores, combined with the optimized air stagnation compensation factor β'. This step is a key execution link in the entire AC cluster control energy-saving system. By accurately understanding the energy demand characteristics of different areas, it ensures optimal operation of the AC system, achieving the dual goals of energy conservation and improving indoor comfort.

[0185] Furthermore, step S3000 includes:

[0186] Step S3100: predict the energy demand trend of the steady-state core in the future based on the energy demand time series of the steady-state core, obtain the expected value of energy demand intensity, and calculate the constant load power based on the expected value of energy demand intensity. ;

[0187] Specifically, the energy demand time series of the steady-state core exhibits a long-term high autocorrelation coefficient ρ and a large decay time constant τ, indicating a strong trend in energy demand over the coming period. To predict future energy demand trends for the steady-state core, time series forecasting methods such as the autoregressive moving average (ARMA) model or the long short-term memory (LSTM) network can be used. The ARMA model is a commonly used time series forecasting model that combines the characteristics of autoregressive (AR) and moving average (MA). The AR component considers the impact of past values on the current value of the time series, while the MA component accounts for the impact of past forecast errors on the current value. Consider a simple ARMA(p,q) model, where p represents the autoregressive order and q represents the moving average order. In practical applications, appropriate p and q values, as well as corresponding model parameters, need to be determined through model training based on the steady-state core energy demand time series data.

[0188] After obtaining the average future energy demand through these prediction methods, it is used as the constant load power. The calculation basis is: ,in, For the future moments The expected value of energy demand intensity, is the predicted time domain length.

[0189] This method of calculating constant load power has several beneficial effects. From an energy utilization perspective, since energy demand in the steady-state core area is relatively stable, a constant load strategy allows the air conditioning system to maintain a constant output power in this area, avoiding the additional energy consumption caused by frequent power adjustments. For example, without a constant load strategy, the air conditioning system might adjust power at the slightest fluctuation in energy demand, resulting in frequent equipment startups and shutdowns and increased energy consumption. Constant load strategy reduces this unnecessary energy waste and improves energy efficiency. From the perspective of equipment operational stability, stable power output helps extend the service life of air conditioning equipment. Frequent power fluctuations can put significant stress and wear on components such as compressors and fans. Constant load strategy reduces component operating stress, reduces the frequency of equipment failures, and reduces maintenance costs. From the perspective of indoor comfort, stable power output ensures a stable indoor temperature, preventing temperature fluctuations from causing discomfort to personnel. In steady-state core areas such as factory production workshops, a stable temperature environment helps employees maintain a good working state and improves production efficiency.

[0190] Step S3200: construct an isochronous integral curve based on the energy demand time series of the transient core and calculate the instantaneous load power. ;

[0191] Specifically, the autocorrelation coefficient ρ of the transient kernel decays rapidly, and the decay time constant τ is small, which indicates that its energy demand time series exhibits obvious non-stationary characteristics, and it is difficult to make long-term predictions on future energy demand trends. For such areas, if a constant loading strategy is adopted, it is very easy to cause an imbalance in energy supply and demand because it cannot keep up with the rapid changes in energy demand. Therefore, this step uses the equal time integration method to calculate the instantaneous loading power. The calculation formula is ,in, represents the expected energy demand intensity, and the integral upper limit M represents the start and end times of the transient core. This method of calculating instantaneous load power allows real-time tracking of energy demand fluctuations within the transient core, enabling flexible regulation. As energy demand increases, the calculated instantaneous load power increases accordingly, allowing the air conditioning system to promptly increase cooling or heating power to meet occupant comfort needs. As energy demand decreases, the instantaneous load power decreases accordingly, avoiding energy waste. From the perspective of improving energy efficiency, compared to a constant load strategy, this approach dynamically adjusts power output based on actual energy demand, reducing unnecessary energy consumption. Furthermore, by adjusting power in response to energy demand, indoor temperatures remain within a comfortable range, providing a better user experience. For example, in shopping mall promotional areas, this approach can ensure a comfortable temperature even when crowded, enhancing customer satisfaction.

[0192] Step S3300: For the dormant core area, based on the temperature control hysteresis And optimize the hovering compensation factor β' to calculate the background loading power ;

[0193] Specifically, the optimization of the airlock compensation factor β' is obtained by using the adversarial generative network to learn the probability distribution of the airlock compensation factor β and sampling it in the low confidence area of the spatiotemporal energy-demand coupling map. It can maximize the matching of the current energy-demand response lag state and provide an adaptive basis for the dynamic configuration of air conditioning control parameters. The calculation formula of the background load power is ,in It is a pre-set time constant, which is used to make reasonable scale adjustments to the calculation results in the formula to make the calculation of background load power more consistent with the actual situation.

[0194] Matching a certain background load power to dormant core areas is crucial. From the perspective of addressing sudden energy demand, when people suddenly enter these areas with lower energy demand, the background load power acts as a reservoir for rapid energy response, enabling the air conditioning system to quickly provide additional energy to meet occupant comfort needs. For example, when cleaning staff enter a normally unoccupied storage room to clean, the pre-set background load power allows the air conditioning system to quickly raise or lower the temperature in that area (depending on the actual situation), preventing temperature discomfort from impacting work. From the perspective of overall air conditioning system operational stability, the background load power setting prevents sudden increases in energy demand from causing a surge in system stress. Without background load power, the simultaneous and temporary gathering of people in multiple dormant core areas could cause a sudden increase in the air conditioning system load, impacting normal operation. The presence of background load power makes the system more stable in these emergencies, reducing system fluctuations and the risk of failure. From an energy-saving perspective, while setting a background load power for dormant nuclear areas will increase energy consumption to a certain extent, this approach allows for a more rational allocation of energy compared to the energy waste and equipment wear that would otherwise occur if power were significantly increased only when people suddenly gather. Because the background load power is relatively low, it doesn't cause excessive energy consumption most of the time, yet it can be used promptly at critical moments, achieving a balance between energy conservation and meeting sudden energy demands.

[0195] Step S3400, based on constant load power , instantaneous load power and background load power ,construct a multi-level energy flow optimization model for collaborative control of air-conditioning groups;

[0196]

[0197] The objective function J is the total load power of the entire air conditioning group, which is the constant load power of all steady-state cores. , the instantaneous load power of all transient cores and the background load power of all dormant cores Composition, among which Indicates the The constant loading power of a steady-state core at time t is, Indicates the The instantaneous load power of a transient core at time t is, Indicates the The background load power of the dormant cores at time t. are the number of steady-state cores, transient cores, and dormant cores respectively. Constraints include the upper and lower limits of the output power of a single air-conditioning terminal. If the load power exceeds this range, the air conditioning equipment may not work properly or even be damaged; as well as the cooling output of the steady-state core, transient core, and dormant core. , , Meet minimum comfort needs, It is the cooling capacity that meets the minimum comfort requirements of indoor occupants. Indicates the The cooling output of a steady-state core at time t is, Indicates the The cooling output of a transient core at time t is, Indicates the The cooling output of the dormant cores at time t. This multi-objective planning model solves the optimal air conditioning cluster loading plan within the control time domain to minimize total energy consumption. These two constraints ensure that the indoor environment can be maintained within a relatively comfortable temperature range regardless of the circumstances.

[0198] The multi-level energy flow optimization model aims to comprehensively consider the energy demand characteristics of different areas and the operating constraints of air conditioning equipment, achieving overall optimized control of the air conditioning fleet. From an energy-saving perspective, minimizing the total distributed power can avoid unnecessary energy consumption. While ensuring indoor comfort, air conditioning power is rationally allocated across various areas, enabling more efficient energy utilization. For example, distributed power can be appropriately reduced in areas with lower energy demand, while power can be flexibly adjusted based on actual conditions in areas with higher and frequently fluctuating energy demand, achieving energy-efficient operation of the entire air conditioning system. From the perspective of ensuring indoor comfort, the constraints ensure that basic comfort requirements are met regardless of energy demand. Even when energy demand in certain areas fluctuates suddenly, the model's optimization calculations ensure that the temperature throughout the building remains within a comfortable range, enhancing the occupant experience. From the perspective of air conditioning system management and maintenance, the model provides scientific guidance for system operation, enabling managers to rationally schedule air conditioning equipment operation based on the model's calculations, reducing overuse and wear, extending equipment life, and lowering maintenance costs.

[0199] Step S3500: solving a multi-level energy flow optimization model for coordinated control of the air conditioning group to generate an air conditioning load power sequence for each core area;

[0200] Specifically, because energy flow optimization control involves complex time-varying environments and random interference, traditional dynamic programming algorithms face the curse of dimensionality when faced with high-dimensional state spaces, making it difficult to effectively solve. Therefore, this embodiment employs an adaptive dynamic programming (ADP) algorithm to address this challenge. ADP employs approximate value iteration to approximate the Bellman optimal value function as a parameterized approximate value function. Using an incremental learning algorithm for policy iteration, ADP effectively overcomes the curse of dimensionality and solves large-scale energy flow optimization control problems. The Bellman optimal value function describes the maximum cumulative reward that can be obtained by adopting the optimal strategy in a given state. In the context of air conditioning group control, the state can be understood as a comprehensive representation of information such as the current energy demand of different core zones (steady-state cores, transient cores, and dormant cores) and the operating status of the air conditioning equipment. The reward can be set as a quantitative indicator related to energy consumption and comfort, such as lower energy consumption and higher comfort, resulting in a larger reward. However, directly solving the Bellman optimal value function in a high-dimensional state space is computationally intensive and difficult to implement.

[0201] The ADP algorithm approximates the Bellman optimal value function by constructing a parameterized approximation function. This approximation function is typically implemented using a function approximator such as a neural network, such as a multilayer perceptron (MLP). The MLP consists of an input layer, a hidden layer, and an output layer. By adjusting the number of neurons and weights in the hidden layer, complex functional relationships can be fitted. In this scenario, the input layer uses the energy demand intensity of each core area, the temperature control hysteresis, and the current power of the air conditioning equipment as inputs. After nonlinear transformations in the hidden layer, the output layer outputs an approximate estimate of the optimal cumulative reward under the current state. Incremental learning algorithms, on the other hand, gradually update the parameters of the approximation function based on newly acquired data during each iteration. Specifically, at each iteration, the algorithm calculates the error based on the current state and action taken, the actual reward obtained, and the next state. Based on this error, the parameters of the approximation function are adjusted to ensure that the approximation function gradually approaches the true Bellman optimal value function. For example, at a certain moment, based on the current energy demand of each core area, a set of loading power control strategies is adopted, and then the actual energy consumption and indoor comfort changes (i.e., the rewards obtained) are observed. Combined with the energy demand status at the next moment, the error between the current approximate value function and the true optimal value is calculated. The approximate value function (such as the weight of the neural network) is adjusted through techniques such as the backpropagation algorithm to reduce this error.

[0202] Through continuous policy iteration, the ADP algorithm gradually finds a more optimal control strategy. In each iteration, a strategy (i.e., the load power allocation for each core area) is determined based on the current approximate value function. This strategy is then used for actual operation, new data is collected, and the approximate value function is updated with this data to obtain a better strategy. After multiple iterations, the algorithm converges to a near-optimal strategy, resulting in a sequence of air conditioner load power allocations for each core area. The resulting load power allocation sequence is as follows:

[0203]

[0204] in, Indicates the The constant load power value of a steady-state core, Indicates the The instantaneous load power value of a transient core, Indicates the The background loading power value of each dormant core is obtained in this way; the loading power sequence is the optimization result under the consideration of complex time-varying environment, random interference and the overall operation constraints of the air-conditioning group.

[0205] This step has significant benefits. From an energy-saving perspective, the ADP algorithm can find near-optimal load distribution solutions in complex energy flow optimization control scenarios, avoiding the energy waste that can result from traditional algorithms. Compared to simpler control strategies, this algorithm, through continuous learning and optimization, can more accurately adjust the load distribution power based on the real-time energy demand of each zone, minimizing energy consumption while ensuring indoor comfort. From a control accuracy perspective, the ADP algorithm, through approximation of the Bellman optimal value function and strategy iteration, can more accurately match the actual energy demand of each core zone. It can account for various complex factors, such as the mutual influence of energy demand between different zones and the response delay of air conditioning equipment, to generate a load distribution power sequence that better meets actual needs. This helps improve the overall control accuracy of the air conditioning system, ensuring that indoor temperatures remain stable within a comfortable range and enhancing occupant comfort. Traditional algorithms often struggle to adapt to complex environments, as energy flow optimization control involves numerous uncertainties, such as the randomness of occupant movement and fluctuations in outdoor ambient temperature. The ADP algorithm's incremental learning characteristics enable real-time strategy adjustments based on evolving environmental information. For example, when the outdoor temperature changes suddenly or an unexpected gathering of people occurs in a certain area, the ADP algorithm can respond quickly, recalculate and adjust the load power of each core area to ensure the stable operation of the air-conditioning system and indoor comfort.

[0206] In step S3600, the air conditioner loading power sequence is parsed into an execution instruction for controlling the air conditioner terminal, and is sent to the corresponding intelligent controller through the bus network to achieve coordinated control of the air conditioner group.

[0207] Specifically, step S3600 converts the theoretical load power calculated previously into an actual operable control signal, which directly acts on the air-conditioning terminal equipment, thereby achieving precise control of the entire air-conditioning group. First, the air-conditioning load power sequence is parsed. For the steady-state core area, the execution instruction contains the control parameters corresponding to each air-conditioning terminal at each time step, such as Air conditioner at the time The frequency setting value is , the fan speed setting value is , the expansion interval opening setting value is These parameters are based on the constant load power of the steady-state core and the characteristics of the air conditioning equipment. For example, if the constant load power of a steady-state core area is calculated to be , according to the region The cooling or heating capacity curve of an air conditioner can be used to find the corresponding compressor frequency To ensure that the air conditioner can stably output the corresponding power in this area. Fan speed and expansion interval In a similar way, based on the working principle and performance parameters of the air-conditioning equipment, they work together to adjust the cooling or heating effect of the air-conditioning to maintain a stable temperature environment in the area. For the transient core area and the dormant core area, they will also be based on their respective load power sequences P t and P b Determine the appropriate control parameters. Because energy demand in transient core regions fluctuates significantly, control parameters must be dynamically adjusted in real time based on the instantaneous load power to quickly respond to energy demand changes. While energy demand in dormant core regions is relatively low, control parameters must also be appropriately set based on the background load power to cope with sudden energy demand.

[0208] After the control parameters are determined, these execution instructions are sent to the corresponding intelligent controllers via a bus network. A bus network is a communication network used to connect multiple devices and facilitate data transmission. In smart buildings, it efficiently transmits control instructions from the central control system to the intelligent controllers at each air conditioning terminal. This network offers high reliability and fast transmission speeds, ensuring that control instructions are delivered accurately and promptly to the target devices. For example, common fieldbuses (such as Modbus) or industrial Ethernet can meet the data transmission requirements of the air conditioning group control system in smart buildings. Execution instructions containing parameters such as frequency setpoints, fan speed setpoints, and expansion valve opening setpoints are accurately transmitted to the corresponding air conditioning terminals via the bus network. Upon receiving the instructions, the intelligent controller converts these parameters into PWM (pulse width modulation) signals. PWM signals control the average power of the output signal by varying the pulse width. For air conditioning equipment, different PWM signals can control the compressor operating frequency, fan speed, expansion valve opening, and other parameters. For example, by adjusting the PWM signal's duty cycle (i.e., the proportion of a high level within a cycle), the compressor motor's power supply duration can be precisely controlled, thereby adjusting the compressor frequency. Increasing the PWM signal's duty cycle increases the compressor motor's power supply duration, speeding up the compressor and increasing cooling or heating power. Conversely, decreasing the duty cycle reduces the compressor's speed and power. This method drives actuators, enabling precise control of the air conditioning system. Actuators are components in air conditioning equipment that directly perform control actions, such as compressors, fans, and expansion valves. Upon receiving a PWM signal, the compressor adjusts its operating frequency based on the signal's duty cycle, thereby adjusting cooling or heating capacity. The fan adjusts its speed based on the PWM signal, controlling air flow for optimal heat exchange. The expansion valve adjusts its opening based on the PWM signal, controlling the refrigerant flow and optimizing refrigeration cycle efficiency.

[0209] Example 2

[0210] This embodiment, based on the first embodiment, provides an intelligent building air conditioning group control energy-saving method based on multi-source data fusion, including:

[0211] Step S1231, using the non-stationary residual component as the observation sequence, identifying the mutation point in the observation sequence;

[0212] Furthermore, step S1231 includes:

[0213] Step S12311, using the agenda association trajectory as a latent variable, and constructing a causal reasoning network based on the observation sequence and latent variables;

[0214] Specifically, a causal inference network is a model structure used to reveal causal relationships between variables. In this step, the non-stationary residual component is used as the observation sequence, which reflects the short-term fluctuations in population density. Because short-term fluctuations in population density can be influenced by multiple factors, with agenda events being a significant factor, agenda-related trajectories are used as latent variables. For example, in an office setting, if the non-stationary residual component indicates a sudden increase in population density during a certain period of time, and the agenda-related trajectories indicate an impending meeting in the area, this suggests that the increase in population density may be related to the meeting.

[0215] When constructing a causal inference network, specific algorithms and model architectures are employed to establish the association between observation sequences and latent variables. Common causal inference algorithms include Bayesian networks and structural causal models. Taking a Bayesian network as an example, it is a directed acyclic graph (DAG), where nodes represent variables (in this case, individual data points in the non-stationary residual component and related attributes in the agenda-related trajectory), and edges represent dependencies between variables. Through training with a large amount of historical data, the conditional probability distribution between nodes in the network is determined, thereby establishing a model of the association between them. In practical applications, a large amount of data on short-term fluctuations in population density and corresponding agenda events is collected. This data is then used to train the Bayesian network, enabling it to learn the inherent connection between population density fluctuations and agenda events.

[0216] The beneficial effect of this step is that, by constructing a causal inference network, it provides a foundation for anomaly identification. In intelligent building air conditioning group control systems, accurately identifying abnormal changes in occupancy density is crucial for proper air conditioning operation. Failure to accurately determine the cause of changes in occupancy density can lead to inappropriate responses from the air conditioning system, resulting in energy waste or decreased indoor comfort. For example, misinterpreting an increase in occupancy density due to a normal meeting as an abnormality can lead to over-adjustment of the air conditioning system, increasing energy consumption. Conversely, misinterpreting a true abnormality (such as an unexpected large gathering of people) as normal can hinder timely satisfaction of indoor occupants' comfort needs. By establishing associations between observation sequences and latent variables, the causal inference network can help infer whether abnormal changes in occupancy density are caused by unexpected events. This improves the accuracy of the cause identification and provides a reliable basis for the subsequent development of appropriate air conditioning control strategies, thereby enhancing the intelligence level and energy efficiency of the air conditioning group control system.

[0217] Step S13212: Using the causal inference network, identify the mutation points in the observation sequence.

[0218] Specifically, when using a constructed causal inference network to identify mutation points in an observation sequence, this is accomplished by leveraging the correlation between the observation sequence (i.e., the non-stationary residual component) and the latent variable (the agenda correlation trajectory) established by the causal inference network. A causal inference network can be understood as a tool that reveals causal relationships between variables through data mining and model building. Common examples include Bayesian networks and structural causal models. This embodiment uses a Bayesian network as an example. A Bayesian network is a directed acyclic graph in which nodes represent variables—in this case, individual data points in the non-stationary residual component and related attributes in the agenda correlation trajectory. Edges represent dependencies between variables. These dependencies are acquired through training with a large amount of historical data. The training process determines the conditional probability distribution between nodes in the network, thereby establishing a correlation model.

[0219] To identify abrupt changes, we first analyze the data point by point in the observation sequence. Since the non-stationary residual component reflects short-term fluctuations in population density, a data point that deviates significantly from the normal fluctuation range predicted by the causal inference network is likely a mutation point. The specific identification method is as follows: Assume that at time t2, based on the conditional probability distribution established in the causal inference network and the attribute information of the agenda-related trajectory at that time and nearby, the normal value range of the non-stationary residual component is predicted to be [a, b]. If the actual observed value xt2 at time t2 is outside this range, that is, xt2 b, then the point can be preliminarily identified as a mutation point. To ensure accuracy, further confirmation can be performed using statistical tests. For example, using the 3σ criterion, if the observed value falls outside the interval centered at the predicted mean and encompassing three standard deviations, it is more likely to be identified as a mutation point.

[0220] This step has several beneficial effects. First, from the perspective of the operation of the air conditioning group control system, accurately identifying sudden changes in occupant density helps the system more accurately grasp abnormal changes in occupant density. In smart building environments, abnormal changes in occupant density directly affect the indoor heat load and, in turn, the energy demand of the air conditioning system. Failure to accurately identify these sudden changes will result in the air conditioning system being unable to respond promptly to changes in occupant density, causing indoor temperatures to be too high or too low, reducing occupant comfort. For example, in a conference room, if the causal inference network fails to identify the sudden change in occupant density caused by the rapid gathering of people at the beginning of a meeting, the air conditioning system may maintain its original low cooling output, causing the room temperature to be too high at the beginning of the meeting, affecting the attendees' experience. Second, from an energy-saving perspective, accurately identifying sudden changes in occupant density can prevent irrational operation of the air conditioning system and achieve energy conservation. When sudden changes in occupant density are misidentified or not identified, the air conditioning system may over-cool or over-heat, resulting in energy waste. For example, in an office area, if the normal decrease in occupant density caused by the brief absence of people during lunch breaks is misidentified as an abnormal sudden change in occupant density, the air conditioning system may continue to maintain high cooling power during this period, increasing unnecessary energy consumption. By accurately identifying mutation points, the air conditioning system can adjust its operating mode in real time based on actual occupancy density, reducing energy consumption. Furthermore, from the perspective of system intelligence, accurately identifying mutation points helps elevate the intelligence level of the air conditioning group control system. Intelligent buildings require systems to be sensitive to environmental changes and intelligently respond to them, and accurately identifying mutation points is crucial to achieving this goal. By continuously optimizing the accuracy of the causal inference network's mutation point identification, the system can better adapt to complex and changing occupant activity, providing a reliable basis for the subsequent development of more scientific and reasonable air conditioning control strategies, and driving the development of intelligent building air conditioning group control systems to a higher level of intelligence.

[0221] Example 3

[0222] This embodiment provides an intelligent building air conditioning group control energy-saving system based on multi-source data fusion on the basis of embodiment 1. Figure 8 Shown, including:

[0223] Energy-demand coupling map construction module: This module is used to collect heterogeneous trajectories reflecting people's activities in the building, perform cross-modal coupling and abnormal outlier suppression on the heterogeneous trajectories, and obtain the heterogeneous trajectories and observation weight matrix after outlier suppression. Based on the heterogeneous trajectories and observation weight matrix after outlier suppression, a spatiotemporal energy-demand coupling map is constructed.

[0224] Region division module: Based on the spatiotemporal energy demand coupling map, the optimized airlock compensation factor β' is obtained; the spatiotemporal energy demand coupling map is clustered at multiple scales to extract energy demand-significant regions and non-energy demand-significant regions; the energy demand-significant regions are divided into steady-state cores and transient cores, and the non-energy demand-significant regions are marked as dormant cores;

[0225] Energy-saving control strategy formulation module: Based on the division results of steady-state cores, transient cores and dormant cores, combined with the optimization of the air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated.

[0226] In the energy-demand coupling graph construction module, the heterogeneous trajectories collected to reflect the activities of people in the building include:

[0227] Step S1110 , collecting the dynamic time series chain of personnel entering and leaving key nodes to generate a time series displacement trajectory;

[0228] Step S1120: Scan the target area with near-infrared bands to capture the spatial distribution characteristics of human radiation energy levels, construct a thermodynamic energy gradient field reflecting the degree of human gathering, and calibrate the hot spots of human gathering through a gradient clustering algorithm to generate a thermal distribution trajectory;

[0229] Step S1130 , extracting agenda multi-dimensional attribute parameters of the agenda event and generating an agenda association trajectory;

[0230] In step S1140 , the temporal displacement trajectory, the thermal distribution trajectory, and the agenda association trajectory are combined to form a heterogeneous trajectory.

[0231] In the area division module, obtaining the optimized hovering compensation factor β' includes:

[0232] Step S2110: extract the air conditioning operating parameter sequence of the same type of area in the same period in history and calculate the ideal temperature control response curve;

[0233] Step S2120 , comparing the deviation between the actual temperature response in the current low confidence zone and the ideal temperature control response curve, and quantifying the deviation as a temperature control hysteresis ΔT;

[0234] Step S2130: constructing a generative adversarial network (ACGAN) based on the temperature control hysteresis ΔT, and learning the probability distribution of the hovering compensation factor β through the generator G in the generative adversarial network (ACGAN);

[0235] Step S2140: sampling from the probability distribution of β to obtain the optimized hovering compensation factor β'.

[0236] In the energy-saving control strategy formulation module, the formulation of the dynamic energy-saving control strategy of the intelligent air-conditioning group includes:

[0237] Step S3100: predict the energy demand trend of the steady-state core in the future based on the energy demand time series of the steady-state core, obtain the expected value of energy demand intensity, and calculate the constant load power based on the expected value of energy demand intensity. ;

[0238] Step S3200: construct an isochronous integral curve based on the energy demand time series of the transient core and calculate the instantaneous load power. ;

[0239] Step S3300: For the dormant core area, based on the temperature control hysteresis And optimize the hovering compensation factor β' to calculate the background loading power ;

[0240] Step S3400, based on constant load power , instantaneous load power and background load power ,construct a multi-level energy flow optimization model for collaborative control of air-conditioning groups;

[0241] Step S3500: solving a multi-level energy flow optimization model for coordinated control of the air conditioning group to generate an air conditioning load power sequence for each core area;

[0242] In step S3600, the air conditioner loading power sequence is parsed into an execution instruction for controlling the air conditioner terminal, and is sent to the corresponding intelligent controller through the bus network to achieve coordinated control of the air conditioner group.

[0243] The methods and systems of the present application may be implemented in many ways. For example, the methods and systems of the present application may be implemented using software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps used in the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified.

[0244] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0245] The above-described specific embodiments further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent building air conditioning group control energy-saving method based on multi-source data fusion, characterized in that: The method comprises: Heterogeneous trajectories reflecting people's activities in the building are collected, and cross-modal coupling and abnormal outlier suppression are performed on the heterogeneous trajectories to obtain the heterogeneous trajectories after outlier suppression and the observation value weight matrix. Based on the heterogeneous trajectories after outlier suppression and the observation value weight matrix, a spatiotemporal energy demand coupling map is constructed. The method of collecting heterogeneous trajectories reflecting the activities of people in a building includes: collecting a dynamic time series chain of people entering and exiting key nodes to generate a time series displacement trajectory; performing near-infrared band scanning on a target area to capture the spatial distribution characteristics of human radiation energy levels, constructing a thermodynamic energy gradient field reflecting the degree of gathering of people, and calibrating the gathering hotspots of people through a gradient clustering algorithm to generate a thermal distribution trajectory; extracting multi-dimensional attribute parameters of agenda events to generate an agenda association trajectory; and combining the time series displacement trajectory, the thermal distribution trajectory, and the agenda association trajectory to form a heterogeneous trajectory. The cross-modal coupling and abnormal outlier suppression processing of heterogeneous trajectories includes: applying a sliding window Fourier decomposition to the time series displacement trajectory to extract a periodic fundamental frequency component and a non-stationary residual component; mapping the thermal distribution trajectory to a three-dimensional gridded energy field, calculating the local thermal entropy growth rate and the thermal core diffusion coefficient; using the non-stationary residual component as an observation sequence, identifying a mutation point in the observation sequence, determining whether the mutation point is a normal mutation point or an abnormal mutation point, and generating a smoothed and corrected non-stationary residual component and an observation value weight matrix; and forming a heterogeneous trajectory that has undergone outlier suppression processing with the periodic fundamental frequency component, the smoothed and corrected non-stationary residual component, the local thermal entropy growth rate, and the thermal core diffusion coefficient. The method of determining whether a mutation point is a normal mutation point or an abnormal mutation point includes: extracting the timestamp of each mutation point in the mutation point sequence , and retrieve the timestamp in the agenda associated track Agenda events within the time range of Δt' nearby; if there is an agenda event within the range of Δt', then the multi-dimensional attribute parameters of the agenda event are correlated with the rate of change of the personnel density at the mutation point, and the correlation coefficient ρ' is calculated; if the correlation coefficient ρ' exceeds the preset correlation threshold ρ0, then the mutation point is marked as a normal mutation point, otherwise it is marked as an abnormal mutation point; Generating the observation value weight matrix includes: calculating the observation value weight coefficient ω according to the correlation coefficient ρ' of the normal mutation point, and constructing the observation value weight matrix. , where λ is the weight decay factor; The construction of the spatiotemporal energy demand coupling map includes: performing a tensor product operation on the heterogeneous trajectories that have undergone outlier suppression processing to generate a spatiotemporal energy demand coupling tensor; performing dimensionality reduction projection on the spatiotemporal energy demand coupling tensor to form a two-dimensional spatiotemporal energy demand coupling map that matches the building plan layout; assigning credibility attributes to pixels in the two-dimensional spatiotemporal energy demand coupling map according to an observation value weight matrix to obtain the spatiotemporal energy demand coupling map; Based on the spatiotemporal energy demand coupling map, the optimized airlock compensation factor β' is obtained; the spatiotemporal energy demand coupling map is clustered at multiple scales to extract energy demand-significant regions and non-energy demand-significant regions; the energy demand-significant regions are divided into steady-state cores and transient cores, and the non-energy demand-significant regions are marked as dormant cores; The method of obtaining the optimized hovering compensation factor β' includes: identifying the reliability attribute of the spatiotemporal energy coupling map below the preset confidence threshold. The low-confidence areas of the spatiotemporal energy-demand coupling map are marked; the air conditioning operating parameter sequences of the same type of area in the same historical period are extracted to calculate the ideal temperature control response curve; the deviation between the measured temperature response in the current low-confidence area and the ideal temperature control response curve is compared and quantified as the temperature control hysteresis ΔT; the adversarial generative network ACGAN is constructed based on the temperature control hysteresis ΔT, and the probability distribution of the air stasis compensation factor β is learned through the generator G in the adversarial generative network ACGAN; the optimized air stasis compensation factor β' is sampled from the probability distribution of β; Generator Mapping to The probability distribution of : ； in, is the Sigmoid activation function, is the entropy weight of the spatiotemporal energy coupling image pixel, is the sub-region temperature gradient, is the hyperbolic tangent function, is the rate of change of air pressure; According to the division results of steady-state core, transient core and dormant core, combined with the optimization of the air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated.

2. The intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to claim 1 is characterized in that: The identifying of mutation points in the observation sequence comprises: Calculate the change rate of the population density at each moment in the observation sequence and generate a change rate sequence; Normalizing the change rate sequence to obtain a normalized change rate sequence; A mutation point identification threshold ξ is set, and the moment when the normalized change rate exceeds ξ is marked as a mutation point to generate a mutation point sequence.

3. The intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to claim 1 is characterized in that: The identifying mutation points in the observation sequence also includes: using the agenda association trajectory as a latent variable, constructing a causal reasoning network according to the observation sequence and the latent variable; and using the causal reasoning network to identify the mutation points in the observation sequence.

4. The intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to claim 2 is characterized in that: The generating of the smoothed and corrected non-stationary residual component comprises: Correlation analysis is performed based on the agenda correlation trajectory to determine whether the mutation point is a normal mutation point or an abnormal mutation point; the corresponding value of the abnormal mutation point in the observation sequence is marked as an outlier; and the data smoothing method is used to obtain the non-stationary residual component after smoothing correction.

5. The intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to claim 4 is characterized in that: The extraction of energy-demanding significant areas and non-energy-demanding significant areas includes: Local binary pattern encoding is applied to the spatiotemporal energy-demand coupling map to extract the energy-demand gradient features and generate an LBP feature map. Morphological filtering is performed on the LBP feature map. Using the morphologically filtered LBP feature map as input, an adaptive spectral clustering algorithm is used to extract energy-demand-significant regions and non-energy-demand-significant regions.

6. The intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to claim 5 is characterized in that: The division of the energy demand significance region into a steady-state core and a transient core includes: Extract the energy demand time series of the significant energy demand area; Based on the periodic fundamental frequency component, the energy demand time series in the energy demand significant area is analyzed to obtain the energy demand autocorrelation coefficient. and decay time constant τ; where, Indicates that the time series can be The value of Indicates that the time series can be The value of It represents the autocorrelation coefficient of the energy time series at time t and time t+Δt, where Δt represents the time lag interval; Setting the autocorrelation threshold , decay time threshold And the required intensity threshold E0, according to ,τ, and , the energy demand significance region is divided into steady-state core and transient core.

7. An intelligent building air conditioning group control energy-saving system based on multi-source data fusion, which is used to implement the intelligent building air conditioning group control energy-saving method based on multi-source data fusion according to any one of claims 1 to 6, characterized in that: The system comprises: Energy-demand coupling map construction module: This module is used to collect heterogeneous trajectories reflecting people's activities in the building, perform cross-modal coupling and abnormal outlier suppression on the heterogeneous trajectories, and obtain the heterogeneous trajectories and observation weight matrix after outlier suppression. Based on the heterogeneous trajectories and observation weight matrix after outlier suppression, a spatiotemporal energy-demand coupling map is constructed. Region division module: Based on the spatiotemporal energy demand coupling map, the optimized airlock compensation factor β' is obtained; the spatiotemporal energy demand coupling map is clustered at multiple scales to extract energy demand-significant regions and non-energy demand-significant regions; the energy demand-significant regions are divided into steady-state cores and transient cores, and the non-energy demand-significant regions are marked as dormant cores; Energy-saving control strategy formulation module: Based on the division results of steady-state cores, transient cores and dormant cores, combined with the optimization of the air stagnation compensation factor β', a dynamic energy-saving control strategy for the intelligent air-conditioning group is formulated.

Images