A method for controlling energy equipment based on regional population distribution

Abstract

This invention belongs to the field of intelligent energy control technology, specifically relating to a method for controlling energy equipment based on regional population distribution. The method involves acquiring real-time population distribution data, environmental parameter data, historical energy consumption data, and historical population flow characteristic data to predict regional population distribution data at different time periods. Based on historical energy consumption data, historical population flow characteristic data, real-time population distribution data, and environmental parameter data, a dynamic energy demand model is constructed. Combining the regional population distribution data at different time periods, a population density heat map is generated, and changes in regional energy equipment load are predicted. Based on load changes and the population density heat map, a differentiated control strategy is generated to dynamically adjust the operating parameters of the energy equipment. According to the differentiated control strategy, the energy equipment is regulated, and regional environmental feedback data and equipment energy consumption data are collected synchronously after regulation to generate an energy efficiency assessment report. This solves the problems of high response delay and limited parameter matching accuracy in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of intelligent energy control technology, specifically relating to a method for controlling energy equipment based on the distribution of people in a region. Background Technology

[0002] With the large-scale construction of green industrial parks, the market urgently needs intelligent control of energy equipment based on regional population distribution, requiring dynamic matching of population and energy consumption, multi-regional collaborative adjustment, and real-time energy efficiency optimization capabilities. However, existing technologies lack deep integration between the population sensing network and energy control modules, as well as a cross-dimensional collaborative reasoning framework and dynamic energy consumption quantification mechanism. This makes it difficult to support precise equipment control in complex population flow scenarios, and there is a significant gap between these technologies and the technical requirements for constructing real-time population sensing maps and dynamically correcting energy output parameters.

[0003] However, traditional energy equipment control suffers from key flaws: it often employs timed start / stop or fixed threshold adjustment modes, resulting in insufficient responsiveness to dynamic changes in the number of people in a region and susceptibility to disturbances caused by tidal flow of people and spatial function transitions; the models lack dynamic perception and adjustment mechanisms for people, leading to significant inconsistencies in the operating parameters of each equipment control node when faced with fluctuations in population density at different times and in different areas; and the data dimension is limited to single energy consumption monitoring or rough population statistics, failing to form a deeply integrated perception-control network and lacking cross-dimensional attention mechanisms for the collaborative processing of population distribution trajectories and equipment operating status data, resulting in a disconnect between dynamic population information and energy consumption parameters. With the accelerated development of smart buildings and green industrial parks, the market urgently needs intelligent control of energy equipment with high precision and high efficiency. However, existing technologies, due to high response latency and limited parameter matching accuracy, struggle to support efficient equipment control applications in complex population flow scenarios. Summary of the Invention

[0004] This application provides a method for controlling energy equipment based on the regional population distribution, in order to solve the problems of high response delay and limited parameter matching accuracy in the prior art.

[0005] The first aspect of this application provides a method for controlling energy equipment based on regional population distribution, comprising the following steps: acquiring real-time population distribution data, environmental parameter data, historical energy consumption data, and historical population flow characteristic data; predicting regional population distribution data for different time periods based on the historical population flow characteristic data and the real-time population distribution data; constructing a dynamic energy demand model based on the historical energy consumption data, the historical population flow characteristic data, the real-time population distribution data, and the environmental parameter data; generating a population density heat map by combining the regional population distribution data for different time periods and predicting regional energy equipment load changes in real time; generating a differentiated control strategy based on the load changes and the population density heat map; dynamically adjusting the operating parameters of the energy equipment according to the differentiated control strategy; dynamically regulating the energy equipment according to the differentiated control strategy; synchronously collecting regional environmental feedback data and equipment energy consumption data after regulation; training an energy efficiency optimization model based on the environmental feedback data and the energy consumption data; generating an energy efficiency assessment report; and updating the parameter weights of the dynamic energy demand model according to the energy efficiency assessment report.

[0006] Preferably, generating a population density heatmap includes: constructing a regional spatial grid division model; based on the regional spatial grid division model, combining the peak time distribution, average dwell time in historical population flow characteristic data, and the instantaneous population in each region in real-time population distribution data, calculating the population density value of each grid unit through a kernel density estimation algorithm, wherein the weight of historical data is dynamically adjusted with the time decay coefficient; mapping the population density value to a preset color gradient range, overlaying it onto a regional two-dimensional planar map, and generating a population density heatmap that labels the real-time population range, density level, and update timestamp of each grid unit.

[0007] Preferably, a dynamic energy demand model is constructed based on the historical energy consumption data, the historical pedestrian flow characteristic data, the real-time population distribution data, and the environmental parameter data. This includes: constructing a multivariate regression model and a long short-term memory network model; fitting multi-dimensional features using the multivariate regression model based on the historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data to output preliminary energy demand predictions; and capturing the temporal dependence characteristics of energy demand using the long short-term memory network model, fusing the preliminary energy demand predictions, and establishing a dynamic energy demand prediction model.

[0008] Preferably, the dynamic energy demand model formula is as follows:

[0009]

[0010] in, ; Real-time population density; Ambient temperature; For ambient humidity; This is a historical energy consumption benchmark value; These are the weighting coefficients.

[0011] Preferably, predicting the population distribution data in different time periods based on the historical population flow characteristic data and the real-time population distribution data includes: constructing a population flow prediction model; based on the population flow prediction model, combining the time periodicity of the historical population flow characteristic data and the spatial distribution characteristics of the real-time population distribution data, extracting time-dependent features through a gated temporal convolutional network, and capturing the spatial correlation between regions through a graph attention network, to predict the population distribution data and confidence intervals in each region at different time periods.

[0012] Preferably, the energy efficiency optimization model is trained based on the environmental feedback data and the energy consumption data, and an energy efficiency assessment report is generated, including: constructing an energy efficiency optimization model; training the energy efficiency optimization model based on regional environmental feedback data and equipment energy consumption data through a multi-agent collaborative optimization algorithm, and outputting energy efficiency assessment indicators; generating an energy efficiency assessment report based on the energy efficiency assessment indicators, wherein the energy efficiency assessment report includes energy consumption per unit area, energy efficiency deviation value, and optimization suggestions.

[0013] Preferably, real-time prediction of regional energy equipment load changes includes: constructing a load prediction model; based on the load prediction model, extracting historical load fluctuation characteristics using the sliding window method, and predicting load change trends using the gradient boosting tree algorithm; when the load change trend exceeds a preset threshold of the equipment's rated load, triggering a load buffering mechanism and generating a pre-adjustment command.

[0014] A second aspect of this application provides a system for controlling energy equipment based on regional population distribution, comprising: an acquisition module for acquiring real-time population distribution data, environmental parameter data, energy equipment operating status data, historical energy consumption data, and historical population flow characteristic data; a prediction module for predicting regional population distribution data for different time periods based on the historical population flow characteristic data and the real-time population distribution data; a generation module for constructing a dynamic energy demand model based on the historical energy consumption data, the historical population flow characteristic data, the real-time population distribution data, and the environmental parameter data, generating a population density heat map by combining the regional population distribution data for different time periods, and predicting regional energy equipment load changes in real time, generating a differentiated control strategy based on the load changes and the population density heat map, and dynamically adjusting the operating parameters of the energy equipment according to the differentiated control strategy; and an update module for dynamically regulating the energy equipment according to the differentiated control strategy, synchronously collecting regional environmental feedback data and equipment energy consumption data after regulation, training an energy efficiency optimization model based on the environmental feedback data and the energy consumption data, generating an energy efficiency assessment report, and updating the parameter weights of the dynamic energy demand model according to the energy efficiency assessment report.

[0015] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a method for controlling energy equipment based on the distribution of people in a region, as described in the above embodiments.

[0016] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a method for controlling energy equipment based on regional population distribution as described in the above embodiments.

[0017] Therefore, this application has the following beneficial effects: The embodiments of this application integrate multi-source data such as real-time population, environmental parameters, historical energy consumption, and pedestrian flow characteristics. By leveraging historical pedestrian flow and real-time data, it accurately predicts the population in different time periods. Combined with a dynamic energy demand model, it generates population density heatmaps and load forecasts. This differentiated control strategy addresses the problems of delayed response and large parameter matching deviations in traditional timed and fixed-threshold adjustments to tidal population flows, breaking down the disconnect between dynamic population information and energy consumption data. Simultaneously, by training with post-adjustment environmental feedback and energy consumption data, the parameter weights of the dynamic energy demand model are continuously updated. This significantly improves energy utilization efficiency while ensuring regional environmental comfort, effectively filling the gap in deep linkage between population perception and energy control in existing technologies. It maximizes the reduction of ineffective energy consumption while ensuring regional environmental comfort. Thus, it solves the problems of high response delay and limited parameter matching accuracy in existing technologies.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0020] Figure 1 This is a flowchart illustrating a method for controlling energy equipment based on regional population distribution, according to an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of an intelligent control scenario for the central air conditioning and lighting system of a large commercial complex according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of a smart business district pedestrian flow management scenario provided according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of a method for controlling energy equipment based on regional population distribution according to an embodiment of this application;

[0024] Figure 5 This is a schematic diagram of the structure of an energy equipment control system based on regional population distribution, according to an embodiment of this application.

[0025] Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0027] The following description, with reference to the accompanying drawings, illustrates a method for controlling energy equipment based on regional population distribution, according to an embodiment of this application. Addressing the issue of high response latency mentioned in the background section, this application provides a method for controlling energy equipment based on regional population distribution. This method integrates multi-source data on real-time population count, environmental parameters, historical energy consumption, and pedestrian flow characteristics. It accurately predicts the population count in different time periods using historical pedestrian flow and real-time data. Combined with a dynamic energy demand model, it generates a population density heatmap and load forecast. This differentiated control strategy solves the problems of delayed response and large parameter matching deviations in traditional timed and fixed-threshold adjustments to tidal population flows, breaking down the disconnect between dynamic population information and energy consumption data. Simultaneously, by training with post-adjustment environmental feedback and energy consumption data, the parameter weights of the dynamic energy demand model are continuously updated. This significantly improves energy utilization efficiency while ensuring regional environmental comfort, effectively filling the gap in deep linkage between population perception and energy control in existing technologies. It maximizes the reduction of ineffective energy consumption while ensuring regional environmental comfort. Thus, it solves the problems of high response latency and limited parameter matching accuracy in existing technologies.

[0028] Specifically, Figure 1 This is a flowchart illustrating a method for controlling energy equipment based on the distribution of people in a region, as provided in an embodiment of this application.

[0029] like Figure 1 As shown, this method for controlling energy equipment based on regional population distribution includes the following steps:

[0030] In step S101, real-time population distribution data, environmental parameter data, historical energy consumption data, and historical population flow characteristic data are acquired.

[0031] Historical energy consumption data refers to a set of relevant data generated during the energy use process of a specific region or energy equipment over a period of time, including consumption amount, consumption period, and corresponding operating status, which can be used for energy analysis, model building, and control strategy optimization.

[0032] It is understood that the embodiments of this application, by acquiring historical energy consumption data and linking it with real-time population distribution, environmental parameters, and historical pedestrian flow characteristics data, provide historical references for dynamic energy demand models to explore energy consumption adaptation patterns under different population distribution scenarios. This avoids the subjectivity of model parameter settings, improves the accuracy of regional energy equipment load prediction, and provides historical experience for the reasonable matching of equipment operating parameters in the formulation of differentiated control strategies, reducing energy losses caused by parameter deviations. At the same time, it can also serve as benchmark data for the training and evaluation of energy efficiency optimization models. By comparing the energy consumption differences before and after regulation, the energy efficiency improvement effect can be clearly quantified, assisting in the dynamic updating of model parameter weights. This provides key data support for the refined control of energy equipment, the improvement of energy utilization efficiency, and the low-carbon operation of smart buildings and green parks.

[0033] In step S102, based on historical pedestrian flow characteristic data and real-time population distribution data, the population distribution data for different time periods is predicted.

[0034] Among them, regional population distribution data refers to the population size and distribution status of different locations within a specific region in real time or at a specific time period. It can provide basic data for scenarios such as energy equipment control and resource allocation, based on the spatial distribution of personnel.

[0035] It is understood that the embodiments of this application, by combining the regional pedestrian flow time-period patterns and tidal changes inherent in historical pedestrian flow characteristic data with the current pedestrian flow dynamics captured by real-time population distribution data, provide energy equipment control with spatial distribution data that is both regular and real-time. This can effectively compensate for the response lag problem caused by the lack of pedestrian flow prediction in traditional energy equipment control, avoiding insufficient equipment load affecting environmental comfort when the number of people increases sharply or equipment idling and causing energy waste when the number of people decreases sharply. It can also provide key time-period population input for the subsequent construction of dynamic energy demand models, supporting the accurate generation of population density heat maps and the advance prediction of regional energy equipment load changes. This allows differentiated control strategies to plan operating parameters in advance for differences in population distribution at different times, improving the accuracy and foresight of energy equipment control. At the same time, it provides a matching population dimension reference for the training of subsequent energy efficiency optimization models, synergistically improving energy utilization efficiency and regional environmental comfort.

[0036] For example, in the intelligent control of the central air conditioning system in a middle school teaching building, CO2 sensors and temperature and humidity sensors are deployed to collect real-time data on the distribution of people in each classroom area. The system combines historical timetables and the flow of people during breaks to predict changes in personnel density at different times. When it detects that students are returning to the classroom after physical education class, it automatically increases the air supply and cooling power of the corresponding area. For empty classrooms or low-person areas during lunch break, it switches to energy-saving mode to reduce the load. At the same time, it starts pre-cooling 30 minutes in advance based on classroom reservation data. This not only avoids the energy waste caused by traditional 24-hour extensive operation, but also achieves annual electricity savings of over 20,000 kWh and carbon reduction of about 42 tons through dynamic matching of the number of people and cooling demand. This fully demonstrates the supporting role of regional personnel distribution data in the precise control of energy equipment.

[0037] In this embodiment of the application, based on historical pedestrian flow characteristic data and real-time population distribution data, the population distribution data of different time periods is predicted, including: constructing a pedestrian flow prediction model; based on the pedestrian flow prediction model, combining the time periodicity of historical pedestrian flow characteristic data and the spatial distribution characteristics of real-time population distribution data, extracting time-dependent features through a gated temporal convolutional network, and capturing spatial correlations between regions through a graph attention network, to predict the population distribution data and confidence intervals of each region in different time periods.

[0038] Among them, the pedestrian flow prediction model refers to a model tool that relies on historical pedestrian flow characteristics, real-time regional population distribution and other relevant data, and uses algorithms to mine the patterns of pedestrian flow changes and capture dynamic trends, so as to output the scale and distribution of pedestrian flow in a specific time period and specific area in the future, and provide decision support for scenarios such as energy equipment regulation and spatial resource allocation.

[0039] It is understood that the embodiments of this application deeply integrate the temporal periodicity of historical pedestrian flow characteristic data with the spatial distribution characteristics of real-time population distribution data. By using a gated temporal convolutional network to accurately extract the temporal dependence features of pedestrian flow and a graph attention network to effectively capture the spatial correlation between regions, this approach breaks the limitations of the separation of time and spatial dimensions in traditional pedestrian flow prediction. It also outputs population distribution data and confidence intervals for different regions at different time periods. This avoids the one-sidedness of prediction caused by relying solely on historical patterns or real-time data, and improves the accuracy and reliability of population distribution prediction for different regions at different time periods. Furthermore, it provides high-quality time-specific population input for generating population density heat maps and extrapolating changes in energy equipment load for subsequent dynamic energy demand models. This allows differentiated control strategies to be planned in advance based on clear population flow predictions, preventing insufficient equipment load from affecting environmental comfort when population increases sharply and reducing energy waste caused by equipment idling when population decreases sharply, thus synergistically improving energy utilization efficiency and environmental comfort.

[0040] It should be noted that the formula for the pedestrian flow prediction model is as follows:

[0041]

[0042] in, To predict the population distribution in different regions at different time periods; The confidence interval for the predicted result; For prediction layer; For graph attention networks; For gated temporal convolutional networks; For feature splicing operations; Historical pedestrian flow characteristics data; This provides real-time population distribution data. It has a time-periodic characteristic; This is the adjacency matrix between regions.

[0043] Formula for gated temporal convolutional networks:

[0044]

[0045] in, The gated output feature for region i at the current time step t; Use the Sigmoid activation function; To update the weight matrix of the gate; This is a 1D convolution operation; For GTCN input data; The kernel size; To update the gate's bias term; It is the hyperbolic tangent activation function; The weight matrix represents the candidate features; The bias term for candidate features; This represents the historical state characteristics of region i at the previous time step t-1.

[0046] Formula for graph attention networks:

[0047]

[0048] in, The output features of region i at time step t; For splicing operations; For neighboring areas; Gather for neighbors; It is an exponential function; Activate ReLU with leakage; The attention vector for the k-th attention head; Let be the linear transformation matrix of the k-th attention head; For the number of attention heads; For all neighboring areas; The input features for region i at time step t; The input features are those of the neighboring region j at time step t; The input features are the features of all neighboring regions k at time step t.

[0049] For example, such as Figure 2 As shown, in the intelligent control scenario of the central air conditioning and lighting system of a large commercial complex, staff use a pedestrian flow prediction model to predict the number of people in each area: First, the historical pedestrian flow characteristic data of the mall over the past 3 months (including the time periodic patterns such as the peak hours of the catering area from 10 am to 12 pm on weekdays and the peak hours of the retail area from 2 pm to 4 pm on weekends) and the real-time collected pedestrian distribution data of each floor and each shop area (such as the spatial distribution characteristics obtained by AI counting through cameras) are input into the model. The model uses a gated temporal convolutional network to extract the time dependence features of pedestrian flow at different times (such as the difference between peak pedestrian flow on weekdays and weekends). At the same time, it uses a graph attention network to capture the spatial correlation of pedestrian flow between adjacent floors and connected shops (such as the relationship between the increase in pedestrian flow at the first-floor entrance and the increase in pedestrian flow at the second-floor escalator entrance). Finally, it outputs the pedestrian distribution data and confidence intervals of each area (such as the catering area, retail area, and underground parking garage) in the next 6 hours. Based on this prediction, the mall's energy control system was adjusted in advance: when it was predicted that the flow of people in the food and beverage area would increase by 30% in one hour, the cooling power and fresh air volume of the corresponding area were increased 15 minutes in advance; when it was predicted that the flow of people in the underground parking garage would drop sharply to 20% after 8 pm, the lighting of non-main passageways was automatically switched to an alternate lighting mode. This not only avoided the problems of insufficient environmental comfort during peak hours and energy waste during off-peak hours in the traditional "one-size-fits-all" control, but also reduced the mall's average monthly energy consumption by 18% while ensuring the experience of customers and staff at different times.

[0050] In step S103, a dynamic energy demand model is constructed based on historical energy consumption data, historical population flow characteristic data, real-time population distribution data, and environmental parameter data. Combined with population distribution data in different time periods, a population density heat map is generated, and the load changes of regional energy equipment are predicted in real time. Based on the load changes and the population density heat map, a differentiated control strategy is generated, and the operating parameters of the energy equipment are dynamically adjusted according to the differentiated control strategy.

[0051] Among them, the dynamic energy demand model refers to a model tool that relies on dynamic variables such as regional population distribution data, real-time energy consumption data, and environmental parameters output by population flow prediction. Through algorithms, it analyzes and extrapolates changes in energy consumption demand in different time periods and regions in real time, outputs accurate time-specific and regional energy demand data, and provides a scientific basis for dynamic regulation of energy equipment and optimal allocation of resources.

[0052] It is understood that the embodiments of this application, by integrating historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data, break through the limitations of fragmented multi-dimensional data in traditional energy control. Simultaneously, by combining the population distribution data of different time periods from the pedestrian flow prediction output, it can not only intuitively present regional population distribution differences through the generation of population density heat maps, but also accurately predict changes in regional energy equipment load in real time. This provides a quantitative basis for the formulation of differentiated control strategies, effectively solving the problems of insufficient energy supply in densely populated areas and energy waste in sparsely populated areas caused by traditional control. The strategies generated based on this model can directly guide the dynamic adjustment of energy equipment operating parameters, ensuring a high degree of matching between equipment operating status and actual regional energy demand. This maximizes energy reduction while ensuring environmental comfort, and also provides a basis for subsequent optimization of model parameters and improvement of long-term energy efficiency through environmental feedback and energy consumption data.

[0053] For example, in the energy management scenario of a smart office building, staff members constructed a dynamic energy demand model based on the building's historical energy consumption data over the past year (including monthly peak energy consumption of air conditioning, lighting, and fresh air systems on each floor, and energy consumption fluctuation patterns at different times), historical human flow characteristic data (such as the pattern that the office area is full from 9:00 to 12:00 on weekdays, the human flow in the lunch break area increases by 30% from 12:00 to 14:00, and the number of people in each area drops sharply to less than 10% after 18:00), combined with real-time data on the distribution of people on each floor (obtained through AI camera counting, such as 85 people in the office area on the 3rd floor and 12 people in the conference room on the 5th floor at 9:00) and environmental parameter data (real-time indoor temperature and humidity, outdoor light intensity, and temperature). The model first combines the population distribution data of different time periods from the traffic flow prediction output (e.g., predicting that the number of people in the 3rd-floor office area will increase to 95 at 10:00, the number of people in the 5th-floor conference room will remain at 12, and the number of people in the 1st-floor lunch break area will reach 60 at 13:00) to generate an intuitive population density heat map—marking the 3rd-floor office area (high-density area) in dark red, the 1st-floor lunch break area (medium-density area) in orange-red, and the 5th-floor conference room (low-density area) in light blue. Then, based on the correlation pattern in historical energy consumption data that "for every 10 additional people in the office area, the air conditioning load increases by 6% and the lighting load increases by 4%", the model adds a correction coefficient of 8% for the real-time outdoor temperature of 32℃, predicting in real-time that at 10:00 the air conditioning load on the 3rd floor will reach 82%, the lighting load will reach 75%, and the air conditioning load in the 5th-floor conference room will only need to be increased by 8%. At 13:00, the air conditioning load in the first-floor nap area reached 58%, with 35% for lighting and 20% for air conditioning. Based on these load changes and the heat map zoning, the model generated differentiated control strategies: for the high-density area on the third floor, the strategy was to set the air conditioning to 25°C, increase the fresh air volume by 25%, and turn on all lights; for the low-density area on the fifth floor, the strategy was to set the air conditioning to 26°C and turn on only the local lighting in the conference room; for the nap area on the first floor, the air conditioning was set to 26°C 15 minutes in advance, and the fresh air volume was adjusted to medium. Finally, the model automatically and dynamically adjusted the operating parameters of the energy equipment in the corresponding areas based on this strategy, such as increasing the air conditioning compressor frequency on the third floor from 50Hz to 65Hz and the fresh air fan speed from 1200r / min to 1500r / min, while reducing the air conditioning compressor frequency in the fifth-floor conference room to 30Hz and turning off the three surrounding lights. According to actual operation statistics, the average monthly energy consumption of the office building in this scenario is reduced by 18%, while the indoor temperature fluctuation in each area is controlled within ±0.5℃, and the comfort satisfaction of the staff is improved to 92%. This fully demonstrates the role of the dynamic energy demand model in accurately matching energy supply with actual demand and balancing energy efficiency and comfort.

[0054] In this embodiment of the application, generating a heatmap of people density includes: constructing a regional spatial grid division model; based on the regional spatial grid division model, combining the peak time distribution, average dwell time in historical pedestrian flow characteristic data, and the instantaneous number of people in each region in real-time people distribution data, calculating the people density value of each grid unit through a kernel density estimation algorithm, wherein the weight of historical data is dynamically adjusted with the time decay coefficient; mapping the people density value to a preset color gradient range, overlaying it onto a regional two-dimensional planar map, and generating a heatmap of people density that labels the real-time people range, density level, and update timestamp of each grid unit.

[0055] Among them, the regional spatial grid division model refers to a technical model that, based on specific objectives and rules, divides a specific regional space into several grid units with relatively uniform attributes to support spatial analysis, simulation, or management decision-making.

[0056] It is understood that the embodiments of this application provide a standardized spatial basis for generating population density heatmaps by using a regional spatial grid partitioning model: by decomposing complex areas into grid units with uniform attributes, it accurately receives and corresponds to historical population flow characteristic data and real-time population distribution data, giving population flow data of different dimensions a unified spatial calculation carrier, ensuring that the kernel density estimation method can efficiently and accurately calculate population density values ​​within each unit; and by relying on the regularity of the grid units, the mapping between density values ​​and color gradients is more intuitive, ensuring that after being superimposed on a two-dimensional map, the real-time population range, density level, and timestamp information of each grid accurately correspond to the spatial location, so that the heatmap can not only clearly present the differences in population flow density within the area, but also provide fine-grained and locationable spatial basis for subsequent decisions such as population flow control and resource allocation, avoiding data calculation deviations and heatmap information distortion caused by ambiguous spatial partitioning.

[0057] It should be noted that the regional spatial grid partitioning model sets the planar coordinate boundary of the target region as follows: Lateral (x-axis, e.g., east-west direction of the region): minimum coordinates Maximum coordinates Vertical axis (y-axis, such as the north-south direction of a region): minimum coordinates Maximum coordinates Total number of horizontal grids Total number of vertical grids .

[0058] Horizontal width of a single grid cell: Vertical height of a single grid cell: h .

[0059] The formula for spatial range is:

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] in, For the row index of the grid; For the column index of the grid; The left boundary (x-axis); The right boundary (x-axis); The lower boundary (y-axis); The upper boundary (y-axis); This is a unique ID for the grid cell.

[0066] Kernel density estimation algorithm formula:

[0067]

[0068] in, Let be the two-dimensional probability density function to be estimated; For sample size; For bandwidth; For kernel functions; Let i be the spatial coordinates of the i-th person's flow event; The target space coordinates; Let be the total number of people in the i-th person flow event.

[0069] For example, such as Figure 3As shown, in the scenario of refined pedestrian flow management in a smart business district, the operator relies on a regional spatial grid division model to first divide the entire business district into regular grid units of 5×5 meters. Taking the upper left corner of the business district as the origin, each grid is assigned a horizontal sequence number i and a vertical sequence number j. The horizontal division is carried out sequentially according to the sequence number, with the left and right boundaries of each grid spaced 5 meters apart, and the same applies to the vertical division. At the same time, a unique identifier "Gij" is set for each grid to achieve precise definition of spatial units. In the data fusion stage, not only is the distribution of historical peak pedestrian flow periods (such as peak data from 12-14 on weekdays and 16-20 on weekends) integrated, but the average dwell time is also dynamically weighted using a time decay coefficient (0.6 for data in the last 7 days and 0.4 for data in the last 30 days). Combined with the real-time instantaneous number of people in the grid updated every 5 minutes, a multi-dimensional pedestrian flow dataset is formed. Subsequently, using a kernel density algorithm, the average pedestrian density of the grid and its three adjacent grids is calculated to smooth out fluctuations. The density values ​​are mapped using a color gradient from light blue (<5 people / ㎡, low density) to yellow (5-15 people / ㎡, medium density) to dark red (>15 people / ㎡, high density), and overlaid with a 2D map of the business district to generate a heatmap that labels the real-time pedestrian range (e.g., "28-45 people"), density level, and minute-level update timestamps (e.g., "2025-08-26 15:32"). In practical applications, beverage shops deploy two additional cashiers based on the high-density signal at midday in their grid (e.g., G-8-5), while security adjusts patrol forces in the high-density areas of the core area (G-5-3 to G-7-6) on weekends, providing spatially accurate pedestrian flow data to support operational decisions.

[0070] In this embodiment, a dynamic energy demand model is constructed based on historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data. This includes: constructing a multivariate regression model and a long short-term memory network model; fitting multi-dimensional features using the multivariate regression model based on historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data to output preliminary energy demand forecasts; and capturing the temporal dependence characteristics of energy demand using the long short-term memory network model, fusing the preliminary energy demand forecasts, and establishing a dynamic energy demand forecasting model.

[0071] Among them, the multivariate regression model is a statistical model that introduces multiple independent variables at the same time, quantifies their impact on a single dependent variable by constructing statistical relationships, and then realizes the prediction of the dependent variable or the correlation analysis between variables.

[0072] It is understood that this application's embodiments, by employing a multivariate regression model, effectively integrate historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data. This overcomes the limitations of single-variable analysis and, by constructing statistical relationships, accurately quantifies the influence of different factors on the dependent variable of energy demand. This avoids prediction bias caused by the omission of key influencing factors and clearly presents the contribution weight of each variable. Simultaneously, the fitting output of multi-dimensional features provides reliable preliminary predictions for dynamic energy demand forecasting. This not only ensures the comprehensiveness and accuracy of the prediction results but also lays a data foundation for subsequent long short-term memory network models to capture time-dependent features and perform multi-model fusion. This enables operators to more scientifically predict energy demand fluctuations, providing precise data support for energy allocation, cost control, and efficient supply in commercial districts, thereby reducing energy waste and supply shortages.

[0073] It should be noted that the formula for the multivariate regression model is as follows:

[0074]

[0075] in, This represents the energy demand value. The bottom line for basic energy consumption; , , , This is an indicator of the strength of the impact of each independent variable on energy demand. Historical pedestrian flow characteristics data; This provides real-time population distribution data. This refers to ambient temperature data. This refers to ambient humidity data. This is the error term.

[0076] Long Short-Term Memory Network Model Formula:

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] in, Output for the forget gate; For activation functions; Here is the weight matrix for the forget gate; For the model in Forecast characteristics of energy demand in the business district at any time; for Input characteristics at any given time; For the bias term of the forget gate; For input gate output; This is the weight matrix of the input gate; This is the bias term for the input gate; This represents the current state of the candidate cells. This is the weight matrix of the currently input features; The bias term for the current input feature; For activation functions; This represents the current state of the cell. Output gate output; This is the weight matrix of the output gate; This is the bias term for the output gate; This is the final output of the LSTM at the current moment; This represents the cell state at the previous moment.

[0084] In this embodiment of the application, the dynamic energy demand model formula is as follows:

[0085]

[0086] in, ; Real-time population density; Ambient temperature; For ambient humidity; This is a historical energy consumption benchmark value; These are the weighting coefficients.

[0087] It is understood that the embodiments of this application utilize a dynamic energy demand model, based on historical energy consumption, historical pedestrian flow characteristics, real-time population distribution, and environmental parameters. By combining the population distribution in different time periods to generate population density heat maps, the model accurately and in real-time predicts changes in regional energy equipment load, generates differentiated control strategies, and dynamically adjusts equipment operating parameters. This precisely matches energy demand with equipment operation, breaks down data silos, establishes a dynamic control link, improves energy utilization efficiency to reduce energy waste, optimizes equipment load allocation to extend equipment lifespan, and aligns with population flow and environmental needs to improve regional comfort. Simultaneously, it reduces manual control costs and promotes the development of energy management towards intelligence and refinement.

[0088] For example, a certain plaza, as a regional commercial complex, relies on a dynamic energy demand model to achieve refined energy management. This model first integrates historical energy consumption data from the past three years (covering core equipment such as air conditioning, lighting, and elevators), and pedestrian flow characteristics data for different time periods (e.g., peak pedestrian flow in the retail area from 10:00-12:00 on weekdays, and peak pedestrian flow in the catering area from 18:00-20:00 on weekends). It also integrates real-time pedestrian distribution data for each floor (collected through an infrared sensing system) and environmental parameters (outdoor temperature and humidity, indoor CO2 concentration), automatically generating a visualized heat map of pedestrian density. This allows for accurate prediction of changes in the load on energy equipment in each area—for example, predicting the lunchtime peak in the catering area one hour in advance and increasing the air conditioning cooling load by 30% from the baseline; and quickly reducing the lighting power in public areas by 50% when pedestrian flow drops sharply after a movie. Simultaneously, the model dynamically adjusts its operating strategy based on peak and off-peak electricity prices, pre-stocking air conditioning cooling capacity during off-peak periods (23:00-07:00 the next day) and prioritizing the use of photovoltaic power stations to supplement power supply during peak periods (10:00-14:00 and 18:00-22:00). Through this solution, the plaza's annual energy consumption has been reduced by 17%, daily electricity savings reached 7,000 kWh during the summer cooling season, equipment failure rate decreased by 12%, indoor temperature fluctuations were controlled within ±0.5℃, and customer comfort and satisfaction increased by 8%, achieving both energy cost savings and ensuring a good business operation experience.

[0089] In this embodiment of the application, real-time prediction of regional energy equipment load changes includes: constructing a load prediction model; based on the load prediction model, extracting historical load fluctuation characteristics using the sliding window method, and predicting load change trends using the gradient boosting tree algorithm; when the load change trend exceeds a preset threshold of the equipment's rated load, triggering a load buffering mechanism and generating a pre-adjustment command.

[0090] Among them, the load forecasting model is a model that predicts the energy load of energy equipment or regions in a specific future period by analyzing historical load data and related influencing factors such as environment, population flow, and economy, and using statistical and machine learning algorithms. It provides data support for energy dispatch, equipment operation optimization and resource allocation.

[0091] It is understood that the embodiments of this application utilize a load forecasting model, relying on historical load data and influencing factors such as environment, population flow, and economy, to accurately extract historical load fluctuation characteristics using the sliding window method, and then efficiently predict the load change trend of regional energy equipment based on the gradient boosting tree algorithm, providing accurate data support for energy dispatching and equipment operation optimization; it can also promptly trigger a load buffering mechanism and generate pre-adjustment instructions when the load change trend exceeds the preset threshold of the equipment's rated load, effectively preventing equipment damage due to overload and extending its service life, while ensuring stable energy supply through advance prediction and adjustment, reducing energy waste, improving energy utilization efficiency, reducing the cost and error of manual control, and helping regional energy management achieve refined and intelligent upgrades.

[0092] It should be noted that the load forecasting model formula is as follows:

[0093]

[0094] in, Predicted load of regional energy equipment at time t; Let be the input feature vector at time t; The baseline function is the base load when there are no characteristic inputs; Let m be the basis function, characterizing the features. Nonlinear or complex effects on the load; Let be the weight coefficients of the m-th decision tree; This represents the total number of decision trees; For iterative indexing.

[0095] Sliding window method formula:

[0096]

[0097]

[0098] in, For a sliding window at time t; The actual load value at time t; For window size; A sliding window for time point t+1; For time points Data points; For time points Data points; For data points at time t−K+1; The data points are at time point t−1.

[0099] Gradient boosting tree algorithm formula:

[0100]

[0101]

[0102]

[0103]

[0104] in, Let be the final predicted value for the i-th sample; The total number of trees; Let be the prediction function for the m-th regression tree; Let be the input feature vector of the i-th sample; Let m be the set of parameters for the m-th regression tree; For iterative indexing; The loss function; This represents the true value of the i-th sample. This is the intermediate predicted value after the m-th iteration; Let be the residual of the i-th sample in the m-th iteration; For the loss function L, the intermediate predicted values The partial derivatives; This represents the difference between the actual value and the current model prediction. This is the intermediate predicted value after the (m-1)th iteration; The learning rate; Let be the prediction function for the m-th regression tree.

[0105] When the load change trend exceeds the preset threshold of the equipment's rated load, the load buffering mechanism is triggered, generating a pre-adjustment command. After the gradient boosting tree model uses the historical load fluctuation characteristics extracted by the sliding window to predict the load change trend of regional energy equipment in the future period, the system will immediately compare the predicted load value with the preset threshold of the equipment's rated load (usually set by considering factors such as equipment safety redundancy and optimal energy cost strategies, such as taking 85%~95% of the rated power, with a buffer space reserved). If the predicted load exceeds the threshold, the load buffering mechanism is immediately triggered: on the one hand, energy reserves are called up (such as energy storage batteries discharging to supplement power, and waste heat recovery devices being put into operation to adjust heat energy) to quickly smooth out the load peak; on the other hand, dynamic control is implemented on core energy equipment (such as reducing the operating load of air conditioning units, adjusting transformer taps to optimize voltage), and at the same time, power control commands are sent to flexible power terminals (smart lighting, adjustable elevators, etc.) to guide them to temporarily reduce load during peak load periods. Ultimately, based on the above strategy, pre-adjustment instructions are generated, covering equipment operating parameters (such as power settings and start-stop sequences) and energy allocation schemes (such as energy storage charging and discharging power and distributed power output adjustment). These instructions are then issued and executed in real time through automated control. This approach avoids equipment burnout due to overload and extends maintenance cycles, while also optimizing energy resource allocation and reducing peak-hour energy procurement costs. It ensures the sustainability of regional energy supply from three dimensions: equipment safety, energy efficiency improvement, and system stability.

[0106] For example, taking a smart industrial park as an example, its load forecasting model integrates historical energy consumption, real-time temperature and humidity, production line operating conditions, and personnel flow data. It uses the sliding window method to uncover load fluctuation patterns and employs a gradient boosting tree algorithm to predict the load trends of each workshop and utility (such as refrigeration stations and substations). When it predicts that the load of a workshop's refrigeration unit is about to exceed the 10% threshold of its rated capacity, the system automatically triggers a buffering mechanism: first, it calls upon the ice storage system to release cooling capacity to smooth out peak loads; second, it simultaneously adjusts the speed of variable frequency equipment in the workshop to optimize energy consumption; and third, it pushes electricity procurement strategies (such as locking in off-peak electricity prices in advance) to the energy management platform. Operational verification shows that the park's load forecasting error is controlled within 8%, equipment overload risk is reduced by 23%, and annual energy cost savings exceed one million yuan, effectively achieving a synergy between precise energy scheduling and cost reduction and efficiency improvement.

[0107] In step S104, energy equipment is dynamically regulated according to a differentiated control strategy. Regional environmental feedback data and equipment energy consumption data are collected synchronously after regulation. An energy efficiency optimization model is trained based on the environmental feedback data and energy consumption data. An energy efficiency assessment report is generated, and the parameter weights of the dynamic energy demand model are updated based on the energy efficiency assessment report.

[0108] Among them, the energy efficiency optimization model is a model that analyzes the operating data of the energy system, equipment characteristics and environmental factors, and uses mathematical modeling and intelligent algorithms to construct the correlation between energy consumption and efficiency, and outputs strategies such as equipment regulation and energy allocation to achieve the goal of improving energy utilization efficiency and optimizing costs.

[0109] It is understood that, by employing an energy efficiency optimization model, this application can construct a dynamic correlation between energy consumption and efficiency based on regional environmental feedback and equipment energy consumption data collected after the implementation of differentiated control strategies. This allows for the precise output of optimized strategies for equipment regulation and energy allocation. The generated energy efficiency assessment report can not only quantify energy utilization levels and pinpoint energy-saving shortcomings, but also update the parameter weights of the dynamic energy demand model in reverse, driving more accurate demand forecasting. This not only improves energy utilization efficiency and reduces energy costs, but also enables model self-optimization through data-driven approaches, strengthening the intelligence and adaptability of energy regulation and promoting the upgrade of energy management from static planning to dynamic and precise regulation.

[0110] For example, taking a smart commercial complex in a city's core business district as an example, it relies on an energy efficiency optimization model to carry out refined energy management: The model first integrates real-time data collected after the implementation of differentiated control strategies—including environmental feedback data from each floor (such as temperature and humidity in the retail area, CO2 concentration in the catering area, and illuminance in the cinema area) and equipment energy consumption data (air conditioning unit power, smart lighting power consumption, and elevator operating energy consumption). Then, it combines historical energy consumption trends, equipment rated parameters, and other information to construct a correlation model of "energy consumption-efficiency-environmental comfort" through machine learning algorithms. Based on this, the model outputs targeted optimization strategies: during weekday midday peak traffic periods, it automatically lowers the air conditioning cooling temperature in the catering area by 1°C and adjusts the lighting brightness to 80%, while reducing the number of idle elevators; during off-peak hours at night, it shuts down some air conditioning terminals in non-core areas, maintaining only basic ventilation. Simultaneously, the model will continuously iterate and train using newly collected feedback data, generating monthly energy efficiency assessment reports that clearly mark key indicators such as "12% energy saving in air conditioning system" and "8% reduction in lighting energy consumption." The assessment results will be transformed into the basis for adjusting parameters of the dynamic energy demand model (such as increasing the weight of "human density" in air conditioning load prediction), achieving the dual goals of improving energy efficiency and ensuring user experience.

[0111] In this embodiment of the application, an energy efficiency optimization model is trained based on environmental feedback data and energy consumption data to generate an energy efficiency assessment report, including: constructing an energy efficiency optimization model; training the energy efficiency optimization model through a multi-agent collaborative optimization algorithm based on regional environmental feedback data and equipment energy consumption data, and outputting energy efficiency assessment indicators; generating an energy efficiency assessment report based on the energy efficiency assessment indicators, wherein the energy efficiency assessment report includes energy consumption per unit area, energy efficiency deviation value, and optimization suggestions.

[0112] Among them, multi-agent cooperative optimization algorithm refers to an algorithm that enables multiple intelligent agents with autonomous decision-making and information perception capabilities to build coordination strategies and computational logic at the algorithm level through collaborative mechanisms such as information interaction, task division or resource allocation, so as to jointly achieve the overall or local optimization goal of the system.

[0113] It is understood that the embodiments of this application, by employing a multi-agent collaborative optimization algorithm, enable agents with different functional roles to break down data silos through real-time information interaction. They collaborate based on the characteristics of different energy devices or functional zones within a region, jointly providing multi-dimensional, scenario-based training data and optimization directions for the energy efficiency optimization model. This ensures the model can accurately output core evaluation indicators such as energy consumption per unit area and energy efficiency deviation. Parallel processing significantly improves data processing and model training efficiency, avoiding latency issues when a single algorithm processes massive amounts of data. The collaborative model allows for more detailed data analysis across various dimensions, effectively reducing energy efficiency assessment errors. Based on the deep adaptation of different agents to specific regions and devices, the optimization suggestions in the generated evaluation report are more targeted, driving continuous performance optimization of the model and laying a solid foundation for subsequent accurate updates to dynamic energy demand model parameters and the realization of efficient regional energy management.

[0114] It should be noted that the formula for the multi-agent cooperative optimization algorithm is as follows:

[0115]

[0116]

[0117]

[0118]

[0119]

[0120] in, Let be the reward value of the i-th agent at time t; The actual energy consumption of the region managed by the i-th agent at time t; Baseline energy consumption; Weighting coefficient for energy conservation contribution; The actual load deviation of the i-th agent; For target load deviation; This is the load deviation penalty coefficient; Let i be the set of neighboring smart agents of the i-th smart agent; Let i be the environmental feature vector of the i-th agent. Let j be the environmental feature vector of the j-th agent; The environmental feature similarity between agents i and j; Environmental consistency reward coefficient , Indexing for intelligent agents; Let be the policy gradient of the i-th agent; These are the policy parameters for the i-th agent; This represents the long-term cumulative reward of the i-th agent. Calculate the expected value over time t; For agent i in policy Below, select the control action at time t. The probability of; The control action at time t; In the previous iteration, agent i selected an action. The old probability; The dominant function at time t; This represents the logarithmic gradient of the policy parameters with respect to the action probabilities; Let be the feature vector of agent i after the (k+1)th iteration; This represents the number of iterations. The information interaction weights between agents i and j; Let be the feature vector of agent j after the kth iteration; For the global decision vector Find the minimum value; The set of control actions for all intelligent agents; The total number of agents; Energy consumption weighting coefficient; Let be the actual energy consumption of agent i at time t; This is the load deviation weighting coefficient; The actual load deviation of agent i; For target load deviation The penalty is the square of the load deviation. Let be the policy parameters of agent i after the (k+1)th iteration; Let be the old policy parameters of agent i in the k-th iteration; The learning rate; Let be the policy gradient of agent i; Let be the coordination coefficient between agents i and j; Let be the policy parameters of agent j in the k-th iteration; Let be the policy parameters of agent i in the k-th iteration.

[0121] Energy efficiency optimization model formula:

[0122]

[0123]

[0124]

[0125]

[0126]

[0127] in, For decision variables; Let be the state vector of the first device at time t; Let N be the state vector of the Nth device at time t; Let i be the state vector of the i-th device at time t; To optimize the total number of time steps in the cycle; The total number of devices; Indexed by time step; For device indexing; Let i be the power model function for the i-th device; The indoor environmental conditions at time t; The outdoor environmental conditions at time t; The actual time length for each time step; Let be the energy unit price at time t; This represents the lower limit permissible by indoor environmental conditions. This represents the upper limit allowed by the indoor environmental conditions. This represents the lower limit of the equipment's operating status. This represents the upper limit of the device's operating status; The maximum allowable change; Let i be the state vector of the i-th device at time t+1; The indoor environmental state at time t+1; This is a system dynamics model.

[0128] This application proposes a method for controlling energy equipment based on regional population distribution. By integrating multi-source data on real-time population count, environmental parameters, historical energy consumption, and pedestrian flow characteristics, it accurately predicts the population size in different time periods using historical and real-time data. Combined with a dynamic energy demand model, it generates population density heatmaps and load forecasts. This differentiated control strategy addresses the problems of delayed response and large parameter matching deviations in traditional timed and fixed-threshold adjustments to tidal population flows, breaking down the disconnect between dynamic population information and energy consumption data. Simultaneously, by training with post-adjustment environmental feedback and energy consumption data, the method continuously updates the parameter weights of the dynamic energy demand model. This significantly improves energy utilization efficiency while ensuring regional environmental comfort, effectively filling the gap in deep integration of population perception and energy control in existing technologies. It maximizes the reduction of ineffective energy consumption while ensuring regional environmental comfort. Therefore, it solves the problems of high response delay and limited parameter matching accuracy in existing technologies.

[0129] The following will illustrate a method for controlling energy equipment based on the distribution of population in a region through a specific embodiment, such as... Figure 4 As shown, it includes:

[0130] Using a large commercial complex located in the city's core business district as an example, this complex has a total construction area of ​​80,000 square meters and six floors. The business layout combines consumption and office functions, specifically divided into three core functional areas: the commercial area covers floors 1-4, with the 1st floor being the atrium and beauty area (3,000 square meters, a core area for weekend foot traffic); the 2nd and 3rd floors being the clothing area (2,500 square meters for women's wear on the 2nd floor and 2,200 square meters for men's wear on the 3rd floor, attracting more foot traffic during weekday lunchtimes); and the 4th floor being the food and beverage area (total area 50 square meters). The building covers an area of ​​00 square meters and includes 10 restaurants of different types, with peak hours from 12:00-13:00 for lunch and 18:00-20:00 for dinner. The office area is concentrated on the 5th floor, housing 20 companies (mainly technology and design companies), with approximately 500 employees working daily, following a workday schedule of 9:00-18:00. The public areas extend throughout the entire building, including corridors on each floor (total length 1200m), elevator lobbies (12), and restrooms (24), making it a key area for pedestrian traffic. The energy equipment system is built around three core systems: air conditioning, lighting, and fresh air. The specific configuration is as follows: The air conditioning system consists of 20 ceiling-mounted air conditioning units, each with a cooling capacity of 50kW, supporting temperature adjustment from 16-30℃ (adjustment accuracy ±0.5℃) and switching between low / medium / high fan speeds. Four units are located in each of the commercial areas on floors 1-4 (corresponding to different zones on each floor), two units are located in the office area on floor 5 (covering the east and west sides of the office area), and two units are located in the public areas (responsible for corridors and elevator lobbies). The lighting system is configured according to the characteristics of each area. The commercial area uses embedded LED light strips (10W per strip, 1200 strips in total). The lighting system features 3000W warm white light with a color temperature of 3000K, supporting stepless brightness adjustment from 0-100%. Office areas use grille lights (300 lights total, 24W each, 4000K neutral light, supporting group switching and brightness adjustment). Public areas use downlights (500 lights total, 8W each, 3500K color temperature, supporting sensor-activated switching and basic brightness adjustment). The fresh air system is equipped with 15 fresh air units, providing independent air supply to each functional zone. Each unit supports 200-1000 m³ / h of fresh air volume adjustment (in 50 m³ / h increments) and has a built-in PM2.5 filter module to ensure fresh air quality. The entire system is connected to the complex's intelligent control system, supporting remote parameter adjustment and status monitoring, covering the energy supply and precise control needs of each functional zone.

[0131] To achieve precise control of energy equipment, a comprehensive data collection system was constructed, encompassing four categories of data: real-time crowd distribution, environmental parameters, historical energy consumption, and historical crowd flow characteristics. Each data collection scheme integrates hardware selection with software processing workflow design. For real-time crowd distribution data collection, a collaborative "hardware + algorithm" solution was adopted, deploying 120 Hikvision DS-2CD3T46WD-L high-definition cameras (2560×1440 resolution), arranged with differentiated density by area—one camera per 50㎡ in the 1st-floor atrium and one camera per 80㎡ in the clothing area. One monitoring unit is used for every 60 square meters in the dining area, one unit for every 100 square meters in the office area, and one unit for every 150 square meters in the public area. Using an AI people counting algorithm (optimized based on the YOLOv8 object detection framework, with added human trajectory tracking and deduplication function to avoid duplicate counting), the number of people in each monitoring unit (smallest unit is 50 square meters) is generated every 5 minutes. During the data processing stage, the 3σ principle is used to remove outliers (if the number of people in a single frame fluctuates by more than 30% compared to the average of the previous 3 frames, the average of the previous 3 frames is used to replace it), and a people matrix for each area is generated to ensure that the counting error is ≤5%. Environmental parameter data were collected using 300 Tuya Smart WS301 IoT sensors, evenly distributed in various areas at a density of 200㎡ / unit (must be installed at air conditioning return air vents and core areas of personnel activity). The monitored indicators included temperature (measurement range -10~60℃, accuracy ±0.5℃), humidity (0~100%RH, accuracy ±3%RH), light intensity (0~100000lux, accuracy ±5%), and PM2.5 (0~1000μg / m³, accuracy ±10μg / m³). Raw data was collected every 2 minutes, and after Z-score standardization (mean=0, standard deviation=1), the mean data every 5 minutes was summarized by area for easy use in subsequent model calls. Historical energy consumption data is obtained through the data interface between the Schneider A9MEM3150 smart meter (accuracy level 1, supporting real-time acquisition of active and reactive power) and the device controller. Real-time power consumption (kW) of each device is collected once every minute and stored in categories of "device type (air conditioning / lighting / fresh air) - area - time period". Historical data of nearly 12 months is retained (including seasonal fluctuation records, such as peak energy consumption of air conditioning in summer and increased lighting duration in winter). At the same time, operation and maintenance information such as device start and stop time and fault alarm are recorded simultaneously, providing complete data support for model analysis. Historical pedestrian flow data is extracted from the complex's property management system. The average number of people in each area is calculated per hour. The pedestrian flow characteristics are further categorized and analyzed for "weekdays / weekends / public holidays." For example, on weekdays, 10:00-12:00 and 18:00-20:00 are peak pedestrian flow times in the commercial area, while 9:00-12:00 and 14:00-18:00 are densely populated times in the office area. At the same time, the peak, trough and fluctuation coefficients of pedestrian flow in each area are calculated to form a structured historical pedestrian flow characteristic database.Data transmission and storage adopt a collaborative architecture of "edge computing + cloud storage": five Huawei Atlas500 edge servers (16 TOPS computing power, supporting multiple device access) are deployed at the edge to process camera traffic data and sensor environmental data in real time, keeping the processing latency within 100ms to avoid data transmission delays affecting control; the cloud uses the Alibaba Cloud IoT platform, configured with 10TB of storage space to store historical data, model parameters and control records, while supporting real-time data access and visualization (the property management backend can view real-time population, energy consumption curves and equipment status in each area through the web interface), and a data backup mechanism is set up (automatic backup every morning at midnight, retaining 30 days of historical backups) to ensure data security.

[0132] Based on the Python 3.9 programming language and the TensorFlow 2.10 deep learning framework, an LSTM (Long Short-Term Memory) time-series prediction model is constructed to predict the distribution of people in different regions during different time periods. The model input features take into account both real-time performance and historical correlation, specifically including two types of core features: real-time features cover the number of people in each region at the current moment (5-minute data obtained from the edge), the trend of the number of people in the previous hour (calculating the difference in the number of people every 15 minutes to form the slope feature), and real-time ambient temperature (corresponding to the sensitivity of the flow of people to the environment); historical features include the average number of people in the same region in the past 7 days (reflecting short-term patterns), the corresponding date type (weekday / weekend / public holiday, converted into numerical features using one-hot encoding), and the population fluctuation coefficient of the same period in history (such as the same month and solar term last year) (reflecting seasonal and periodic characteristics). The two types of features together constitute a 12-dimensional input vector. The model structure adopts a three-layer architecture of "input layer-hidden layer-output layer": The input layer receives a 12-dimensional feature vector, and maps the dimension to 24 dimensions through a fully connected layer to enhance the feature expression capability; The hidden layer sets up a 2-layer LSTM structure with 64 neurons in each layer, and the activation function is tanh. At the same time, a Dropout layer with a probability of 0.2 is introduced (randomly discarding some neurons) to avoid overfitting. The output of the first LSTM layer is normalized and then fed into the second LSTM layer to enhance the ability to capture temporal dependencies; The output layer maps the output of the hidden layer to a 6-dimensional vector through a fully connected layer, which corresponds to the prediction of the number of people in each area in the next 6 time periods (1 hour each) (e.g., predicting 280 people in the 1F atrium from 10:00-11:00, 350 people from 11:00-12:00, and 320 people from 12:00-13:00, etc.). The model training process uses the Adam optimizer (learning rate 0.001, weight decay 1e-5), and the loss function is mean squared error (MSE). The training dataset consists of 800,000 historical data points from January to April 2024, which are divided into a training set (560,000 data points), a validation set (80,000 data points), and a test set (160,000 data points) in a 7:1:2 ratio. The training epochs are set to 50, and an early stopping strategy is enabled (training stops when the validation set loss does not decrease for 5 consecutive epochs) to ensure a balance between convergence and generalization ability of the model.The test period was from May 1st to May 7th, 2024 (including 4 working days, 2 weekends, and 1 adjusted holiday). The model's prediction performance was excellent: In peak commercial areas (such as the 1st floor atrium), the prediction error was ≤8% within 1 hour (for example, if the actual number of people was 350, the predicted value was 322-378 people), the prediction error was ≤12% within 3 hours, and the prediction error was ≤15% within 6 hours; In off-peak commercial areas (such as the 2nd floor women's clothing area), due to smaller fluctuations in pedestrian flow, the error was ≤6% within 1 hour, ≤9% within 3 hours, and ≤11% within 6 hours; The office area had the strongest pedestrian flow pattern, with an error of ≤5% within 1 hour, ≤7% within 3 hours, and ≤9% within 6 hours. The model fully meets the accuracy requirements for time-based population prediction. The prediction results are pushed to the complex's intelligent control platform in real time, providing data support for subsequent energy demand analysis.

[0133] Based on historical data (including 12 months of energy consumption, population flow, and environmental data), a dynamic energy demand model integrating "multivariate linear regression + XGBoost (extreme gradient boosting tree)" is constructed to achieve accurate load prediction for different regions and different equipment. The model construction process consists of three steps: The first step is data preprocessing. Outliers (such as instantaneous high power consumption caused by air conditioner malfunctions) are removed from historical energy consumption data using box plots. Environmental parameters and population data are normalized. Key input features are selected through Pearson correlation analysis (redundant features with correlation < 0.3 are removed, and four core inputs are retained: population, ambient temperature, humidity, and light intensity). The second step is model training. First, a multiple linear regression model is used to fit the linear relationship between energy consumption and input features (such as air conditioner load = 0.12 × population + 0.08 × temperature - 0.03 × humidity + 5.2). Then, an XGBoost model is used to capture nonlinear correlations (such as the nonlinear feature that the air conditioner load increases faster after the population exceeds 5 people / 50㎡). The prediction results of the two models are fused with a weight of 6:4 to improve prediction accuracy. The third step is model validation. The model is validated on a test set (data from April 2024). The prediction error of air conditioner load is ≤ 8%, lighting load is ≤ 6%, and fresh air load is ≤ 7%, which meets the needs of practical applications. The model inputs are the number of people in the area (people / 50㎡) and environmental parameters (temperature, humidity, etc.), and the output is the real-time load prediction value (kW) of each device. Combined with the previously predicted "distribution data of people in the area at different times", a visualized heat map of people density is generated. Using the heat map generation function of Alibaba Cloud IoT platform, with 50㎡ as the smallest unit, the density of people is represented by color gradient (blue: ≤2 people / 50㎡, low density; yellow: 2-5 people / 50㎡, medium density; red: >5 people / 50㎡, high density), and the density change trend of each area in the next 1-6 hours is marked, which makes it easy for maintenance personnel to intuitively grasp the dynamics of people flow. Based on the load prediction results and density heat map, differentiated control strategies are formulated by area, time period, and device, which are executed by Siemens S7-1200 PLC controller (response delay ≤30s). The control command transmission process is as follows: the edge generates the control strategy according to the model output → transmits it to the PLC controller through the Modbus-TCP protocol → the PLC parses the command and sends the control signal to each device → the device executes the parameter adjustment and feeds back the status.Specific strategies are designed differently based on regional density types: For low-density areas (such as the 1F atrium before 9:00 AM on weekdays and the 5th-floor office area after 6:00 PM), air conditioning is set to 28℃ (summer) / 22℃ (winter) with low fan speed; lighting is limited to the main corridor light strips (30% brightness); and fresh air volume is 200-300 m³ / h (only emergency fresh air volume of 300 m³ / h is maintained after the office area is closed); for medium-density areas (such as the 2F women's clothing area from 2:00 PM to 4:00 PM on weekdays), air conditioning is set to 26℃ (summer) / 23℃ (winter). In winter, with medium wind speed, lighting is set to 80% (70% brightness) for light strips / grille lights, and fresh air volume is 500-600 m³ / h. In high-density areas (such as the 1F atrium from 15:00-18:00 on weekends and the 4F dining area from 12:00-13:00), air conditioning is set to 24℃ (summer) / 21℃ (winter), with high wind speed, lighting is 100% on (100% brightness), and fresh air volume is 800-1000 m³ / h (in the dining area, due to the large amount of oil fumes, the fresh air volume is set to 800 m³ / h and the exhaust frequency is increased). Seasonal and time-of-day adjustments are also considered, such as lowering the air conditioning temperature by 1℃ in high-density areas in summer and raising it by 1℃ in winter; at night (after 22:00), all areas are controlled according to the low-density area strategy, retaining only necessary lighting and fresh air.

[0134] After the equipment control is executed, a real-time feedback data collection mechanism is activated. Through the sensors and smart meters deployed above, two types of core data are collected synchronously at a frequency of once per minute: Environmental feedback data includes the actual temperature (°C), humidity (%RH), and PM2.5 concentration (μg / m³) of each area. At the same time, a personnel comfort questionnaire survey is conducted twice a day at 10:00 and 15:00 (targeting merchant employees and consumers, the questionnaire includes 5 dimensions: temperature comfort, humidity comfort, air quality, lighting comfort, and overall satisfaction, with a full score of 10 points), and the average comfort score is calculated; Equipment energy consumption data includes the real-time power consumption (kW) and cumulative energy consumption (kWh) of each air conditioner, lighting, and fresh air unit, as well as the equipment operating status (such as the number of times the air conditioner compressor starts and stops, and the number of times the lights are switched on and off), ensuring that the data covers the three-dimensional evaluation dimensions of "environment-energy consumption-experience" of the control effect. Based on the collected feedback data, a random forest algorithm was used to construct an energy efficiency optimization model. The model was trained using "control strategy parameters (such as air conditioning temperature, lighting brightness, and fresh air volume) + environmental feedback data" as input, with the goal of "optimal energy consumption-comfort balance" (defined as the optimal range where energy consumption is 10% lower than the regional average and comfort score is ≥8). The model outputs the optimal range of equipment operating parameters for each region. The training dataset consisted of 100,000 data points from one month after the adjustment. Five-fold cross-validation was used to evaluate the model performance. The key hyperparameters of the random forest (such as 100 decision trees, maximum depth of 10, and minimum number of sample splits of 5) were optimized using a grid search method. The final model achieved an accuracy of 89% and could accurately identify the energy efficiency optimization space in different regions (e.g., when the fresh air volume in the 4th-floor catering area was reduced from 800 m³ / h to 750 m³ / h during high-density periods, energy consumption decreased by 3% and comfort score did not decrease significantly). An energy efficiency assessment report is generated weekly, covering four core modules: overall energy consumption analysis (comparing energy consumption changes before and after adopting this method, such as a 18.5% decrease in overall energy consumption in the first week of May 2024 compared to the same period in April, with air conditioning energy consumption decreasing by 22%, lighting energy consumption by 30%, and fresh air energy consumption by 15%); environmental compliance statistics (92% compliance rate for temperature 22-28℃, 88% compliance rate for humidity 40%-60%, and 95% compliance rate for PM2.5≤50μg / m³); comfort assessment (average comfort score of 8.2 / 10, an increase of 0.7 points from before, and a 40% decrease in complaints); and area optimization suggestions (such as increasing the air conditioning fan speed from medium to high during peak hours in the 1F atrium, which would increase energy consumption by 2% but improve comfort by 0.5 points; and reducing the fresh air volume in the 4F dining area by 5%).Based on the analysis conclusions of the energy efficiency assessment report, the parameter weights of the dynamic energy demand model were iteratively updated. The update process followed four steps: "data analysis - weight adjustment - verification testing - formal application." The first step identified model deviations through the report (e.g., finding that the actual increase in the number of people had a greater impact on air conditioning load than the model predicted). The second step adjusted the weights of the corresponding features (e.g., increasing the weight of "number of people on air conditioning load" from 0.6 to 0.65, and decreasing the weight of "temperature on air conditioning load" from 0.3 to 0.25 to balance the impact of number of people and temperature). The third step tested the prediction accuracy of the adjusted model on the validation set (new data from the past week) to ensure reduced error. The fourth step formally deployed the adjusted model to the system, replacing the original model. Through continuous iteration, the model's prediction accuracy gradually improved: the prediction error for air conditioning load decreased from 10% to 7%, the prediction error for lighting load decreased from 8% to 5%, and the prediction error for fresh air load decreased from 9% to 6%. Meanwhile, a comprehensive model optimization is carried out once a month, and the model parameters are retrained to adapt to seasonal changes (such as increasing the weight of temperature features in summer and increasing the weight of humidity features in winter) to ensure that the model always fits the actual operating scenario, realizes the closed-loop management of "control-feedback-optimization", and continuously improves energy utilization efficiency and personnel comfort.

[0135] In summary, this invention constructs a multi-dimensional data collection system covering real-time population distribution, environmental parameters, historical energy consumption, and historical population flow characteristics. Combined with an efficient data processing and storage architecture, it provides precise data support for subsequent regulation. Relying on a time-series prediction model, it accurately predicts regional population changes, and uses a fusion-based dynamic energy demand model to predict equipment load. Furthermore, by combining population density heat maps, it formulates differentiated control strategies for different regions, time periods, and equipment, effectively improving energy utilization efficiency and reducing unnecessary energy consumption. After equipment regulation, real-time feedback data collection and energy efficiency optimization models balance energy consumption and personnel comfort, improving environmental quality, enhancing user experience, and reducing related complaints. The intelligent control system supports remote monitoring and parameter adjustment, and a data security mechanism ensures stable operation while reducing ineffective equipment start-ups and shutdowns and malfunctions, extending equipment lifespan, and reducing the labor costs of property energy management. Through regular energy efficiency assessments and iterative updates of model parameters, a closed-loop management system of "regulation-feedback-optimization" is formed, ensuring that the model adapts to seasonal changes and actual operating scenarios, continuously improving energy utilization efficiency and personnel comfort, and achieving long-term optimization and sound operation of the complex's energy management.

[0136] Next, referring to the accompanying drawings, a system for controlling energy equipment based on the regional population distribution is described according to an embodiment of this application.

[0137] Figure 5 This is a schematic diagram of the structure of an energy equipment control system based on the regional population distribution according to an embodiment of this application.

[0138] like Figure 5 As shown, the energy equipment control system 10 based on regional population distribution includes: an acquisition module 100, a prediction module 200, a generation module 300, and an update module 400.

[0139] The module 100 is used to acquire real-time population distribution data, environmental parameter data, energy equipment operating status data, historical energy consumption data, and historical population flow characteristic data. The prediction module 200 is used to predict population distribution data in different time periods based on historical population flow characteristic data and real-time population distribution data. The generation module 300 is used to construct a dynamic energy demand model based on historical energy consumption data, historical population flow characteristic data, real-time population distribution data, and environmental parameter data. It generates a population density heat map by combining population distribution data in different time periods and predicts changes in regional energy equipment load in real time. Based on load changes and population density heat map, it generates differentiated control strategies and dynamically adjusts the operating parameters of energy equipment according to the differentiated control strategies. The update module 400 is used to dynamically regulate energy equipment according to the differentiated control strategies, synchronously collect regional environmental feedback data and equipment energy consumption data after regulation, train an energy efficiency optimization model based on environmental feedback data and energy consumption data, generate an energy efficiency assessment report, and update the parameter weights of the dynamic energy demand model based on the energy efficiency assessment report.

[0140] It should be noted that the foregoing explanation of an embodiment of a method for controlling energy equipment based on regional population distribution also applies to a system for controlling energy equipment based on regional population distribution in this embodiment, and will not be repeated here.

[0141] The energy equipment control system based on regional population distribution proposed in this application integrates multi-source data such as real-time population, environmental parameters, historical energy consumption, and population flow characteristics. It accurately predicts the population in different time periods using historical and real-time data, and generates population density heat maps and load forecasts using a dynamic energy demand model. This differentiated control strategy addresses the problems of delayed response and large parameter matching deviations in traditional timed and fixed-threshold adjustments to tidal population flows, breaking down the disconnect between dynamic population information and energy consumption data. Simultaneously, it continuously updates the parameter weights of the dynamic energy demand model through post-adjustment environmental feedback and energy consumption data training. This significantly improves energy utilization efficiency while ensuring regional environmental comfort, effectively filling the gap in deep linkage between population perception and energy control in existing technologies, and minimizing ineffective energy consumption while ensuring regional environmental comfort. Therefore, it solves the problems of high response delay and limited parameter matching accuracy in existing technologies.

[0142] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0143] The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0144] When the processor 602 executes the program, it implements the energy equipment control method based on the regional population distribution provided in the above embodiments.

[0145] Furthermore, electronic devices also include:

[0146] Communication interface 603 is used for communication between memory 601 and processor 602.

[0147] The memory 601 is used to store computer programs that can run on the processor 602.

[0148] The memory 601 may include high-speed RAM (Random Access Memory) and may also include non-volatile memory, such as at least one disk storage.

[0149] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0150] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0151] The processor 602 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0152] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for controlling energy equipment based on the distribution of population in a given area.

[0153] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0154] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0155] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0156] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0157] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0158] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for controlling energy equipment based on regional population distribution, characterized in that, include: Acquire real-time population distribution data, environmental parameter data, historical energy consumption data, and historical population flow characteristic data; Based on the historical pedestrian flow characteristic data and the real-time population distribution data, predict the population distribution data in different time periods and regions; Based on the historical energy consumption data, historical pedestrian flow characteristic data, real-time population distribution data, and environmental parameter data, a dynamic energy demand model is constructed. Combined with the population distribution data for different time periods, a population density heatmap is generated. This involves constructing a regional spatial grid division model. Based on this model, and combining the peak time distribution and average dwell time from the historical pedestrian flow characteristic data with the instantaneous population in each region from the real-time population distribution data, a kernel density estimation algorithm is used to calculate the population density value for each grid unit. The historical data weights are dynamically adjusted with a time decay coefficient. The population density values ​​are mapped to a preset color gradient range and overlaid onto a regional two-dimensional planar map to generate a population density heatmap that labels the real-time population range, density level, and update timestamp for each grid unit. Real-time prediction of regional energy equipment load changes is also performed. Based on the load changes and the population density heatmap, a differentiated control strategy is generated, and the operating parameters of the energy equipment are dynamically adjusted according to the differentiated control strategy. According to the differentiated control strategy, energy equipment is dynamically regulated, and regional environmental feedback data and equipment energy consumption data are collected synchronously after regulation. Based on the environmental feedback data and the energy consumption data, an energy efficiency optimization model is trained, an energy efficiency assessment report is generated, and the parameter weights of the dynamic energy demand model are updated according to the energy efficiency assessment report.

2. The method for controlling energy equipment based on regional population distribution according to claim 1, characterized in that, Based on the historical energy consumption data, the historical pedestrian flow characteristic data, the real-time population distribution data, and the environmental parameter data, a dynamic energy demand model is constructed, including: Construct multivariate regression models and long short-term memory network models; Based on historical energy consumption data, historical population flow characteristics data, real-time population distribution data, and environmental parameter data, the multivariate regression model is used to fit the multidimensional characteristics and output a preliminary energy demand forecast. By capturing the temporal dependence characteristics of energy demand through the Long Short-Term Memory network model and integrating the preliminary energy demand forecasts, a dynamic energy demand forecasting model is established.

3. The method for controlling energy equipment based on regional population distribution according to claim 2, characterized in that, The dynamic energy demand model formula is as follows: ； in, ; Real-time population density; Ambient temperature; For ambient humidity; This is a historical energy consumption benchmark value; These are the weighting coefficients.

4. The method for controlling energy equipment based on regional population distribution according to claim 1, characterized in that, Based on the historical pedestrian flow characteristic data and the real-time population distribution data, predict the population distribution data for different time periods and regions, including: Construct a pedestrian flow prediction model; Based on the aforementioned pedestrian flow prediction model, combining the temporal periodicity of historical pedestrian flow characteristic data with the spatial distribution characteristics of real-time pedestrian flow distribution data, the model extracts time-dependent features through a gated temporal convolutional network and captures spatial correlations between regions through a graph attention network, thereby predicting the pedestrian flow distribution data and confidence intervals for each region at different time periods.

5. The method for controlling energy equipment based on regional population distribution according to claim 1, characterized in that, Based on the environmental feedback data and the energy consumption data, an energy efficiency optimization model is trained, and an energy efficiency assessment report is generated, including: Construct an energy efficiency optimization model; Based on regional environmental feedback data and equipment energy consumption data, the energy efficiency optimization model is trained through a multi-agent collaborative optimization algorithm, and energy efficiency evaluation indicators are output. An energy efficiency assessment report is generated based on the energy efficiency assessment indicators. The energy efficiency assessment report includes energy consumption per unit area, energy efficiency deviation value, and optimization suggestions.

6. The method for controlling energy equipment based on regional population distribution according to claim 1, characterized in that, Real-time prediction of regional energy equipment load changes, including: Build a load forecasting model; Based on the load forecasting model, the sliding window method is used to extract historical load fluctuation characteristics, and the gradient boosting tree algorithm is used to predict the load change trend. When the load change trend exceeds the preset threshold of the equipment's rated load, the load buffering mechanism is triggered, generating a pre-adjustment command.

7. A system for controlling energy equipment based on regional population distribution, characterized in that, include: The acquisition module is used to acquire real-time population distribution data, environmental parameter data, energy equipment operating status data, historical energy consumption data, and historical population flow characteristic data. The prediction module is used to predict the population distribution data in different time periods based on the historical population flow characteristic data and the real-time population distribution data. The generation module is used to construct a dynamic energy demand model based on the historical energy consumption data, the historical pedestrian flow characteristic data, the real-time population distribution data, and the environmental parameter data. It combines the population distribution data of different time periods to generate a population density heat map and predict the load changes of regional energy equipment in real time. Based on the load changes and the population density heat map, it generates a differentiated control strategy and dynamically adjusts the operating parameters of energy equipment according to the differentiated control strategy. The update module is used to dynamically adjust energy equipment according to the differentiated control strategy, synchronously collect regional environmental feedback data and equipment energy consumption data after adjustment, train an energy efficiency optimization model based on the environmental feedback data and the energy consumption data, generate an energy efficiency assessment report, and update the parameter weights of the dynamic energy demand model according to the energy efficiency assessment report.

Images