HVAC intelligent regulation and control method and system based on personnel dynamic prediction
By deploying monitoring nodes in the HVAC system and combining LSTM and linear regression models for multi-dimensional prediction, the operation of HVAC equipment can be dynamically adjusted, solving the problems of energy waste and insufficient comfort, and achieving efficient energy saving and dynamic adjustment of comfort.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 青岛和泰精密工业有限公司
- Filing Date
- 2026-03-25
- Publication Date
- 2026-05-12
AI Technical Summary
Existing HVAC systems lack dynamic optimization mechanisms, resulting in energy waste and insufficient comfort. They cannot quickly respond to changes in the number of people indoors, and traditional forecasting methods have large errors and lack the ability to predict future demand.
By deploying multiple monitoring nodes in the building area to collect data in real time, and combining LSTM and linear regression models to predict the number of people in the short and historical dimensions, the system dynamically allocates weights and generates control commands using PID control algorithms to optimize the operation of HVAC equipment.
It reduces prediction errors, improves system energy efficiency, ensures that regional comfort can be adjusted as needed, and reduces hysteresis caused by thermal inertia.
Smart Images
Figure CN122015239A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent building technology, and in particular to an intelligent HVAC control method and system based on dynamic prediction of personnel. Background Technology
[0002] Numerous studies have shown that most HVAC (Heating, Ventilation, and Air Conditioning) systems rely heavily on single temperature and humidity sensors or simple feedback control, failing to consider the number of people in each area. This leads to overheating or overcooling. For example, in shopping malls, the distribution of people varies across different areas at different times, but current technologies often implement uniform control, resulting in some areas being overheated or undercooled. Similarly, in office areas where no one is working during lunchtime, the system continues to operate in its usual mode. Research indicates that this rigid operating mode leads to 20%-40% energy waste. When the number of people indoors changes suddenly, HVAC equipment cannot react quickly due to thermodynamic inertia. For instance, if a meeting suddenly starts, the equipment needs to run for a period of time to adjust the temperature to a comfortable range, leading to decreased comfort. Furthermore, traditional systems rely on current data for reactive control, lacking anticipation of future needs, resulting in delayed response and energy waste. For example, in the last few minutes before a meeting ends, the equipment may still be running at relatively high power, unable to predictably reduce power consumption.
[0003] Existing control methods rely on fixed, manually set parameters, lacking dynamic optimization mechanisms and making it difficult to adapt to long-term environmental changes. For example, shopping mall traffic patterns differ between holidays and weekdays. The rigidity of HVAC control methods leads to energy waste due to over-adjustment or insufficient comfort due to under-adjustment. Furthermore, traditional crowd forecasting methods often use single forecasting models, resulting in significant discrepancies between predicted and actual values. Summary of the Invention
[0004] This application provides an intelligent HVAC control method and system based on dynamic personnel prediction, which solves the technical problem that existing HVAC equipment control methods are prone to energy waste.
[0005] On the one hand, this application provides an intelligent HVAC control method based on dynamic personnel prediction, the method comprising the following steps: Step S1: Deploy multiple monitoring nodes in the building control area through the sensing layer to collect real-time data on the number of people and environmental parameters in the area, and transmit the data to the integrated control module in a standardized format; Step S2: Based on the data collected by the perception layer, the number of people is predicted in the short-term and historical dimensions through the prediction layer. In the short-term dimension, the LSTM model is used to predict the number of people in the future second preset time period based on the data of the first preset time period in the past. In the historical dimension, the number of people in the same period is predicted through time clustering and linear regression models. The weights are dynamically allocated according to the prediction error, and the weighted and fused comprehensive predicted number of people is output. Step S3: The decision-making level calculates the target temperature setpoint based on the real-time number of people and the comprehensive predicted number of people, combined with the thermal comfort temperature and the energy-saving temperature, and generates regional HVAC equipment control instructions based on the PID control algorithm; Step S4: The execution layer adjusts the operating parameters of the HVAC equipment in the corresponding area according to the HVAC equipment control instructions of the area, and simultaneously calculates the total load of the building air conditioning to adjust the power of the air conditioning unit; Step S5: Optimize control parameters by analyzing historical operating data through the evaluation layer, detect abnormal data and generate fault diagnosis reports to achieve iterative optimization.
[0006] In one implementation of this application, in step S1, each node includes a visual sensor, an infrared sensor, a CO2 sensor, and a temperature and humidity sensor; the deployment of monitoring nodes specifically includes: uniformly distributing them on the ceiling of the control area in units of a preset area to ensure full coverage; hierarchically spatially encoding the building area, and synchronizing the clocks of each sensor through the NTP protocol, and setting timestamp errors.
[0007] In one implementation of this application, step S2, the short-term dimension prediction process, specifically includes: performing linear interpolation of missing values and outlier removal on sensor data; standardizing the number of people using the Z-score normalization method, inputting it into an LSTM model for training, the model structure including an input layer, a 32-neuron LSTM layer, and a fully connected output layer; training the model with the goal of minimizing MAE, and outputting the predicted number of people in the second future time period after inverse normalization.
[0008] In one implementation of this application, the historical dimension prediction process in step S2 specifically includes: clustering timestamps into Monday to Sunday and holidays, and filtering special weather periods by associating them with historical meteorological data; encoding historical data of the same category by minute, inputting them into a linear regression model, and outputting the predicted number of people in the second future time period.
[0009] In one implementation of this application, the weight allocation formula in step S2 is:
[0010]
[0011] Where a represents short-term prediction error and b represents historical prediction error. To find the minimum value, avoid having a denominator of 0.
[0012] In one implementation of this application, the formula for calculating the target temperature setpoint in step S3 is:
[0013] in, The target temperature set for the region based on energy conservation and comfort objectives; The standard operating temperature for the functional area corresponds to the first-level thermal comfort rating range. This is the energy-saving temperature, which is the lower limit of the temperature range for Level 2 thermal comfort. The number of people is a combination of the real-time and predicted number of people; k is the proportion of the trigger zone to adjust to a comfortable temperature; Design capacity for the area or the number of regular users.
[0014] In one implementation of this application, the anomaly detection process in step S5 specifically includes: setting threshold ranges for personnel quantity, environmental parameters, and equipment operating parameters; analyzing abnormal data characteristics through association rule algorithms to locate the root cause of sensor failure or network interruption, and automatically generating a diagnostic report.
[0015] In one implementation of this application, the method further includes: automatically switching back to intelligent control mode after the area enters manual control mode and continues for a third preset time period; if the area remains unattended for an extended period of time, an energy-saving temperature setting is triggered.
[0016] This application also provides an HVAC intelligent control system based on dynamic personnel prediction. The system includes: a sensing module, consisting of multiple monitoring nodes, each node integrating a visual sensor, an infrared sensor, a CO2 sensor, and a temperature and humidity sensor, for real-time collection of personnel number and environmental parameter data; a prediction module, deployed on edge computing nodes, including short-term prediction units and historical prediction units, which output a weighted fusion of the predicted number of personnel through LSTM and linear regression models, respectively; a decision-making module, which calculates the target temperature setpoint based on real-time data and prediction results, and generates control commands through a PID algorithm; an execution module, which adjusts the operating parameters of the regional HVAC equipment according to the commands and dynamically adjusts the power of the main air conditioning unit; and an evaluation module, which uses machine learning to optimize control parameters, detects abnormal data, and generates fault reports.
[0017] In one implementation of this application, the data transmission rules of the sensing module are as follows: real-time uploading is triggered when the number of people changes, and temperature and humidity data are uploaded at a fixed frequency; the transmission format is: spatial code-timestamp-number of people-temperature-humidity, and the data is compressed and transmitted after noise reduction by Kalman filtering.
[0018] The HVAC intelligent control method and system based on dynamic personnel prediction provided in this application have the following beneficial effects: (1) By combining short-term and historical predictions, the future number of people can be predicted in multiple dimensions. At the same time, the weight of the dimensions is changed by the error with the actual number of people, the predicted value of the dimension with higher error is reduced, and the predicted number of people is corrected. This solves the problem of excessive error caused by the single prediction model of the original prediction method.
[0019] (2) Based on real-time personnel numbers and environmental parameters, combined with future demand forecasts, adjust the air conditioning operation mode in advance to avoid excessive cooling or heating and improve the system's energy efficiency.
[0020] (3) A weighted comprehensive control strategy is adopted to dynamically adjust the temperature setting value according to the number of people, ensuring that comfort is prioritized in high-density areas and energy conservation is emphasized in low-density areas, thus achieving on-demand adjustment.
[0021] (4) By predicting future population, the temperature of the region can be adjusted in advance to offset the lag caused by real-time control of thermal inertia. Attached Figure Description
[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of an HVAC intelligent control method based on personnel dynamic prediction provided in this application embodiment; Figure 2 The overall logic flowchart provided for the embodiments of this application; Figure 3 This is a diagram illustrating the composition of an HVAC intelligent control system based on dynamic personnel prediction, provided as an embodiment of this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] This application provides an HVAC intelligent control method and system based on dynamic personnel prediction. The technical solution proposed in this application will be described in detail below with reference to the accompanying drawings.
[0025] Figure 1 A flowchart illustrating an HVAC intelligent control method based on dynamic personnel prediction, provided as an embodiment of this application. Figure 1 As shown, the method mainly includes the following steps: Step S1: Deploy multiple monitoring nodes in the building control area through the sensing layer to collect real-time data on the number of people and environmental parameters in the area, and transmit the data to the integrated control module in a standardized format.
[0026] In this embodiment, the sensor deployment and installation are configured to be based on a preset area according to the building structure, specifically 1m². 2 or 4m 2 (The specific value of the preset area is not limited in this manual, and other values can also be used.) Each node is evenly distributed in the ceiling of the control area (the control area is the control range of the controllable unit HVAC equipment) to ensure that the node fully covers the area to be controlled. Each node is pre-set with a dynamic monitoring system, which includes a visual sensor (to achieve anonymized people counting), an infrared sensor (to assist in calibrating the number of people), a CO2 sensor, and a temperature and humidity sensor to ensure comprehensive perception of the number of people and environmental parameters in the area, and to perform time alignment of each sensor (based on the NTP protocol or manually synchronized to unify the clocks of all sensors to ensure that the data timestamp error is ≤1s). Furthermore, a global monitoring network is established, specifically by hierarchically coding building areas in a format of building-floor-monitoring module, such as B1-F2-Z3. This coding rule cleverly implies spatial proximity, with areas like Z3 and Z4 representing adjacent regions. This coding method facilitates precise data location during system operation and also enables collaborative control between different areas.
[0027] Furthermore, the data processing and transmission process is as follows: the monitoring data is standardized and noise-reduced (e.g., Kalman filtering), and compressed and transmitted to the integrated control system according to different regions and different sensor data through different transmission rules. The transmission rules are as follows: (1) Number of people: changes in the number of people in the area trigger real-time upload. (2) Temperature and humidity: data is uploaded in real time. (3) Transmission format: {spatial code-time code-number of people-temperature-humidity}.
[0028] Step S2: Based on the data collected by the perception layer, the number of people is predicted in the short-term and historical dimensions through the prediction layer. In the short-term dimension, the LSTM model is used to predict the number of people in the future second preset time period based on the data of the first preset time period in the past. In the historical dimension, the number of people in the same period is predicted through time clustering and linear regression models. The weights are dynamically allocated according to the prediction error, and the weighted and fused comprehensive predicted number of people is output.
[0029] In this embodiment of the application, in order to reduce costs or reduce network latency caused by large amounts of data transmission, the prediction layer can be set up on the edge computing node, and the population of each region can be predicted independently.
[0030] Specifically, the data preprocessing for short-term dimension prediction is as follows: (1) Extract timestamps and personnel quantity data from the data collected by sensors. For missing data, if a minute's data is missing, linear interpolation between the two points before and after is used to fill it. For outliers, box plots (IQR method) are used to identify and remove them, and then interpolation is used to complete the data. (2) Convert timestamps into continuous numerical features, specifically, they are encoded into minutes from 0 to 59 in chronological order. These values serve as time dimension identifiers for the model input, providing an intuitive time reference for the model when analyzing the pattern of personnel quantity changes over time. (3) In the data processing stage, the Z-score normalization method is used to standardize the personnel quantity data. Through this method, the personnel quantity data is converted into a standard normal distribution with a mean of 0 and a standard deviation of 1. The purpose of this processing is to eliminate the difference in scale between different data, improve the stability of the model training process, and avoid poor model training results due to data scale issues.
[0031] Furthermore, in the process of building and training the model, the first step is to prepare the data by selecting historical data from the past three months and generating training samples in a sliding time window manner. Each sample is input by the sequence of the number of people in the past 60 minutes (the specific value of the first preset time period is not specified in this manual, but can also be 30 minutes or 45 minutes), and output by the actual number of people in the next 10 minutes (the specific value of the second preset time period is not specified in this manual, but can also be 15 minutes or 5 minutes).
[0032] Furthermore, this embodiment employs a Long Short-Term Memory (LSTM) network model structure. The input layer receives the population count sequence data from the past 60 minutes. The LSTM layer consists of one LSTM unit with 32 neurons to fully capture long-term dependencies in the time series. The output layer is a fully connected layer that processes the data output from the LSTM layer to output a predicted population count for the next 10 minutes.
[0033] Furthermore, during model training, the optimization objective is to minimize the mean absolute error (MAE) between the predicted and actual values. Training cycles and batches are set to 100 epochs, with each batch containing 32 samples. Validation set configuration: 10% of the data is reserved as a validation set to monitor for overfitting during model training.
[0034] In this embodiment of the application, the values for the number of layers and the number of neurons in the LSTM neural network are provided only as a reference. No specific limit is made on the specific number of layers and the number of neurons, and other values may also be used.
[0035] Furthermore, during real-time prediction, (1) Input data: The sequence of the number of people over the past 60 minutes is obtained from the perception layer in real time, with the data format being N_{t-59}, N_{t-58}, …, N_t. (2) Prediction output: The obtained sequence of the number of people is input into the trained model, and the model outputs the predicted number of people for the next 10 minutes, with the format being N_{t+1}, N_{t+2}, …, N_{t+10}. (3) Denormalization: Since the data input to the model has undergone Z-score normalization, the predicted value needs to be converted from the Z-score range back to the predicted number of people. The short-term prediction result is output, that is, the set of predicted number of people in the short-term dimension for the next 10 minutes. Its elements include N1, N2, ..., N 10 .
[0036] In this application embodiment, the prediction of the historical dimension specifically includes: (1) Time clustering: The timestamps are clustered into the same group of datasets. The datasets are divided into 8 categories, from Monday to Sunday and holidays, and coded as 1-8. Priority is given to whether it is a holiday. For example, if the Dragon Boat Festival is on a Monday, then the timestamp is in the holiday group.
[0037] (2) Data selection: For each time cluster, select time data from each sensor over the past year. (In the early stages of system operation, if there is insufficient data in the time cluster set, data from the past 6 months can be selected. If there are less than 6 months, the time cluster set will not be used for historical dimension prediction. For example, if the system only runs for 4 weeks, or runs for 6 weeks but the data in the holiday cluster set is less than 20 times, the data will only be collected and will not be used for historical reference.)
[0038] (3) Data Filtering: Historical meteorological information is obtained using meteorological data APIs. Based on this information, data from periods with special weather (such as rain or haze) are filtered out. The purpose is to ensure that the data referenced when making historical-dimensional forecasts comes from typical scenarios, avoiding interference from abnormal factors such as special weather and thus improving the accuracy of the forecasts.
[0039] (4) Data compression and import: Divide the historical number of people into nodes according to every minute, and code them as 1-840 (assuming the system operates for 14 hours a day). Calculate and generate the number of people in each time period and align it with the same code in the same group.
[0040] (5) Personnel prediction output: Take the same coded data and embed it into the linear regression model (e.g., 8-15, that is, the 15th minute of the system operation during the holiday), and output the predicted number of people in the next 10 minutes based on the historical dimension. Its elements include N1, N2, ..., N 10 .
[0041] It should be noted that the historical dimension database needs to be updated periodically, and only data from the past year is retained.
[0042] Furthermore, the predicted number of people is output based on weights. First, the predicted number of people for the currently concluded prediction period is obtained from the short-term and historical prediction models. The actual number of people within the currently concluded prediction period is obtained from the perception layer. The prediction errors of the two prediction models within the currently concluded prediction period are calculated separately, including short-term prediction error a: the prediction error between the predicted value of the past time window and the current actual value. Historical prediction error b: the prediction error between the predicted value of the same period in the historical dimension and the current actual value.
[0043] The weighting formula is as follows:
[0044]
[0045] in, Minimum value (e.g.) ), to avoid the denominator being 0.
[0046] It should be noted that: In the first prediction period after the system starts, there is no historical prediction error data, and the initial weights can be set to 50% each. Starting from the second prediction period, the above dynamic weight allocation method will be officially used to update the weights based on the prediction error of the previous prediction period. The smaller the error, the higher the weight.
[0047] The weighted fusion formula is as follows:
[0048] Then output the predicted number of people per minute for the next 10 minutes, i.e., the set. Its elements include N1, N2, ..., N 10 ;where N i For the predicted number of people in the i-th minute (i=1, 2, ..., 10), subsequent calculations are performed according to the minute sequence number, and the predicted number of people in different dimensions in the same minute are matched and calculated. The weights are dynamically updated according to the prediction error every minute.
[0049] Step S3: The decision-making level calculates the target temperature setpoint based on the real-time number of people and the comprehensive predicted number of people, combined with the thermal comfort temperature and the energy-saving temperature, and generates regional HVAC equipment control instructions based on the PID control algorithm.
[0050] In this embodiment, the core formula for setting the temperature regulation of the decision-making layer is as follows:
[0051] The target temperature set for the region based on energy conservation and comfort objectives; The standard operating temperature for the functional area can be set to 25℃ in summer and 23℃ in winter, corresponding to the first-level thermal comfort level range of GB / T 50785-2012 "Evaluation Standard for Indoor Thermal and Humidity Environment of Civil Buildings"; For energy saving, the temperature can be set to 28℃ in summer and 18℃ in winter, which is the lower limit of the temperature range of the above-mentioned standard level 2 thermal comfort level. The number of people is a combination of the real-time and predicted number of people; k is the proportion of the trigger zone to adjust to a comfortable temperature; This is for the area design or the number of regular users. It should be noted that this applies to cooling operation. The specific range of values for this should be within [the specified range]. and The specific rules are as follows: when ≥ The system executes directly. Prioritize ensuring the thermal comfort of indoor occupants; when 0 < ≤ At that time, the set temperature is calculated according to the above formula to achieve a dynamic balance between comfort and thermal equilibrium; when When ≤0, the system executes directly. To minimize ineffective energy consumption.
[0052] Example: Let N be the number of times the classroom area is used regularly. 基础 =40 people, k=15%, the number of people (k·N) whose temperature is adjusted to a comfortable level in the trigger zone. 基础 =6 people, then when N 综合 When there are ≥6 people, T 设定= T 舒适 ; 0 < ≤ When the area is unoccupied, calculate the set temperature according to the above formula; when the area is unoccupied, execute... .
[0053]
[0054] in, The weight assigned to the current number of people and the temperature (can be set to 0.6). For the current number of people, The weighting of temperature for the predicted number of people in the i-th minute; N i Let N be the number of people in the i-th minute. 基础The number of people in the area is designed for or used regularly (e.g., the number of seats in a meeting room), and m is the total number of minutes in the predicted time window (e.g., 10).
[0055]
[0056] in, The weighting is exponentially decaying, meaning the weight of the predicted number decreases rapidly in the early stages of the time window and then levels off, emphasizing the greater weight of near-term predictions. This is the exponential decay coefficient, which represents the rate at which the predicted weight decays over time. It can be set based on thermodynamic inertia, initially set to 3. If the thermal inertia is large, The dosage may be increased as appropriate.
[0057] also, This is the normalization coefficient, which forces the total weight of the current number of people and the predicted number of people to be 1.
[0058] In this embodiment, a comfort compensation mechanism is also set up. A manual mode can be configured, allowing personnel to manually adjust the temperature. After the area enters unmanned or manual mode for two hours (the specific duration is not limited in this specification), it re-enters intelligent control mode to compensate for the comfort issue of a very small number of people being at the edge of comfort temperature for an extended period. The area control command is based on the traditional PID control method, issuing control commands to the area HVAC equipment to bring the area temperature to T. 设定 At the same time, it ensures appropriate humidity. The total power of the air conditioning unit is calculated by summing the loads of each area to determine the total load of the area to be controlled, and then issuing control commands to the air conditioning unit according to the total load.
[0059] Step S4: The execution layer adjusts the operating parameters of the HVAC equipment in the corresponding area according to the HVAC equipment control instructions of the area, and simultaneously calculates the total load of the building air conditioning to adjust the power of the air conditioning unit.
[0060] In this embodiment of the application, the HVAC equipment (fan frequency, valve opening, and main air conditioning unit power) corresponding to the region is adjusted according to the decision command.
[0061] Step S5: Optimize control parameters by analyzing historical operating data through the evaluation layer, detect abnormal data and generate fault diagnosis reports to achieve iterative optimization.
[0062] In this embodiment, the evaluation layer utilizes existing AI technology to set normal range thresholds for data such as personnel numbers, environmental parameters, fan frequency, and valve opening degree. Data at each level is detected, and data exceeding the threshold is marked as abnormal. When evaluating system operation and troubleshooting, an association rule algorithm is used to conduct in-depth analysis of the abnormal data characteristics. This algorithm can quickly and accurately locate the source of the fault, such as determining whether it is a sensor malfunction or a network interruption. Furthermore, the system automatically generates detailed diagnostic reports, providing strong support for subsequent maintenance and system optimization.
[0063] Based on historical system operating data, evaluation indicators such as temperature control accuracy and energy consumption are calculated. Machine learning algorithms are used to analyze their relationship with control parameters, thereby optimizing decision-making level control commands, such as adjusting valve opening control parameters. The personnel number prediction model at the forecasting level is evaluated, and the processing methods for historical forecast data in various dimensions are optimized to improve prediction accuracy.
[0064] In this embodiment of the application, the overall control flowchart is as follows: Figure 2 As shown.
[0065] The above describes an HVAC intelligent control method based on personnel dynamic prediction, as provided in this application. Based on the same inventive concept, this application also provides an HVAC intelligent control system based on personnel dynamic prediction. Figure 3 A diagram illustrating the composition of an HVAC intelligent control system based on personnel dynamic prediction, as provided in this application embodiment, is shown below. Figure 3 As shown, the system mainly includes: a perception module 301, consisting of multiple monitoring nodes, each node integrating a visual sensor, an infrared sensor, a CO2 sensor, and a temperature and humidity sensor, used to collect real-time data on the number of people and environmental parameters; a prediction module 302, deployed on edge computing nodes, including a short-term prediction unit and a historical prediction unit, which respectively output a weighted fusion of the predicted number of people through an LSTM model and a linear regression model; a decision module 303, which calculates the target temperature setpoint based on real-time data and prediction results, and generates control instructions through a PID algorithm; an execution module 304, which adjusts the operating parameters of the regional HVAC equipment according to the instructions, and dynamically adjusts the power of the main air conditioning unit; and an evaluation module 305, which uses machine learning to optimize control parameters, detects abnormal data, and generates fault reports.
[0066] Furthermore, the data transmission rules of the sensing module are as follows: real-time upload is triggered when the number of people changes, and temperature and humidity data are uploaded at a fixed frequency; the transmission format is: spatial code-timestamp-number of people-temperature-humidity, and the data is compressed and transmitted after noise reduction by Kalman filtering.
[0067] This application provides an intelligent HVAC control method and system based on dynamic personnel prediction. By combining short-term and historical predictions, it forecasts future personnel numbers from multiple dimensions. Simultaneously, it adjusts the weights of dimensions based on the error compared to the actual number of people, reducing the predicted values of dimensions with higher errors and correcting the predicted number of people. This solves the problem of excessive errors caused by the single prediction model in traditional prediction methods. Based on real-time personnel numbers and environmental parameters, combined with future demand predictions, it adjusts the air conditioning operation mode in advance to avoid excessive cooling or heating, improving system energy efficiency. A weighted comprehensive control strategy is adopted to dynamically adjust the temperature setpoint according to the number of people, ensuring that comfort is prioritized in high-density areas while energy conservation is emphasized in low-density areas, achieving on-demand adjustment. By predicting future personnel numbers, it adjusts the temperature of the area in advance, offsetting the lag caused by real-time control due to thermal inertia.
[0068] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0069] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0070] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for intelligent HVAC control based on dynamic personnel prediction, characterized in that, The method includes the following steps: Step S1: Deploy multiple monitoring nodes in the building control area through the sensing layer to collect real-time data on the number of people and environmental parameters in the area, and transmit the data to the integrated control module in a standardized format; Step S2: Based on the data collected by the perception layer, the number of people is predicted in the short-term and historical dimensions through the prediction layer. In the short-term dimension, the LSTM model is used to predict the number of people in the future second preset time period based on the data of the first preset time period in the past. In the historical dimension, the number of people in the same period is predicted through time clustering and linear regression models. The weights are dynamically allocated according to the prediction error, and the weighted and fused comprehensive predicted number of people is output. Step S3: The decision-making level calculates the target temperature setpoint based on the real-time number of people and the comprehensive predicted number of people, combined with the thermal comfort temperature and the energy-saving temperature, and generates regional HVAC equipment control instructions based on the PID control algorithm; Step S4: The execution layer adjusts the operating parameters of the HVAC equipment in the corresponding area according to the HVAC equipment control instructions of the area, and simultaneously calculates the total load of the building air conditioning to adjust the power of the air conditioning unit; Step S5: Optimize control parameters by analyzing historical operating data through the evaluation layer, detect abnormal data and generate fault diagnosis reports to achieve iterative optimization.
2. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S1, each node includes a visual sensor, an infrared sensor, a CO2 sensor, and a temperature and humidity sensor; the deployment of monitoring nodes specifically includes: uniformly distributing them on the ceiling of the control area in units of a preset area to ensure full coverage; hierarchically spatially encoding the building area and synchronizing the clocks of each sensor through the NTP protocol, and setting timestamp errors.
3. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S2, the short-term dimensional prediction process specifically includes: Perform linear interpolation of missing values and outlier removal on sensor data; The number of people was standardized by Z-score normalization and input into the LSTM model for training. The model structure includes an input layer, a 32-neuron LSTM layer and a fully connected output layer. The model is trained with the goal of minimizing MAE, and the output is the predicted number of people in the second future time period after inverse normalization.
4. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S2, the historical dimension prediction process specifically includes: The timestamps are clustered into Monday through Sunday and public holidays, and historical meteorological data is linked to filter out special weather periods; Encode historical data of the same category by minute, input it into a linear regression model, and output the predicted number of people in the second future time period.
5. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S2, the weight allocation formula is as follows: Where a represents short-term prediction error and b represents historical prediction error. To find the minimum value, avoid having a denominator of 0.
6. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S3, the formula for calculating the target temperature setpoint is: in, The target temperature set for the region based on energy conservation and comfort objectives; The standard operating temperature for the functional area corresponds to the first-level thermal comfort rating range. This is the energy-saving temperature, which is the lower limit of the temperature range for Level 2 thermal comfort. The number of people is a combination of the real-time and predicted number of people; k is the proportion of the trigger zone that adjusts to a comfortable temperature. Design capacity for the area or the number of regular users.
7. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, In step S5, the anomaly detection process specifically includes: Set threshold ranges for personnel numbers, environmental parameters, and equipment operating parameters; By analyzing the characteristics of abnormal data through association rule algorithms, the root cause of sensor failures or network outages can be located, and a diagnostic report can be automatically generated.
8. The HVAC intelligent control method based on dynamic personnel prediction according to claim 1, characterized in that, The method further includes: when the area enters the manual control mode and continues for a third preset time period, automatically switching back to the intelligent control mode; if the area remains unoccupied for an extended period of time, the energy-saving temperature setting is triggered.
9. An intelligent HVAC control system based on dynamic personnel prediction, characterized in that, The system includes: The sensing module consists of multiple monitoring nodes, each of which integrates a visual sensor, an infrared sensor, a CO2 sensor, and a temperature and humidity sensor to collect real-time data on the number of people and environmental parameters. The prediction module, deployed on edge computing nodes, includes short-term prediction units and historical prediction units, which output a weighted and fused comprehensive predicted number of people through LSTM and linear regression models, respectively. The decision-making module calculates the target temperature setpoint based on real-time data and prediction results, and generates control commands through a PID algorithm. The execution module adjusts the operating parameters of the regional HVAC equipment according to the instructions and dynamically adjusts the power of the main air conditioning unit; The evaluation module uses machine learning to optimize control parameters, detect abnormal data, and generate fault reports.
10. The HVAC intelligent control system based on dynamic personnel prediction according to claim 9, characterized in that, The data transmission rules of the sensing module are as follows: real-time upload is triggered when the number of people changes, and temperature and humidity data are uploaded at a fixed frequency; the transmission format is: spatial code-timestamp-number of people-temperature-humidity, and the data is compressed and transmitted after noise reduction by Kalman filtering.