Intelligent heating and ventilation load regulation method and system based on big data analysis

Abstract

The application discloses a heating and ventilation load intelligent regulation method and system based on big data analysis, and relates to the technical field of industrial data analysis.The method comprises the following steps: collecting image sequences in a target area, identifying the number of users and the activity intensity of users, and obtaining the number of users and the activity intensity of users; obtaining a warm air regulation parameter space and a ventilation regulation parameter space for heating and ventilation load regulation, and generating first warm air regulation parameters and first ventilation regulation parameters; performing warm air comfort degree analysis and ventilation comfort degree analysis according to the first warm air regulation parameters and the first ventilation regulation parameters, in combination with the number of users and the activity intensity of users respectively, and obtaining first warm air comfort degree and first ventilation comfort degree; and performing iterative optimization of the warm air regulation parameters and the ventilation regulation parameters based on the first warm air comfort degree and the first ventilation comfort degree, obtaining optimal warm air regulation parameters and optimal ventilation regulation parameters, and performing heating and ventilation load regulation.The application effectively improves the comprehensive comfort degree of warm air and ventilation.

Description

Technical Field

[0001] This invention relates to the field of industrial data analysis technology, specifically to a method and system for intelligent control of HVAC loads based on big data analysis. Background Technology

[0002] With the rapid development of smart building and IoT technologies, the demand for energy saving and refined control of HVAC systems is becoming increasingly prominent. Currently, in most civil and public building scenarios, the control of heating and ventilation in HVAC systems is still mainly based on manual management by property management, and the control parameters are mostly set according to fixed time periods, experience thresholds, or uniform standards.

[0003] However, manual control cannot detect and match changes in the number of indoor users and the intensity of their activities in real time, which can easily lead to a disconnect between heating output and ventilation volume and actual demand. This directly results in indoor heating and ventilation comfort failing to meet the actual user experience, making it impossible to guarantee a stable and comfortable indoor environment and to achieve refined and intelligent control of HVAC load. Summary of the Invention

[0004] This invention provides a method and system for intelligent control of HVAC load based on big data analysis, aiming to solve the technical problem that insufficient environmental comfort is caused by the manual control of HVAC load in the prior art.

[0005] In view of the above problems, the present invention provides a method and system for intelligent control of HVAC load based on big data analysis.

[0006] In a first aspect, the present invention provides a method for intelligent control of HVAC load based on big data analysis, comprising:

[0007] Collect image sequences within the target area, perform user count identification and user activity intensity identification, and obtain the user count and user activity intensity; Obtain the heating air control parameter space and ventilation control parameter space for HVAC load control, and generate the first heating air control parameter and the first ventilation control parameter. Based on the first heating control parameters and the first ventilation control parameters, and combined with the number of users and the intensity of user activities, heating comfort analysis and ventilation comfort analysis are performed to obtain the first heating comfort and the first ventilation comfort. Based on the first heating comfort level and the first ventilation comfort level, the heating control parameters and ventilation control parameters are iteratively optimized to obtain the optimal heating control parameters and optimal ventilation control parameters, and then the HVAC load is controlled. In this process, comfort contradiction analysis and control contradiction analysis are performed, and the process is iterated.

[0008] Secondly, the present invention provides an intelligent control system for HVAC loads based on big data analysis, comprising: The user status recognition module is used to collect image sequences within the target area, perform user count recognition and user activity intensity recognition, and obtain the user count and user activity intensity. The initial parameter generation module is used to obtain the heating air control parameter space and ventilation control parameter space of HVAC load control, and generate the first heating air control parameter and the first ventilation control parameter. The comfort analysis module is used to perform heating comfort analysis and ventilation comfort analysis based on the first heating control parameters and the first ventilation control parameters, combined with the number of users and the intensity of user activities, to obtain the first heating comfort and the first ventilation comfort. The iterative optimization control module is used to iteratively optimize the heating and ventilation control parameters based on the first heating comfort level and the first ventilation comfort level to obtain the optimal heating and ventilation control parameters and to control the HVAC load. In this process, comfort contradiction analysis and control contradiction analysis are performed, and the process is iterated.

[0009] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a method and system for intelligent control of HVAC load based on big data analysis. It identifies the number of users and their activity intensity by collecting image sequences of the target area, accurately obtaining the real-time status of indoor occupants to provide a basis for control. It establishes a scientific control benchmark by acquiring the space of heating and ventilation control parameters and generating initial control parameters. Combining occupant status with the initial parameters, it conducts heating and ventilation comfort analysis to clarify the gap between current control and user needs. Based on the analysis of comfort and control contradictions, iterative parameter optimization is performed to obtain the optimal control parameters and implement HVAC load control. This effectively overcomes the lag and crudeness of traditional manual control, accurately matches the real-time needs of indoor occupants, improves heating and ventilation comfort, and achieves intelligent and precise control of HVAC load. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating the intelligent control method for HVAC load based on big data analysis provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the structure of the intelligent control system for HVAC load based on big data analysis provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: User status recognition module 11, initial parameter generation module 12, comfort analysis module 13, and iterative optimization and control module 14. Detailed Implementation

[0012] This invention provides a method and system for intelligent control of HVAC load based on big data analysis, which is used to address the technical problem that insufficient environmental comfort is caused by the manual control of HVAC load in the prior art.

[0013] Example 1, as Figure 1 As shown, this invention provides a method for intelligent control of HVAC load based on big data analysis, the method comprising: S100: Collect image sequences within the target area, identify the number of users and the intensity of user activity, and obtain the number of users and the intensity of user activity.

[0014] In this embodiment of the invention, image sequences within a target area are acquired to identify the number of users and the intensity of their activities, thus obtaining the number of users and the intensity of their activities. Traditional HVAC load control often relies on manual, fixed parameter control, which cannot perceive the status of people within the target area in real time. This easily leads to a mismatch between control parameters and actual needs: the more people there are and the greater their activity intensity, the faster the indoor carbon dioxide concentration rises, requiring increased ventilation and fresh air volume to ensure air quality; simultaneously, increased heat production by the human body necessitates reducing heating output to avoid excessively high indoor temperatures and stuffy conditions. If the number of users and their activity intensity cannot be accurately obtained, it can result in insufficient fresh air supply and excessive heating control, reducing indoor comfort and wasting energy. Therefore, it is necessary to first quantify the number of users and their activity intensity within the target area through image acquisition and intelligent recognition, providing a basis for subsequent coordinated control of heating and ventilation.

[0015] Step S100 in the method provided in this embodiment of the invention includes: Acquire image sequences within the target area, which is the area where HVAC load control is performed; User identification is performed on each image within the image sequence to obtain the user coordinate distribution sequence; Based on the user coordinate distribution sequence, the number of users and the intensity of user activity are calculated.

[0016] First, image sequences are acquired within the target area, which is the area where HVAC load control is implemented. The target area refers to indoor spaces requiring intelligent HVAC load control, such as offices, factory workshops, and public areas in shopping malls—enclosed or semi-enclosed spaces where people are active. The image sequence is a combination of multiple frames containing scenes of the target area, continuously acquired by monitoring equipment at fixed time intervals, reflecting dynamic changes in personnel. Through monitoring cameras deployed in the target area, image sequences of the current scene are continuously acquired in real time; simultaneously, historical image monitoring records of the area are retrieved, and images from different time periods and with different personnel densities are filtered to form a sample image set. For example, if the target area is 200m... 2 In an open-plan office, a ceiling-mounted surveillance camera continuously captures images for 30 seconds at a rate of 1 frame per second, resulting in an image sequence containing 30 frames.

[0017] Secondly, user identification is performed on each image in the image sequence to obtain the user coordinate distribution sequence.

[0018] Specifically, user identification is performed on each image within the image sequence to obtain a user coordinate distribution sequence, including: Based on the image monitoring records of the target area, a set of sample images is collected, and the user coordinates in each sample image are labeled to obtain a set of sample user coordinate distributions. A user identifier is constructed based on a convolutional neural network, which includes convolutional layers, pooling layers, and fully connected layers. The user recognizer is trained and tested using the set of sample images as training input and the set of sample user coordinate distributions as supervision of correct results. The training is completed after convergence. Each image in the image sequence is input into the user identifier, and the output is a user coordinate distribution sequence.

[0019] First, based on image monitoring records of the target area, a sample image set is collected. The user coordinates within each sample image are labeled, resulting in a sample user coordinate distribution set. The sample image set refers to a collection of historical images covering different numbers of people and different activity states within the target area, used for model training. User coordinates refer to the horizontal and vertical coordinates of a user's corresponding position in a single frame image, used to locate the user's position. The sample user coordinate distribution set is a summary of user coordinates from all sample images, representing the distribution pattern of personnel positions under different scenarios. For each image in the sample image set, the horizontal and vertical coordinates of each user are manually labeled. All user coordinates in a single image are organized into a coordinate distribution, and the sum of the coordinate distributions of all images forms the sample user coordinate distribution set.

[0020] For example, retrieve the surveillance footage of the office for the past month, and filter out 1000 images with 0-15 people and different activity statuses to form a sample image set. Label each of the 1000 sample images. For example, for an image containing 8 office workers, label it with 8 sets of user coordinates such as (120, 300), (250, 280), etc., and use these 8 sets of coordinates as a coordinate distribution. Summarize the coordinate distributions of the 1000 images to obtain the sample user coordinate distribution set.

[0021] Secondly, a user identifier is constructed based on a convolutional neural network (CNN), which includes convolutional layers, pooling layers, and fully connected layers. A CNN is a deep learning network suitable for image feature extraction, capable of automatically identifying people in images. The user identifier is a CNN-based model used to extract person features from the input image and output the user coordinate distribution. The convolutional, pooling, and fully connected layers are the main structures of the CNN. Convolutional layers extract person features such as image edges and contours; pooling layers compress feature data and improve computational efficiency; and fully connected layers integrate features and output the coordinate results.

[0022] Specifically, a convolutional neural network model is constructed using a hierarchical structure of stacked convolutional layers, pooling layers, and fully connected layers, and this model is encapsulated as a user identifier specifically for personnel localization. The convolutional layers are responsible for extracting features such as contours and textures of people in the image; the pooling layers compress the feature dimensions to reduce computational complexity; and the fully connected layers integrate global features and map them to the personnel coordinate output dimension. The core parameters and activation rules of each layer are as follows: The convolutional layers use 3×3 two-dimensional convolutional kernels with a stride of 1 and the same padding method, maintaining the feature map size consistent with the input. Each convolutional operation is followed by a ReLU activation function to introduce non-linear feature mapping. The pooling layers use 2×2 max-pooling kernels with a stride of 2, deployed at intervals after the convolutional layers, and have no activation function, only performing feature downsampling. The fully connected layers contain multiple hidden layers and one output layer. The hidden layers all use the ReLU activation function, while the output layer has no activation function and is directly mapped to personnel coordinate values.

[0023] For example, for 200m 2For a user location scenario in an open office, a convolutional neural network (CNN) is constructed, consisting of 3 convolutional layers, 2 pooling layers, and 2 fully connected hidden layers, to act as a user identifier. This network is specifically designed to identify people in office images and output their coordinates. The complete architecture parameters are as follows: Input layer: Receives an RGB format office surveillance image with a size of 640×480×3; First convolutional layer: Configured with 32 3×3 convolutional kernels, stride 1, uniform padding, ReLU activation, output feature map size 640×480×32; First pooling layer: 2×2 max pooling, stride 2, output feature map size 320×240×32; Second convolutional layer: Configured with 64 3×3 convolutional kernels, stride 1, uniform padding, ReLU activation, output feature map size 320×240×64; Second pooling layer: 2×2 max pooling, stride 2, output feature map size 1. 60×120×64; Third convolutional layer: configured with 128 3×3 convolutional kernels, stride 1, same padding, ReLU activation, output feature map size is 160×120×128; Flattening layer: converts the 160×120×128 feature map output by the third convolutional layer into a one-dimensional vector with a dimension of 160×120×128=2457600; First fully connected hidden layer: set with 2048 neurons, ReLU activation, output dimension is 2048; Second fully connected hidden layer: set with 512 neurons, ReLU activation, output dimension is 512; Output layer: set with 40 neurons, corresponding to a maximum of 20 people can be identified in a single frame image, each person outputs x and y two-dimensional coordinates, no activation function, directly output the coordinate values ​​of people in the office image such as (120,300), (250,280), etc.

[0024] Next, using the set of sample images as training input and the set of sample user coordinate distributions as supervision of correct results, the user recognizer undergoes supervised training and testing, completing training upon convergence. Supervised training refers to using sample images as input and the sample user coordinate distribution as the standard label to guide the model to learn the mapping relationship between image features and personnel coordinates. Model convergence means that the model's prediction error decreases to a preset threshold, and the deviation between the prediction result and the standard label stabilizes and no longer decreases significantly, indicating that training is complete.

[0025] Specifically, the sample image set is divided into a training set and a test set in an 8:2 ratio. The sample images in the training set are used as the model input, and the sample user coordinate distribution set is used as the supervision label. Mini-batch gradient descent combined with the Adam optimizer is used to iteratively train the user recognizer. During training, the mean squared error loss function (MSE) is used to quantify the deviation between the model's predicted coordinates and the actual labeled coordinates. Simultaneously, the model accuracy is verified using the test set after each round of training until the convergence condition is met, at which point training stops. The convergence condition is as follows: the mean absolute error between the model's predicted coordinates and the labeled coordinates on the test set is consistently lower than a preset threshold, and the fluctuation range of the loss value is less than 0.001 in 30 consecutive iterations, or the number of iterations reaches a preset maximum. The model is considered to have converged if either condition is met.

[0026] For example, 1000 sample images are divided into an 8:2 ratio: 800 training images and 200 test images. Using the 800 training images as input and the corresponding 800 user coordinate distributions as labels, the user recognizer is trained using mini-batch gradient descent (batch size set to 32) + Adam optimizer (learning rate set to 0.001). The loss function is mean squared error (MSE). Assuming a batch of 32 office sample images with 128 labeled coordinates, and the sum of squared deviations between the model's predicted coordinates and the actual labeled coordinates is 51.2, then the MSE loss for that batch is 51.2 / 128 = 0.4. After each training round, 200 test images are used to verify the accuracy, and the mean absolute error between the predicted and labeled coordinates is calculated. When the test set error is consistently below 2% for 30 consecutive rounds, and the loss value gradually decreases from the initial 5.2 to 0.08 with a fluctuation range of less than 0.001, the model is considered converged, training is stopped, and the user recognizer training is complete.

[0027] Then, each image in the image sequence is input into the user recognizer, and a user coordinate distribution sequence is output. The user coordinate distribution sequence refers to the user coordinate distribution corresponding to each frame of the real-time image sequence, arranged in chronological order to reflect the dynamic changes in personnel positions. The real-time image sequence is input frame by frame into the trained user recognizer, and the user recognizer outputs the user coordinate distribution frame by frame, sorted by acquisition time to form the user coordinate distribution sequence. For example, 30 frames of real-time images from an office are input frame by frame into the user recognizer, each frame outputting the corresponding personnel coordinate distribution; the 30 coordinate distributions are sorted by time to obtain the user coordinate distribution sequence.

[0028] Furthermore, based on the user coordinate distribution sequence, the number of users and the intensity of user activity are calculated.

[0029] The number of users and the intensity of user activity are calculated based on the user coordinate distribution sequence, including: Based on the user coordinate distribution sequence, the number of users is counted to obtain the user coordinate number sequence, and the average value is calculated to obtain the number of users. Based on the user coordinate distribution sequence, the two user coordinates with the smallest distance within the adjacent user coordinate distribution are paired, and the coordinate change amplitude is calculated to obtain multiple sets of user coordinate change amplitudes. The average value is then calculated to obtain the user activity intensity.

[0030] First, based on the user coordinate distribution sequence, user counts are performed to obtain a user coordinate quantity sequence, and the average value is calculated to obtain the total number of users. The user coordinate quantity sequence refers to the sequence of the number of people corresponding to each frame of the image in the user coordinate distribution sequence, arranged chronologically. The user coordinate distribution sequence is traversed, and the number of user coordinate groups for each frame is counted to obtain the user coordinate quantity sequence. The average value of this sequence is calculated as the real-time number of users in the target area. For example, if the user coordinate quantities of 30 frames are 7, 8, 8, 7…9 respectively, and the average value is 8, then the real-time number of users in the office is 8.

[0031] Secondly, based on the user coordinate distribution sequence, the two user coordinates with the smallest distance within adjacent user coordinate distributions are paired, and the coordinate change amplitude is calculated to obtain multiple sets of user coordinate change amplitudes. The average value is then calculated to obtain the user activity intensity. Coordinate change amplitude refers to the displacement distance of the corresponding coordinates of the same user in two adjacent image frames, representing the degree of movement of a single person. User activity intensity is the average value of all user coordinate change amplitudes; a larger value indicates more frequent overall movement and activity of the personnel. Adjacent user coordinates in two frames are paired according to the minimum distance, and the displacement distance of each pair of coordinates, i.e., the coordinate change amplitude, is calculated. All change amplitudes are summarized and the average value is calculated to obtain the user activity intensity. For example, in a 30-frame coordinate distribution sequence, adjacent frames are paired, and the displacement distance of each pair of coordinates is calculated (e.g., 5 pixels, 10 pixels, etc.). The average value of all displacement distances is 12 pixels, indicating that the user activity intensity in this office is moderate.

[0032] In this embodiment of the invention, by image acquisition, convolutional neural network recognition and data statistics, the number of users and the intensity of user activities in the target area can be accurately and in real time quantified, eliminating the error of subjective human judgment; the quantification results directly match the HVAC control logic, providing objective and dynamic data support for increasing fresh air volume and optimizing heating parameters, solving the problem that traditional control cannot perceive the real-time status of people, and laying the foundation for improving indoor comfort and reducing energy consumption.

[0033] S200: Obtain the heating air control parameter space and ventilation control parameter space for HVAC load control, and generate the first heating air control parameter and the first ventilation control parameter.

[0034] In this embodiment of the invention, the space of heating control parameters and the space of ventilation control parameters for HVAC load regulation are obtained, and first heating control parameters and first ventilation control parameters are generated. S100 has accurately obtained the number of users and activity intensity within the target area. Intelligent HVAC load regulation requires precise matching of heating and ventilation through specific parameters. The core parameter for heating control is temperature, and the core parameter for ventilation control is the fresh air volume per unit time and space. Traditional manual regulation lacks a clear parameter control range, relying solely on experience to set fixed initial parameters. This fails to generate first-generation control parameters adapted to the real-time status of personnel, easily leading to a lack of reasonable benchmarks for subsequent comfort analysis. Therefore, it is necessary to first define the adjustable parameter space for heating and ventilation, and based on adjustments to the initial parameters from the previous period, generate first-generation first heating and ventilation control parameters, providing a quantifiable and adjustable initial benchmark for subsequent comfort analysis and iterative optimization.

[0035] Step S200 in the method provided in this embodiment of the invention includes: Obtain the space of heating air control parameters and ventilation control parameters for HVAC load regulation; Obtain the initial heating and ventilation control parameters set in the previous time period; Within the space of heating control parameters and ventilation control parameters, the initial heating control parameters and initial ventilation control parameters are adjusted by adjusting the step size to obtain the first heating control parameters and the first ventilation control parameters.

[0036] First, obtain the heating air control parameter space and ventilation control parameter space for HVAC load regulation. The heating air control parameter space refers to the operable range of the HVAC system's heating air regulation, i.e., the range of indoor temperatures that the HVAC system can stably output. This needs to be determined in conjunction with the rated power of the HVAC equipment and the insulation performance of the target area. The ventilation control parameter space refers to the operable range of the HVAC system's ventilation regulation, i.e., the range of fresh air volume that the HVAC system can deliver to the target area per unit time. This needs to be matched with the rated airflow of the ventilation equipment. Retrieve the equipment specifications of the HVAC system in the target area, and combine them with industry indoor HVAC regulation standards to determine the reasonable adjustable range of the heating air temperature, thus obtaining the heating air control parameter space; determine the reasonable adjustable range of the fresh air volume per unit time, thus obtaining the ventilation control parameter space; ensure that the parameter range is within the rated load capacity of the equipment and meets the basic requirements of the indoor environment.

[0037] For example, retrieve 200m 2 HVAC equipment specifications for open-plan offices: Heaters with a rated temperature adjustment range of 18-28℃; fresh air system with a rated air volume of 200-800 m³ / h. 3 / h; Based on indoor environmental standards for office scenarios, the heating control parameters for the space are determined to be 18-26℃, and the ventilation control parameters for the space are determined to be 200-800m. 3 / h.

[0038] Secondly, obtain the initial heating and ventilation control parameters set in the previous time period. The previous time period refers to the preceding continuous control period of the current control cycle; for example, it can be set to 1 hour and adjusted according to actual control needs. The initial heating temperature control parameter refers to the heating temperature setpoint during the stable operation of the HVAC system in the previous time period, which is a historical parameter within the heating temperature control parameter space. The initial ventilation control parameter refers to the fresh air volume setpoint per unit time during the stable operation of the HVAC system in the previous time period, which is a historical parameter within the ventilation control parameter space. By querying the HVAC system's control record database, the heating temperature and fresh air volume setpoints during the stable operation of the system in the previous time period are used as the initial parameters for this control, ensuring that the initial parameters are within the corresponding parameter space. If they exceed the parameter space, the boundary values ​​of the parameter space are used. For example, querying the 200m... 2 The HVAC control records for the open-plan office were from one hour ago. The initial heating control parameters were 24°C and the initial ventilation control parameters were 300m³. 3 / h.

[0039] Furthermore, within the aforementioned heating and ventilation control parameter spaces, the initial heating and ventilation control parameters are adjusted using an adjustment step size to obtain the first heating and ventilation control parameters. The adjustment step size refers to the smallest unit of parameter adjustment, used for precise fine-tuning of the initial parameters to ensure that the generated first-generation parameters have a reasonable difference from the initial parameters, avoiding large fluctuations that could affect the stability of the indoor environment. Specifically, the heating adjustment step size is in units of 0.5°C, and the ventilation adjustment step size is in units of 50m. 3 The unit is / h. The first heating control parameter and the first ventilation control parameter refer to the first generation of control parameters generated by adjusting the initial parameters of the previous period within the corresponding parameter space.

[0040] Specifically, within the heating control parameter space, the initial heating parameters of the previous period are fine-tuned using the temperature adjustment step size; within the ventilation control parameter space, the initial ventilation parameters of the previous period are fine-tuned using the fresh air volume adjustment step size. The direction of fine-tuning can be initially adapted to the number of users and activity intensity obtained by S100. For example, when there are many people or the activity intensity is high, the heating temperature can be appropriately reduced and the fresh air volume can be increased. The parameters obtained after fine-tuning are the first heating control parameters and the first ventilation control parameters, ensuring that the parameters are still within the corresponding parameter space.

[0041] For example, based on the information obtained by S100 regarding the current number of users in the office (8 people), moderate activity level, increased heat generation, and the need for more fresh air, the initial parameters are adjusted using a set step size: Heating parameter adjustment: initial 24℃, step size 0.5℃, appropriately lowering the temperature to 23.5℃, which is used as the first heating parameter; Ventilation parameter adjustment: initial 300m... 3 / h, step size 50m 3 / h, appropriately increase the fresh air volume to 350m³. 3 / h is used as the first ventilation control parameter.

[0042] In this embodiment of the invention, by clearly defining the parameter space for heating and ventilation, the problem of control parameters exceeding the equipment's capacity or failing to meet environmental standards is avoided. Based on the initial parameters of the previous period, precise fine-tuning is performed to generate the first heating control parameters and the first ventilation control parameters. This ensures the stability of the indoor environment and initially adapts to the number of users and activity intensity obtained by S100. It provides a specific and quantifiable benchmark for the subsequent comfort analysis of S300, solving the shortcomings of traditional manual control that lacks reasonable initial parameters and cannot adapt to the real-time status of personnel, and providing a foundation for subsequent parameter iteration and optimization.

[0043] S300: Based on the first heating control parameters and the first ventilation control parameters, and combined with the number of users and the intensity of user activities, perform heating comfort analysis and ventilation comfort analysis to obtain the first heating comfort and the first ventilation comfort.

[0044] In this embodiment of the invention, based on the first heating control parameters and the first ventilation control parameters, and combined with the number of users and the intensity of user activity, heating comfort analysis and ventilation comfort analysis are performed to obtain the first heating comfort and the first ventilation comfort. S100 obtains the number of users and activity intensity in the target area, and S200 generates the first-generation first heating control parameters and the first ventilation control parameters. However, whether these two sets of parameters can meet the users' heating and ventilation comfort needs lacks quantitative analysis. Traditional manual control relies solely on subjective feelings to judge comfort, failing to distinguish between the individual comfort levels of heating and ventilation, and also failing to quantify the degree of comfort attainment, easily leading to a lack of clear direction for subsequent parameter optimization. Therefore, it is necessary to first map the actual operating values ​​corresponding to the parameters through the HVAC control data table, and then use a specially configured HVAC comfort analysis intelligent agent to analyze the comfort levels of heating and ventilation separately, quantifying the first heating comfort and the first ventilation comfort. This provides an accurate judgment basis for the subsequent parameter iteration optimization in S400, ensuring that the optimization direction aligns with the actual comfort needs of users.

[0045] Step S300 in the method provided in this embodiment of the invention includes: Based on the HVAC control data table, the first temperature mapped by the first heating control parameter and the first fresh air volume mapped by the first ventilation control parameter are obtained. The HVAC control data table is constructed based on the mapping relationship between the sample heating control parameter set and the sample temperature set, and the mapping relationship between the sample ventilation control parameter set and the sample fresh air volume set. The first temperature and the first fresh air volume are combined with the number of users and the intensity of user activity, respectively, and input into the HVAC comfort analysis agent to obtain the first heating comfort and the first ventilation comfort. The HVAC comfort analysis agent includes a heating comfort analysis network and a ventilation comfort analysis network.

[0046] First, based on the HVAC control data table, the first temperature mapped by the first heating control parameter and the first fresh air volume mapped by the first ventilation control parameter are obtained. The HVAC control data table is constructed based on the mapping relationship between the sample heating control parameter set and the sample temperature set, and the mapping relationship between the sample ventilation control parameter set and the sample fresh air volume set.

[0047] First, a heating, ventilation, and air conditioning (HVAC) control data table is constructed. This HVAC control data table is built based on the mapping relationships between sample sets of heating control parameters and sample sets of temperatures, and between sample sets of ventilation control parameters and sample sets of fresh air volumes. The HVAC control data table is a dataset used to store the mapping relationships between heating control parameters and temperatures, and between ventilation control parameters and fresh air volumes. Its function is to achieve a precise correspondence between control parameters and actual operating values, avoiding discrepancies between parameters and actual outputs. The sample set of heating control parameters refers to the summary of different set parameters in the historical heating control of the target area, such as various temperature set values ​​within the range of 18-26℃. The sample set of temperatures is a summary of the actual indoor temperatures output by the HVAC system, corresponding one-to-one with the sample heating control parameters. Similarly, the sample sets of ventilation control parameters and fresh air volumes are summaries of the correspondence between ventilation set parameters and actual fresh air volumes.

[0048] Specifically, historical control data of the HVAC system in the target area are retrieved, sample heating air control parameters and corresponding actual output temperatures are collected to form a mapping pair of sample heating air control parameters and actual output temperatures; sample ventilation control parameters and corresponding actual output fresh air volume are collected to form a mapping pair of ventilation control parameters and actual output fresh air volume; the two types of mapping pairs are sorted and summarized to construct a standardized HVAC control data table to ensure that each control parameter can be accurately mapped to a unique actual operating value.

[0049] For example, retrieve the historical HVAC control data for the open-plan office over the past three months, collect sample heating control parameters and corresponding actual output temperatures, collect sample ventilation control parameters and corresponding actual fresh air volumes, and compile them into an HVAC control data table. An example of the mapping relationship is: heating control parameter 23.5℃: actual first temperature 23.5℃; ventilation control parameter 350m³ / h. 3 / h: Actual first fresh air volume 350m³ 3 / h; ensure that the control parameters accurately correspond to the actual operating values.

[0050] Secondly, based on the HVAC control data table, the first temperature mapped to the first heating air control parameter and the first fresh air volume mapped to the first ventilation control parameter are obtained. The first temperature refers to the actual indoor temperature that the HVAC system can output, obtained by mapping the first heating air control parameter through the HVAC control data table, and precisely corresponds to the first heating air control parameter. The first fresh air volume refers to the actual fresh air volume that the HVAC system can output per unit time, obtained by mapping the first ventilation control parameter through the HVAC control data table, and precisely corresponds to the first ventilation control parameter. The constructed HVAC control data table is queried, and the first heating air control parameter generated by S200 is input, extracting its mapped actual temperature as the first temperature; the first ventilation control parameter is generated by inputting S200, extracting its mapped actual fresh air volume as the first fresh air volume.

[0051] For example, querying the above HVAC control data table, inputting the first heating air control parameter of 23.5℃ generated by S200, the actual output first temperature of 23.5℃ is obtained; inputting the first ventilation control parameter of 350m 3 / h, mapped to the actual output first fresh air volume of 350m³ 3 / h, these two sets of actual operating values ​​are used for subsequent precise analysis of heating and ventilation comfort.

[0052] The configuration steps of the HVAC comfort analysis intelligent agent include: Based on the historical HVAC control data of the target area, sample temperature set, sample fresh air volume set, sample number of users set, sample user activity intensity set, and sample user comfort set are collected. A machine learning-based heating comfort analysis network and a ventilation comfort analysis network are constructed. Both the heating comfort analysis network and the ventilation comfort analysis network include an input layer, a hidden layer and an output layer. The hidden layer contains multiple processing nodes. Using the sample temperature set, sample fresh air volume set, and sample user number set as inputs, and the sample user comfort set as labels, the heating comfort analysis network is subjected to supervised training and testing until convergence. Using the sample fresh air volume set, sample user number set, and sample user activity intensity set as inputs, and the sample user comfort set as labels, the ventilation comfort analysis network is subjected to supervised training and testing until convergence. By combining convergent machine learning networks for heating comfort analysis and ventilation comfort analysis, a heating, ventilation, and air conditioning comfort analysis agent is obtained.

[0053] First, based on historical HVAC control data for the target area, sample temperature sets, sample fresh air volume sets, sample user quantity sets, sample user activity intensity sets, and sample user comfort sets are collected. Historical HVAC control data refers to all data on the control parameters, actual operating status, personnel status, and user comfort feedback of the HVAC system in the target area over a past period. The sample user comfort set is a summary of comfort scores collected through user surveys and sensory perception tests for different sample temperatures, sample fresh air volumes, and sample personnel statuses, quantified using a 0-10 scale, with higher scores indicating better comfort. The sample temperature sets, sample fresh air volume sets, sample user quantity sets, and sample user activity intensity sets are all extracted from the historical HVAC control data and correspond one-to-one with the sample user comfort sets, used for training the HVAC comfort analysis agent.

[0054] Specifically, historical HVAC control data for the target area is retrieved, and various data types are extracted according to the correspondence between temperature, fresh air volume, number of users, user activity intensity, and comfort score. These data are then organized into sample temperature sets, sample fresh air volume sets, sample number of users sets, sample user activity intensity sets, and corresponding sample user comfort sets. This ensures that the number of samples for each type is sufficient, covers different scenarios, and meets the training needs of the intelligent agent.

[0055] For example, retrieve the office's historical HVAC control data for the past three months and extract various sample sets, such as: sample temperature set {18℃, 18.5℃, ..., 26℃}, and sample fresh air volume set {200m³ / s}. 3 / h,250m 3 / h,…,800m 3 The sample user set is {0, 1, ..., 15}, and the sample user activity intensity set is {5 pixels, 10 pixels, ..., 12 pixels}. Simultaneously, user comfort scores for the corresponding scenarios are extracted to form a sample user comfort set, such as 23.5℃-350m. 3 / h-8 people-12 pixels corresponds to a user comfort score of 8 points, and all sample sets correspond one-to-one.

[0056] Secondly, machine learning-based heating comfort analysis networks and ventilation comfort analysis networks are constructed. Both networks include an input layer, a hidden layer, and an output layer, with multiple processing nodes within the hidden layer. The HVAC comfort analysis agent is a machine learning-based intelligent analysis model composed of two independent sub-networks: the heating comfort analysis network and the ventilation comfort analysis network. These sub-networks analyze the comfort of heating and ventilation respectively, outputting quantitative scores. The heating comfort analysis network specifically analyzes the impact of temperature and occupant status on heating comfort, with inputs including temperature, fresh air volume, and number of users, and outputting a heating comfort score. The ventilation comfort analysis network specifically analyzes the impact of fresh air volume and occupant status on ventilation comfort, with inputs including fresh air volume, number of users, and user activity intensity, and outputting a ventilation comfort score. The input, hidden, and output layers form the main structure of the machine learning network. The input layer receives various parameters required for analysis, the hidden layer performs feature integration and analysis on the parameters, and the output layer outputs a quantitative comfort score.

[0057] Specifically, two structurally consistent and independently operating machine learning networks are constructed, using BP neural networks. Each network contains an input layer, a hidden layer, and an output layer. The hidden layer has two layers, each containing 128 processing nodes, and uses the ReLU activation function. The number of nodes in the input and output layers is set according to their respective analysis needs. Finally, the two sub-networks are combined to form a complete HVAC comfort analysis intelligent agent.

[0058] For example, two independent backpropagation (BP) neural networks are constructed to form an intelligent agent for HVAC comfort analysis. The heating comfort analysis network has three input layers corresponding to sample temperature, sample fresh air volume, and sample number of users; two hidden layers with 128 processing nodes per layer, activated by ReLU; and one output layer with one node, corresponding to the output heating comfort score. The ventilation comfort analysis network has three input layers corresponding to sample fresh air volume, sample number of users, and sample activity intensity; two hidden layers with 128 processing nodes per layer, activated by ReLU; and one output layer with one node, corresponding to the output ventilation comfort score. The two sub-networks operate independently and do not interfere with each other, focusing respectively on the analysis of heating comfort and ventilation comfort.

[0059] Next, using the aforementioned sample temperature set, sample fresh air volume set, and sample user number set as inputs, and the sample user comfort set as labels, the heating comfort analysis network is subjected to supervised training and testing until convergence. Supervised training refers to using various sample sets as inputs and the sample user comfort set as standard labels to guide the network to learn the mapping relationship between input parameters and comfort scores, ensuring the accuracy of the comfort scores output by the network. Convergence is consistent with the model convergence definition in S100, meaning that the error between the heating comfort score predicted by the network and the sample labels is consistently lower than a preset threshold, and the loss value fluctuation is less than 0.001 for 30 consecutive iterations, or the maximum number of iterations is reached, at which point training is considered complete.

[0060] Specifically, a mini-batch gradient descent combined with the Adam optimizer is used. The set of sample temperatures, sample fresh air volumes, and sample user numbers are used as inputs, and the set of sample user comfort levels is used as the supervision label to iteratively train the heating comfort analysis network. After each training round, the accuracy is verified using a test set until the network converges and the training is completed.

[0061] For example, various sample sets are divided into training and testing sets in an 8:2 ratio. The inputs to the training set are sample temperature, sample fresh air volume, and sample number of users, and the supervision label is the user comfort score. The heating comfort analysis network is trained with a batch size of 32 and a learning rate of 0.001. When the error of the test set is less than 2% for 30 consecutive rounds and the fluctuation of the loss value is less than 0.001, the heating comfort analysis network converges, the training is completed, and the heating comfort score can be accurately output.

[0062] Furthermore, using the sample fresh air volume set, sample user number set, and sample user activity intensity set as inputs, and the sample user comfort set as labels, the ventilation comfort analysis network is subjected to supervised training and testing until convergence. Using the same training method as the heating comfort analysis network, iterative training is performed using the sample fresh air volume set, sample user number set, and sample user activity intensity set as inputs, and the sample user comfort set as supervision labels, with simultaneous accuracy verification using a test set, until the network converges, completing the training. The training set inputs are sample fresh air volume, sample user number, and sample activity intensity, with the supervision label being the user comfort score; using the same training parameters as the heating network, the ventilation comfort analysis network is trained, and training is complete when the convergence condition is met, accurately outputting ventilation comfort scores.

[0063] Finally, the converged machine learning networks for heating comfort analysis and ventilation comfort analysis are combined to obtain an intelligent agent for HVAC comfort analysis. The converged heating comfort analysis network and ventilation comfort analysis network are encapsulated and integrated, their respective input and output interfaces are clearly defined, ensuring that after inputting the corresponding parameters, the two sub-networks can run independently and synchronously output their respective comfort scores. After integration, this becomes a usable intelligent agent for HVAC comfort analysis.

[0064] Based on this, the first temperature and the first fresh air volume, combined with the number of users and the intensity of user activity, are input into the HVAC comfort analysis agent, resulting in the output of the first heating comfort score and the first ventilation comfort score. The HVAC comfort analysis agent includes a heating comfort analysis network and a ventilation comfort analysis network. The first heating comfort score is a quantitative comfort score output by the heating comfort analysis network after analyzing the current first temperature, first fresh air volume, and number of users, reflecting the degree of suitability of the current heating parameters. The first ventilation comfort score is a quantitative comfort score output by the ventilation comfort analysis network after analyzing the current first fresh air volume, number of users, and intensity of user activity, reflecting the degree of suitability of the current ventilation parameters.

[0065] Specifically, the first temperature, the first fresh air volume, and the number of users and user activity intensity obtained by S100 are simultaneously input into the trained HVAC comfort analysis agent. The two sub-networks in the HVAC comfort analysis agent analyze independently. The heating analysis network outputs the first heating comfort level, and the ventilation analysis network outputs the first ventilation comfort level. The two sets of scores are recorded for subsequent optimization.

[0066] For example, the first temperature is 23.5℃ and the first fresh air volume is 350m³. 3 The input parameters are: / h, 8 users, and moderate activity intensity. The heating comfort analysis network outputs a first heating comfort score of 8.2; the ventilation comfort analysis network outputs a first ventilation comfort score of 7.8. Both scores are quantitative values, clearly reflecting the current comfort levels of the heating and ventilation parameters, providing a clear basis for subsequent parameter iteration and optimization.

[0067] In this embodiment of the invention, by constructing a HVAC control data table, a precise mapping between control parameters and actual operating values ​​is achieved, avoiding errors caused by analyzing comfort using theoretical parameters. By configuring the HVAC comfort analysis agent step-by-step, heating and ventilation comfort are analyzed separately. The training method of these two independent sub-networks ensures the relevance and accuracy of the comfort analysis. The final obtained first heating comfort and first ventilation comfort quantify the adaptation effect of the current first-generation control parameters, clearly presenting the respective comfort shortcomings of heating and ventilation. This provides a basis for judging the parameter iteration and optimization of the S400, ensuring that subsequent optimizations accurately meet users' dual comfort needs for heating and ventilation, and laying the foundation for the coordinated optimization of heating and ventilation.

[0068] S400: Based on the first heating comfort level and the first ventilation comfort level, iterative optimization of heating and ventilation control parameters is performed to obtain the optimal heating and ventilation control parameters, and then the HVAC load is controlled. In this process, comfort contradiction analysis and control contradiction analysis are performed, and the process is iterated.

[0069] In this embodiment of the invention, based on the first heating comfort level and the first ventilation comfort level, the heating control parameters and ventilation control parameters are iteratively optimized to obtain the optimal heating control parameters and optimal ventilation control parameters, and then the HVAC load is controlled. This involves comfort conflict analysis and control conflict analysis, followed by iterative optimization. S300 has already obtained the first heating comfort level and the first ventilation comfort level, but there are numerical differences between the two comfort levels. Furthermore, changes in outdoor temperature and ventilation operation can remove indoor heat, directly interfering with the heating output effect, resulting in a deviation between the preset heating control parameters and the actual required heating parameters under the coupling effect of outdoor and ventilation. Therefore, it is necessary to first quantify the comfort conflict and the control conflict to obtain comprehensive conflict parameters; then, the fitness correction coefficient and convergence tolerance are dynamically configured through the conflict parameters. The larger the conflict, the lower the fitness and the stricter the convergence judgment; subsequently, the control parameters are iteratively adjusted with the corrected comfort level as the target until the convergence condition is met, ultimately obtaining the optimal heating and ventilation control parameters that balance comfort and control coordination, achieving precise and coordinated control of the HVAC load.

[0070] Step S400 in the method provided in this embodiment of the invention includes: Based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort contradiction analysis and a control contradiction analysis are performed to obtain the contradiction parameters. Based on the contradictory parameters, an fitness correction coefficient is configured, and the first warm air comfort and the first ventilation comfort are corrected and calculated to obtain the first comfort level. Obtain a preset convergence tolerance, configure a convergence tolerance adjustment coefficient according to the contradictory parameters, and adjust the preset convergence tolerance to obtain the convergence tolerance; Continue to adjust the first heating air control parameters and the first ventilation control parameters to obtain the second heating air control parameters and the second ventilation control parameters, and continue to analyze and calculate the second comfort level and update the convergence tolerance; The optimization process continues until the latest convergence tolerance is met or the convergence iteration count is reached, at which point the optimization is complete, and the optimal heating and ventilation control parameters that maximize comfort are obtained.

[0071] First, based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameter and the first ventilation control parameter, a comfort contradiction analysis and a control contradiction analysis are performed to obtain the contradiction parameters.

[0072] Specifically, based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort conflict analysis and a control conflict analysis are performed to obtain conflicting parameters, including: Calculate the difference between the first heating comfort level and the first ventilation comfort level to obtain the comfort contradiction parameter; Obtain the first heating power and the first ventilation power corresponding to the first heating control parameters and the first ventilation control parameters; The outdoor temperature of the target area is obtained, and the first heating power and the first ventilation power are combined with the input of the HVAC load impact analysis agent to obtain the first predicted heating parameters. The HVAC load impact analysis agent is constructed based on machine learning, and the training data includes a sample outdoor temperature set, a sample ventilation load set, and a sample heating load set as inputs, and a sample predicted heating parameter set as outputs. The deviation between the first predicted heating air parameter and the first heating air control parameter is calculated and used as the control contradiction parameter; Based on the aforementioned comfort-related conflict parameters and control-related conflict parameters, the conflict parameters are calculated.

[0073] First, the difference between the first heating comfort level and the first ventilation comfort level is calculated to obtain the comfort conflict parameter. The comfort difference range is the absolute difference between the first heating comfort level and the first ventilation comfort level, used to characterize the degree of mismatch between the two types of comfort. The larger the difference, the more uncoordinated the heating and ventilation comfort levels are. The comfort conflict parameter is a conflict index directly quantified by the comfort difference range, reflecting the degree of conflict between heating and ventilation in terms of perceived comfort. The first heating comfort level and the first ventilation comfort level output by S300 are read, and the absolute difference between them is calculated. This difference is used as the comfort conflict parameter; the larger the parameter value, the more prominent the comfort conflict.

[0074] For example, the first heating comfort score is 8.2 points, and the first ventilation comfort score is 7.8 points; the comfort difference range = |8.2-7.8| = 0.4; therefore, the comfort inconsistency parameter = 0.4, indicating that there is a slight inconsistency between the current heating and ventilation comfort.

[0075] Secondly, obtain the first heating power and first ventilation power corresponding to the first heating control parameters and first ventilation control parameters. Heating power refers to the operating power required by the heating system to reach the target temperature; it is positively correlated with temperature, and the higher the temperature setting, the greater the required heating power. Ventilation power refers to the operating power required by the fresh air system to reach the target fresh air volume; it is positively correlated with the fresh air volume per unit time, and the greater the fresh air volume, the greater the required ventilation power. Retrieve the power characteristic curves of the HVAC equipment, and map the first heating control parameters generated by S200 to the first heating power, and the first ventilation control parameters to the first ventilation power, providing a power basis for subsequent load impact analysis. For example, the equipment mapping relationship for the target area: first heating control parameter 23.5℃ corresponds to first heating power = 3.2kW; first ventilation control parameter 350m... 3 / h corresponds to a first ventilation power of 0.9kW.

[0076] Further, the outdoor temperature of the target area is obtained, and combined with the first heating power and the first ventilation power, the HVAC load impact analysis agent is input to obtain the first predicted heating parameters. The HVAC load impact analysis agent is constructed based on machine learning. The training data includes a set of sample outdoor temperatures, a set of sample ventilation loads, and a set of sample heating loads as inputs, and a set of sample predicted heating parameters as outputs. The outdoor temperature is the real-time temperature of the external environment of the target area. In low-temperature environments, ventilation fresh air will bring in cold air, lower the indoor temperature, and increase the heating load demand. The HVAC load impact analysis agent is a regression model constructed based on machine learning, used to learn the mapping relationship between outdoor temperature, heating power, ventilation power, and reasonable heating parameters, quantifying the coupled influence of outdoor temperature and ventilation on heating regulation. The first predicted heating parameters refer to the theoretically reasonable heating temperature, adapted to outdoor-ventilation interference, output by the agent after comprehensive calculation based on outdoor temperature, heating power, and ventilation power.

[0077] The construction of the intelligent agent for HVAC load impact analysis includes: First, the HVAC load impact analysis agent is constructed using a three-layer fully connected neural network. The overall structure consists of an input layer, a hidden layer, and an output layer, clearly adapting to the prediction requirements of the impact of outdoor temperature and HVAC power on heating parameters. The input layer contains three neurons, corresponding to the sample outdoor temperature, sample heating power, and sample ventilation power, respectively, used to receive three types of feature parameters. The hidden layer has two layers: the first hidden layer has 128 neurons, and the second hidden layer has 64 neurons. Both hidden layers use the ReLU activation function to extract the coupling features between input parameters, complete the nonlinear mapping, and avoid gradient vanishing. The output layer contains one neuron with no activation function, directly outputting the sample predicted heating parameters to ensure the continuity and accuracy of the prediction results.

[0078] Secondly, the training datasets are all derived from historical HVAC system control data of the target area, ensuring that the samples closely match actual application scenarios: the outdoor temperature sample set is extracted from historical outdoor temperature monitoring records of the target area, covering temperature ranges of different seasons and time periods; the heating power sample set and ventilation power sample set are extracted from historical operating data of the HVAC system, corresponding to the actual operating power under different heating and ventilation control parameters; the predicted heating parameter sample set is a summary of the optimal heating temperatures suitable for corresponding outdoor temperatures, heating power, and ventilation power in historical control, serving as supervision labels for model training. The training method employs mini-batch gradient descent combined with the Adam optimizer, dividing the three types of sample input sets into training and test sets in an 8:2 ratio, setting the batch size to 32 and the learning rate to 0.001. The model training is completed through an iterative process of forward propagation to extract features, calculate loss, and backpropagation to update weights.

[0079] Secondly, the mean squared error loss function (MSE) is used for model training to quantify the deviation between the model's predicted heating parameters and the sample's predicted heating parameters. The smaller the loss value, the higher the model's prediction accuracy. Its calculation logic is the mean of the squared deviations between the predicted value and the label value, which is suitable for the regression prediction requirements of continuous values. The convergence condition is explicitly set as the model's prediction error being stably below 2%, supplemented by dual constraints: the fluctuation range of the loss value is less than 0.001 in 30 consecutive iterations, or the number of iterations reaches the preset maximum. If either condition is met, the model training is considered to have converged, and iteration is stopped. This ensures that the trained model can accurately capture the mapping relationship between outdoor temperature, HVAC power, and heating parameters, while avoiding overfitting or insufficient training, and can stably output predicted heating parameters that fit actual needs.

[0080] Then, the outdoor temperature of the target area is obtained, and the first heating power and the first ventilation power are input into the HVAC load impact analysis agent to output the first predicted heating parameter. For example, the current real-time outdoor temperature is 5℃; the outdoor temperature of 5℃, the heating power of 3.2kW, and the ventilation power of 0.9kW are input into the HVAC load impact analysis agent; the first predicted heating parameter is output as 24.3℃.

[0081] Subsequently, the deviation range between the first predicted heating air parameter and the first heating air control parameter is calculated, serving as the control conflict parameter. The control deviation range refers to the absolute difference between the first predicted heating air parameter and the first heating air control parameter, reflecting the degree of mismatch between the preset heating air parameters after external interference. The control conflict parameter is a conflict index quantified by the control deviation range, characterizing the degree of conflict between the preset control parameters and the actual load demand. The absolute difference between the first predicted heating air parameter and the first heating air control parameter is calculated, and this difference is used as the control conflict parameter; the larger the difference, the more severe the control conflict.

[0082] For example, the first predicted heating parameter is 24.3℃, and the first heating control parameter is 23.5℃; the control deviation range is |24.3-23.5|=0.8℃; therefore, the control contradiction parameter is 0.8, which indicates that there is a significant deviation between the preset heating parameter and the outdoor-ventilation coupling demand.

[0083] Furthermore, based on the aforementioned comfort-related conflict parameters and control-related conflict parameters, conflict parameters are calculated. The comprehensive conflict parameter refers to the total conflict index obtained after balancing and integrating the comfort-related and control-related conflicts. It is calculated using the arithmetic mean and uniformly represents the overall degree of incoordination in the HVAC system. The comprehensive conflict parameter is obtained by adding the comfort-related conflict parameter and the control-related conflict parameter and taking the average value, which is used to uniformly guide the adaptation correction and convergence tolerance adjustment. For example, if the comfort-related conflict parameter = 0.4 and the control-related conflict parameter = 0.8, the comprehensive conflict parameter = (0.4 + 0.8) / 2 = 0.6.

[0084] Based on this, according to the contradictory parameters, an fitness correction coefficient is configured to correct the first heating comfort and the first ventilation comfort, thus obtaining the first comfort level. The fitness correction coefficient is a weighting coefficient dynamically determined by the comprehensive contradictory parameters. It is used to reduce the comfort weight in high-contradiction scenarios to avoid misjudging incompatible states as high-quality solutions; the greater the contradiction, the smaller the correction coefficient. The first comfort level refers to the comprehensive comfort score after contradiction correction, which serves as the objective function for iterative optimization. The correction coefficient is set to be negatively correlated with the contradictory parameters: fitness correction coefficient = 1 - comprehensive contradictory parameters, with a correction coefficient range of 0~1; first comfort level = (first heating comfort + first ventilation comfort) / 2 × fitness correction coefficient. For example, comprehensive contradictory parameters = 0.6; fitness correction coefficient = 1 - 0.6 = 0.4; original average comfort level = (8.2 + 7.8) / 2 = 8.0 points; corrected first comfort level = 8.0 × 0.4 = 3.2 points.

[0085] Furthermore, a preset convergence tolerance is obtained, and a convergence tolerance adjustment coefficient is configured according to the conflict parameters to adjust the preset convergence tolerance, thus obtaining the convergence tolerance. The preset convergence tolerance is a default comfort fluctuation threshold, for example, a preset convergence tolerance of 0.1. Convergence is achieved when the comfort change is less than 0.1 for multiple consecutive rounds. The convergence tolerance adjustment coefficient is negatively correlated with the comprehensive conflict parameters; the greater the conflict, the smaller the convergence tolerance adjustment coefficient, the smaller the convergence tolerance, and the stricter the optimization judgment. The convergence tolerance is a dynamic threshold adjusted for conflict, used to control the iteration termination condition. Convergence tolerance adjustment coefficient = 1 - comprehensive conflict parameters; final convergence tolerance = preset convergence tolerance × convergence tolerance adjustment coefficient; convergence rule: optimization stops when the comfort change amplitude is less than the final convergence tolerance for 10 consecutive iterations, or when the maximum number of iterations, such as 100, is reached. For example, the preset convergence tolerance is 0.1, the comprehensive contradiction parameter is 0.6, the convergence tolerance adjustment coefficient is 1-0.6=0.4, and the convergence tolerance is 0.1×0.4=0.04.

[0086] Next, the first heating and ventilation control parameters are adjusted to obtain the second heating and ventilation control parameters. The second comfort level and convergence tolerance are then analyzed and calculated. The second heating and ventilation control parameters refer to the second-generation parameters obtained by making small adjustments to the first heating and ventilation control parameters with a fixed step size within the parameter space. These parameters are used for a new round of comfort and conflict analysis. Using the adjustment step size of S200, the first heating and ventilation control parameters are adjusted within the parameter space to generate the second heating and ventilation control parameters. The steps of comfort analysis, conflict calculation, fitness correction, and convergence tolerance update are then repeated to obtain the second comfort level.

[0087] For example, within the parameter space, the first heating air control parameter and the first ventilation control parameter are adjusted: the first heating air control parameter of 23.5℃ is adjusted to the second heating air control parameter of 24.0℃; the first ventilation control parameter is adjusted to 350m. 3 / h is adjusted to the second ventilation control parameter of 370m 3 / h; After reanalysis by S300, the second warm air comfort level is 8.5 and the second ventilation comfort level is 8.3; Repeat the above steps to calculate the new contradictory parameters, correct the comfort level, update the convergence tolerance, and enter the next iteration.

[0088] Finally, optimization is completed when the latest convergence tolerance is met or the convergence iteration number is reached, yielding the optimal heating and ventilation control parameters that maximize comfort. Convergence conditions: Iteration terminates when either of the following conditions is met: the change in corrected comfort over 10 consecutive iterations is less than the final convergence tolerance; or the number of iterations reaches a preset maximum value, such as 100 iterations. The optimal heating and ventilation control parameters refer to the set of parameters that maximizes corrected comfort and minimizes conflicts during the iteration process. The corrected comfort and corresponding control parameters for each iteration are recorded. When the convergence condition is met, iteration stops, and the set of parameters maximizing comfort is selected as the optimal parameters, which are then sent to the HVAC system for load control.

[0089] For example, in the 32nd iteration, the comfort level variation for 10 consecutive iterations was less than the final convergence tolerance of 0.04, satisfying the convergence condition; the parameters with the highest comfort level and lowest conflict were selected for this iteration: optimal heating control parameter = 24.0℃, optimal ventilation control parameter = 370m 3 / h, sent to the HVAC system, to execute precise control of heating air temperature and fresh air volume.

[0090] In this embodiment of the invention, by quantifying the contradictions between comfort and regulation respectively, a comprehensive contradiction parameter is obtained, achieving an accurate characterization of the degree of incoordination in the HVAC system. By dynamically configuring the adaptability correction coefficient and convergence tolerance through the contradiction parameter, the greater the contradiction, the lower the comfort weight and the stricter the convergence judgment, effectively avoiding optimization distortion caused by inflated comfort levels and regulation conflicts. Through multiple rounds of iterative adjustments and dynamic convergence control, the contradiction between heating and ventilation is gradually reduced, while maximizing user comfort. The final output optimal regulation parameter can be adapted to the indoor occupant status, outdoor temperature, and the impact of ventilation and heat dissipation, solving the problems of incoordination between heating and ventilation, significant outdoor interference, and low optimization accuracy in traditional regulation. This achieves refined, intelligent, and collaborative regulation of HVAC load, improving indoor comfort while ensuring stable operation and reasonable energy consumption of the HVAC system.

[0091] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a method and system for intelligent control of HVAC load based on big data analysis. Through image sequence recognition and convolutional neural network models, it accurately acquires the number of users and activity intensity in the target area, providing a basis for control based on actual needs and overcoming the shortcomings of traditional manual control that cannot perceive human dynamics in real time. It clarifies the parameter space for heating and ventilation control and generates first-generation initial control parameters, ensuring that the parameters are adapted to equipment performance and scenario requirements, laying a reliable benchmark for subsequent optimization. It uses HVAC control data tables to achieve accurate mapping between parameters and actual operating values, and quantifies heating and ventilation comfort separately through a specially configured HVAC comfort analysis agent. By analyzing the contradictions in comfort and the control contradictions under the influence of outdoor temperature and ventilation, it integrates and generates comprehensive contradiction parameters, dynamically configures the fitness correction coefficient and convergence tolerance, and obtains the optimal control parameters through multiple rounds of iterative optimization, effectively solving the problems of uncoordinated heating and ventilation control, significant interference from the outdoor environment, and insufficient optimization accuracy. Ultimately, it enhances both indoor heating and ventilation comfort, accurately adapts to real-time occupant status and changes in the outdoor environment, and ensures stable operation and reasonable energy consumption of the HVAC system. It eliminates the lag, subjectivity, and crudeness of traditional manual control, achieving efficient and intelligent management of HVAC load.

[0092] Example 2, as Figure 2 As shown, this invention provides an intelligent HVAC load control system based on big data analysis, the system comprising: User status recognition module 11 is used to collect image sequences within the target area, perform user number recognition and user activity intensity recognition, and obtain user number and user activity intensity; The initial parameter generation module 12 is used to obtain the heating air control parameter space and ventilation control parameter space of the heating and ventilation load control, and generate the first heating air control parameter and the first ventilation control parameter. The comfort analysis module 13 is used to perform heating comfort analysis and ventilation comfort analysis based on the first heating control parameters and the first ventilation control parameters, combined with the number of users and the intensity of user activities, to obtain the first heating comfort and the first ventilation comfort. The iterative optimization control module 14 is used to iteratively optimize the heating control parameters and ventilation control parameters based on the first heating comfort level and the first ventilation comfort level, to obtain the optimal heating control parameters and the optimal ventilation control parameters, and to control the heating, ventilation and air conditioning load. In this process, comfort contradiction analysis and control contradiction analysis are performed, and iterative optimization is carried out.

[0093] In one embodiment, the user state recognition module 11 is further configured to: Acquire image sequences within the target area, which is the area where HVAC load control is performed; User identification is performed on each image within the image sequence to obtain the user coordinate distribution sequence; Based on the user coordinate distribution sequence, the number of users and the intensity of user activity are calculated.

[0094] Specifically, user identification is performed on each image within the image sequence to obtain a user coordinate distribution sequence, including: Based on the image monitoring records of the target area, a set of sample images is collected, and the user coordinates in each sample image are labeled to obtain a set of sample user coordinate distributions. A user identifier is constructed based on a convolutional neural network, which includes convolutional layers, pooling layers, and fully connected layers. The user recognizer is trained and tested using the set of sample images as training input and the set of sample user coordinate distributions as supervision of correct results. The training is completed after convergence. Each image in the image sequence is input into the user identifier, and the output is a user coordinate distribution sequence.

[0095] The number of users and the intensity of user activity are calculated based on the user coordinate distribution sequence, including: Based on the user coordinate distribution sequence, the number of users is counted to obtain the user coordinate number sequence, and the average value is calculated to obtain the number of users. Based on the user coordinate distribution sequence, the two user coordinates with the smallest distance within the adjacent user coordinate distribution are paired, and the coordinate change amplitude is calculated to obtain multiple sets of user coordinate change amplitudes. The average value is then calculated to obtain the user activity intensity.

[0096] In one embodiment, the initial parameter generation module 12 is further configured to: Obtain the space of heating air control parameters and ventilation control parameters for HVAC load regulation; Obtain the initial heating and ventilation control parameters set in the previous time period; Within the space of heating control parameters and ventilation control parameters, the initial heating control parameters and initial ventilation control parameters are adjusted by adjusting the step size to obtain the first heating control parameters and the first ventilation control parameters.

[0097] In one embodiment, the comfort analysis module 13 is further used for: Based on the HVAC control data table, the first temperature mapped by the first heating control parameter and the first fresh air volume mapped by the first ventilation control parameter are obtained. The HVAC control data table is constructed based on the mapping relationship between the sample heating control parameter set and the sample temperature set, and the mapping relationship between the sample ventilation control parameter set and the sample fresh air volume set. The first temperature and the first fresh air volume are combined with the number of users and the intensity of user activity, respectively, and input into the HVAC comfort analysis agent to obtain the first heating comfort and the first ventilation comfort. The HVAC comfort analysis agent includes a heating comfort analysis network and a ventilation comfort analysis network.

[0098] The configuration steps of the HVAC comfort analysis intelligent agent include: Based on the historical HVAC control data of the target area, sample temperature set, sample fresh air volume set, sample number of users set, sample user activity intensity set, and sample user comfort set are collected. A machine learning-based heating comfort analysis network and a ventilation comfort analysis network are constructed. Both the heating comfort analysis network and the ventilation comfort analysis network include an input layer, a hidden layer and an output layer. The hidden layer contains multiple processing nodes. Using the sample temperature set, sample fresh air volume set, and sample user number set as inputs, and the sample user comfort set as labels, the heating comfort analysis network is subjected to supervised training and testing until convergence. Using the sample fresh air volume set, sample user number set, and sample user activity intensity set as inputs, and the sample user comfort set as labels, the ventilation comfort analysis network is subjected to supervised training and testing until convergence. By combining convergent machine learning networks for heating comfort analysis and ventilation comfort analysis, a heating, ventilation, and air conditioning comfort analysis agent is obtained.

[0099] In one embodiment, the iterative optimization control module 14 is further configured to: Based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort contradiction analysis and a control contradiction analysis are performed to obtain the contradiction parameters. Based on the contradictory parameters, an fitness correction coefficient is configured, and the first warm air comfort and the first ventilation comfort are corrected and calculated to obtain the first comfort level. Obtain a preset convergence tolerance, configure a convergence tolerance adjustment coefficient according to the contradictory parameters, and adjust the preset convergence tolerance to obtain the convergence tolerance; Continue to adjust the first heating air control parameters and the first ventilation control parameters to obtain the second heating air control parameters and the second ventilation control parameters, and continue to analyze and calculate the second comfort level and update the convergence tolerance; The optimization process continues until the latest convergence tolerance is met or the convergence iteration count is reached, at which point the optimization is complete, and the optimal heating and ventilation control parameters that maximize comfort are obtained.

[0100] Specifically, based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort conflict analysis and a control conflict analysis are performed to obtain conflicting parameters, including: Calculate the difference between the first heating comfort level and the first ventilation comfort level to obtain the comfort contradiction parameter; Obtain the first heating power and the first ventilation power corresponding to the first heating control parameters and the first ventilation control parameters; The outdoor temperature of the target area is obtained, and the first heating power and the first ventilation power are combined with the input of the HVAC load impact analysis agent to obtain the first predicted heating parameters. The HVAC load impact analysis agent is constructed based on machine learning, and the training data includes a sample outdoor temperature set, a sample ventilation load set, and a sample heating load set as inputs, and a sample predicted heating parameter set as outputs. The deviation between the first predicted heating air parameter and the first heating air control parameter is calculated and used as the control contradiction parameter; Based on the aforementioned comfort-related conflict parameters and control-related conflict parameters, the conflict parameters are calculated.

[0101] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent control of HVAC load based on big data analysis, characterized in that, The method includes: Collect image sequences within the target area, perform user count identification and user activity intensity identification, and obtain the user count and user activity intensity; Obtain the heating air control parameter space and ventilation control parameter space for HVAC load control, and generate the first heating air control parameter and the first ventilation control parameter. Based on the first heating control parameters and the first ventilation control parameters, and combined with the number of users and the intensity of user activities, heating comfort analysis and ventilation comfort analysis are performed to obtain the first heating comfort and the first ventilation comfort. Based on the first heating comfort level and the first ventilation comfort level, the heating control parameters and ventilation control parameters are iteratively optimized to obtain the optimal heating control parameters and optimal ventilation control parameters, and then the HVAC load is controlled. In this process, comfort contradiction analysis and control contradiction analysis are performed, and the process is iterated.

2. The intelligent control method for HVAC load based on big data analysis according to claim 1, characterized in that, Image sequences within the target area are acquired, and user count and activity intensity are identified to obtain user count and activity intensity, including: Acquire image sequences within the target area, which is the area where HVAC load control is performed; User identification is performed on each image within the image sequence to obtain the user coordinate distribution sequence; Based on the user coordinate distribution sequence, the number of users and the intensity of user activity are calculated.

3. The intelligent control method for HVAC load based on big data analysis according to claim 2, characterized in that, User identification is performed on each image within the image sequence to obtain a user coordinate distribution sequence, including: Based on the image monitoring records of the target area, a set of sample images is collected, and the user coordinates in each sample image are labeled to obtain a set of sample user coordinate distributions. A user identifier is constructed based on a convolutional neural network, which includes convolutional layers, pooling layers, and fully connected layers. The user recognizer is trained and tested using the set of sample images as training input and the set of sample user coordinate distributions as supervision of correct results. The training is completed after convergence. Each image in the image sequence is input into the user identifier, and the output is a user coordinate distribution sequence.

4. The intelligent control method for HVAC load based on big data analysis according to claim 2, characterized in that, Based on the user coordinate distribution sequence, the number of users and the intensity of user activity are calculated, including: Based on the user coordinate distribution sequence, the number of users is counted to obtain the user coordinate number sequence, and the average value is calculated to obtain the number of users. Based on the user coordinate distribution sequence, the two user coordinates with the smallest distance within the adjacent user coordinate distribution are paired, and the coordinate change amplitude is calculated to obtain multiple sets of user coordinate change amplitudes. The average value is then calculated to obtain the user activity intensity.

5. The intelligent control method for HVAC load based on big data analysis according to claim 1, characterized in that, Obtain the heating air control parameter space and ventilation control parameter space for HVAC load regulation, and generate the first heating air control parameter and the first ventilation control parameter, including: Obtain the space of heating air control parameters and ventilation control parameters for HVAC load regulation; Obtain the initial heating and ventilation control parameters set in the previous time period; Within the space of heating control parameters and ventilation control parameters, the initial heating control parameters and initial ventilation control parameters are adjusted by adjusting the step size to obtain the first heating control parameters and the first ventilation control parameters.

6. The intelligent control method for HVAC load based on big data analysis according to claim 1, characterized in that, Based on the first heating control parameters and the first ventilation control parameters, and in conjunction with the number of users and the intensity of user activity, heating comfort analysis and ventilation comfort analysis are performed to obtain the first heating comfort level and the first ventilation comfort level, including: Based on the HVAC control data table, the first temperature mapped by the first heating control parameter and the first fresh air volume mapped by the first ventilation control parameter are obtained. The HVAC control data table is constructed based on the mapping relationship between the sample heating control parameter set and the sample temperature set, and the mapping relationship between the sample ventilation control parameter set and the sample fresh air volume set. The first temperature and the first fresh air volume are combined with the number of users and the intensity of user activity, respectively, and input into the HVAC comfort analysis agent to obtain the first heating comfort and the first ventilation comfort. The HVAC comfort analysis agent includes a heating comfort analysis network and a ventilation comfort analysis network.

7. The intelligent control method for HVAC load based on big data analysis according to claim 6, characterized in that, The configuration steps for the HVAC comfort analysis intelligent agent include: Based on the historical HVAC control data of the target area, sample temperature set, sample fresh air volume set, sample number of users set, sample user activity intensity set, and sample user comfort set are collected. A machine learning-based heating comfort analysis network and a ventilation comfort analysis network are constructed. Both the heating comfort analysis network and the ventilation comfort analysis network include an input layer, a hidden layer and an output layer. The hidden layer contains multiple processing nodes. Using the sample temperature set, sample fresh air volume set, and sample user number set as inputs, and the sample user comfort set as labels, the heating comfort analysis network is subjected to supervised training and testing until convergence. Using the sample fresh air volume set, sample user number set, and sample user activity intensity set as inputs, and the sample user comfort set as labels, the ventilation comfort analysis network is subjected to supervised training and testing until convergence. By combining convergent machine learning networks for heating comfort analysis and ventilation comfort analysis, a heating, ventilation, and air conditioning comfort analysis agent is obtained.

8. The intelligent control method for HVAC load based on big data analysis according to claim 1, characterized in that, Based on the first level of heating comfort and the first level of ventilation comfort, iterative optimization of the heating and ventilation control parameters is performed to obtain the optimal heating and ventilation control parameters, including: Based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort contradiction analysis and a control contradiction analysis are performed to obtain the contradiction parameters. Based on the contradictory parameters, an fitness correction coefficient is configured, and the first warm air comfort and the first ventilation comfort are corrected and calculated to obtain the first comfort level. Obtain a preset convergence tolerance, configure a convergence tolerance adjustment coefficient according to the contradictory parameters, and adjust the preset convergence tolerance to obtain the convergence tolerance; Continue to adjust the first heating air control parameters and the first ventilation control parameters to obtain the second heating air control parameters and the second ventilation control parameters, and continue to analyze and calculate the second comfort level and update the convergence tolerance; The optimization process continues until the latest convergence tolerance is met or the convergence iteration count is reached, at which point the optimization is complete, and the optimal heating and ventilation control parameters that maximize comfort are obtained.

9. The intelligent control method for HVAC load based on big data analysis according to claim 8, characterized in that, Based on the first heating comfort level and the first ventilation comfort level, as well as the first heating control parameters and the first ventilation control parameters, a comfort conflict analysis and a control conflict analysis are performed to obtain conflicting parameters, including: Calculate the difference between the first heating comfort level and the first ventilation comfort level to obtain the comfort contradiction parameter; Obtain the first heating power and the first ventilation power corresponding to the first heating control parameters and the first ventilation control parameters; The outdoor temperature of the target area is obtained, and the first heating power and the first ventilation power are combined with the input of the HVAC load impact analysis agent to obtain the first predicted heating parameters. The HVAC load impact analysis agent is constructed based on machine learning, and the training data includes a sample outdoor temperature set, a sample ventilation load set, and a sample heating load set as inputs, and a sample predicted heating parameter set as outputs. The deviation between the first predicted heating air parameter and the first heating air control parameter is calculated and used as the control contradiction parameter; Based on the aforementioned comfort-related conflict parameters and control-related conflict parameters, the conflict parameters are calculated.

10. A smart HVAC load control system based on big data analysis, characterized in that, The system is used to implement the intelligent control method for HVAC load based on big data analysis as described in any one of claims 1-9, the system comprising: The user status recognition module is used to collect image sequences within the target area, perform user count recognition and user activity intensity recognition, and obtain the user count and user activity intensity. The initial parameter generation module is used to obtain the heating air control parameter space and ventilation control parameter space of HVAC load control, and generate the first heating air control parameter and the first ventilation control parameter. The comfort analysis module is used to perform heating comfort analysis and ventilation comfort analysis based on the first heating control parameters and the first ventilation control parameters, combined with the number of users and the intensity of user activities, to obtain the first heating comfort and the first ventilation comfort. The iterative optimization control module is used to iteratively optimize the heating and ventilation control parameters based on the first heating comfort level and the first ventilation comfort level to obtain the optimal heating and ventilation control parameters and to control the HVAC load. In this process, comfort contradiction analysis and control contradiction analysis are performed, and the process is iterated.

Images