Machine vision driven central air conditioner energy-saving control method

Abstract

The invention relates to a central air conditioner energy-saving control system driven by machine vision. The system comprises the steps that a data collection dimension is preset, and multi-dimensional original data of the central air conditioner is obtained; standardized data are obtained through preprocessing, personnel distribution and indoor environment feature data are obtained through target detection and environment feature extraction, and a data analysis result is formed through fusion analysis; constructing an indoor load demand evaluation model based on an analysis result to obtain a real-time dynamic load demand value; in combination with air conditioner operation parameter analysis, the matching deviation between the load demand and air conditioner operation is obtained, and a multi-stage visual driving instruction is generated; the multi-stage visual driving instruction is analyzed, and the air conditioner operation adjusting parameter requirement is obtained; optimization calculation is conducted by combining a preset energy-saving optimization mechanism, energy-saving operation parameters are obtained, control signals are converted, and the air conditioner is driven to operate in an energy-saving mode; energy-saving control over the central air conditioner is achieved through machine vision.

Description

Technical Field

[0001] This invention belongs to the fields of automatic control and artificial intelligence technology, and specifically relates to a machine vision-driven energy-saving control method for central air conditioning. Background Technology

[0002] Currently, the following areas still need improvement in the energy-saving control of central air conditioning systems: Against the backdrop of global energy transition and the advancement of dual-carbon goals, building energy consumption, as a core area of ​​energy consumption, has attracted much attention for its energy-saving potential. Central air conditioning systems, as core energy-consuming equipment in commercial buildings, office parks, and large venues, account for a large proportion of total building energy consumption, making them a key breakthrough point for building energy conservation. However, current central air conditioning energy-saving control technologies still face many significant bottlenecks, making it difficult to meet the actual needs of both precise energy saving and a comfortable experience. First, the perception dimension is limited and the data collection is incomplete. Traditional control systems rely on a limited number of physical sensors such as temperature and humidity, which can only capture single environmental indicators. They lack effective perception of key load-influencing factors such as the dynamic distribution, activity trajectories, and density of indoor personnel. At the same time, they do not fully integrate the central air conditioning system's own operating parameters with detailed environmental characteristics, resulting in incomplete data dimensions and an inability to fully reflect the dynamic changes in actual load demand.

[0003] Secondly, load assessment is static and lacks adaptability. Existing load assessment models are mostly based on fixed building parameters, historical energy consumption data, or static algorithms, making it difficult to respond to dynamic scenarios such as personnel movement, temporary gatherings, and sudden environmental changes. For example, the personnel density in office areas differs significantly between peak hours and lunch breaks. Traditional models cannot adjust the load assessment results in real time, leading to a serious mismatch between air conditioning load supply and actual demand—densely populated areas may experience insufficient cooling / heating, while unoccupied or low-load areas continue to operate at high power, resulting in significant energy waste.

[0004] Third, the control strategies lack differentiation and the ability to coordinate and optimize is weak. Most systems adopt a one-size-fits-all unified control mode, failing to differentiate regulation based on the population density, environmental characteristics, and functional needs of different areas, thus failing to achieve precise energy supply. At the same time, existing technologies often fail to balance energy efficiency and indoor comfort. Either they excessively reduce the operating load in pursuit of energy conservation, causing indoor temperature and humidity to deviate from the comfortable range and affecting the user experience; or they maintain high-load operation to ensure comfort, ignoring energy consumption, thus falling into the dilemma of "energy conservation and comfort cannot be achieved at the same time."

[0005] Fourth, the connection between command generation and execution is not smooth, resulting in low control accuracy. Existing control commands are mostly generated based on single deviation signals, lacking scientific priority division and regional specificity. Furthermore, the parameter optimization process does not fully incorporate equipment safety constraints, leading to delayed control command response and excessively large or small adjustment ranges. In addition, some systems do not consider the differences in communication protocols and regional adaptability of central air conditioning actuators, making it difficult to accurately drive equipment operation after parameter conversion, further reducing control effectiveness and failing to achieve a balance between high efficiency, energy saving, and stable operation.

[0006] These problems not only lead to low energy efficiency in central air conditioning systems and increase user operating costs, but also contradict the current green and low-carbon development concept. Therefore, there is an urgent need for an intelligent control method that integrates multi-dimensional perception, dynamic load assessment, differentiated control and collaborative optimization to break through existing technical bottlenecks and promote the upgrading of central air conditioning energy-saving control towards precision, intelligence and efficiency. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a machine vision-driven energy-saving control method for central air conditioning. The objective of this invention can be achieved through the following technical solutions: S1: Obtain multi-dimensional raw data on the operation of the central air conditioning system by pre-setting data collection dimensions; S2: Preprocess the multi-dimensional raw data to obtain standardized data; perform target detection and environmental feature extraction based on the standardized data to obtain feature data of personnel distribution and indoor environment; perform fusion analysis on the feature data to obtain data analysis results; S3: Based on the data analysis results, construct an indoor load demand assessment model to obtain real-time dynamic load demand values; combine the central air conditioning operating parameters for analysis to obtain the matching deviation between load demand and air conditioning operation, and use the matching deviation to generate multi-level visual drive commands. S4: Parse the multi-level vision driving instructions to obtain the air conditioner operation adjustment parameter requirements; optimize the parameter requirements in combination with the preset energy-saving optimization mechanism to obtain energy-saving operation parameters; convert the energy-saving operation parameters to obtain the control signal of the air conditioner energy-saving operation mechanism.

[0008] As a preferred technical solution of the present invention, the specific process of obtaining multi-dimensional raw data of central air conditioning operation includes: collecting basic regional environmental data to obtain first environmental data; using a machine vision camera to capture indoor scene images, recording the activity area of ​​personnel and environmental details to obtain second visual data; collecting the operating parameters of central air conditioning through the air conditioning controller bus interface to generate third central air conditioning operation data; adding a unified timestamp to the first environmental data, second visual data, and third central air conditioning operation data to obtain multi-dimensional raw data with time-series marking.

[0009] Specifically, the acquisition of standardized data includes: performing outlier removal and normalization on the first environmental data in the multi-dimensional original data to obtain standardized environmental data; performing denoising, grayscale conversion, and size normalization operations on the second visual data to obtain standardized image data; performing unit unification conversion on the operating data of the third central air conditioning system to obtain standardized operating parameters; and integrating the standardized environmental data, standardized image data, and standardized operating parameters to obtain a structured standardized dataset.

[0010] Specifically, the target detection involves: performing frame-by-frame analysis of the standardized image data using target detection rules to identify human targets in the image and obtain the coordinates, number, and movement trajectory data of the human targets; obtaining the activity density of people based on the pixel ratio and distance conversion of the human targets; and integrating the personnel distribution information using the result data from the target detection rules.

[0011] Specifically, the environmental feature extraction involves: extracting detailed environmental features of densely populated areas based on the personnel distribution information; performing grayscale analysis on the wall temperature distribution and the heat dissipation area of ​​the central air conditioning in the standardized image data to obtain indoor thermal environment feature data; and combining the non-visual dimension environmental features in the standardized environmental data to obtain complete indoor environmental feature data.

[0012] Specifically, the process of fusing and analyzing the feature data includes: based on the personnel distribution information, indoor environmental feature data and standardized operating parameters, using a weighted fusion mechanism to assign weights to the corresponding data dimensions, and calculating three indicators: indoor comfort score, personnel load ratio, and central air conditioning operation adaptability; integrating the three indicators to obtain data analysis results.

[0013] Specifically, obtaining the real-time dynamic load demand value includes: constructing a dynamic load demand assessment model based on the three indicators; using the difference between the real-time temperature and humidity in the standardized environmental data and the set comfort threshold as the model correction parameter; substituting the central air conditioning operation adaptability data to calculate the basic load demand value; and dynamically correcting the basic load demand value by combining the prediction of personnel movement trends with personnel movement trajectories to obtain the real-time dynamic load demand value.

[0014] Specifically, the process of analyzing the central air conditioning system based on its operating parameters includes: calculating the actual load supply value of the air conditioning system based on the standardized operating parameters; comparing the real-time dynamic load demand value with the actual load supply value of the air conditioning system to obtain a preliminary matching deviation; analyzing the operating adaptability data of the central air conditioning system to obtain a deviation causal tree; and combining the indoor thermal environment characteristic data to correct the preliminary matching deviation and obtain the load demand and air conditioning operation matching deviation.

[0015] Specifically, the process of generating multi-level visual driving instructions using matching deviation includes: presetting high, medium, and low deviation thresholds; obtaining the instruction adjustment priority based on the level of the air conditioning operation matching deviation; assigning differentiated adjustment weights to corresponding areas based on the personnel distribution information to generate regional differentiated adjustment parameters; and integrating the instruction adjustment priority, regional differentiated adjustment parameters, and load demand to obtain multi-level visual driving instructions.

[0016] Specifically, the process of obtaining the air conditioning operation adjustment parameter requirements includes: parsing the adjustment priority and regionally differentiated adjustment parameters in the multi-level visual drive instructions, and extracting the load correction direction in the multi-level visual drive instructions; calculating the basic adjustment range by combining the real-time dynamic load demand value; and applying a safety threshold constraint to the basic adjustment range based on the central air conditioning operation adaptability data to obtain the air conditioning operation adjustment parameter requirements.

[0017] Specifically, the optimization calculation of parameter requirements includes: using a preset particle swarm optimization algorithm, taking the air conditioning operation adjustment parameter requirements as the optimization objective and the input equipment safe operation parameter range as the constraint; combining the indoor comfort score threshold, constructing an energy-saving-comfort dual-objective optimization function to obtain energy-saving operation parameters for both energy efficiency and comfort.

[0018] Specifically, the process of obtaining the control signal for the air conditioning energy-saving operation mechanism includes: converting the energy-saving operation parameters of energy efficiency and comfort to a format suitable for the communication protocol of the central air conditioning actuator; adjusting the weights based on regional differences, decomposing the parameters into independent control parameters for the actuators in the corresponding regions, and generating the energy-saving operation control signal for the central air conditioning.

[0019] The beneficial effects of this invention are as follows: By relying on machine vision technology to analyze the coordinates, number, movement trajectory, and activity density of indoor human targets frame by frame, and combining this with thermal environment characteristics such as wall temperature distribution and heat dissipation areas, a dynamic load demand assessment model is constructed. This model is then dynamically corrected using temperature and humidity differences, enabling real-time and accurate calculation of load demand. Compared to traditional control methods that rely on fixed thresholds or single sensors, this invention avoids problems such as idling and excessive energy supply, ensuring a high degree of alignment between air conditioning supply and actual load demand, thus reducing ineffective energy consumption at the source.

[0020] By integrating personnel distribution, environmental characteristics, and air conditioning operating parameters through a weighted fusion mechanism, core indicators such as comfort scores and personnel load ratios are generated. Then, a particle swarm optimization algorithm is used to construct a dual-objective optimization function of energy saving and comfort, outputting optimal operating parameters under the constraint of equipment safety operating parameters. This ensures that indoor comfort is not lower than a set threshold while significantly reducing air conditioning energy consumption. Long-term use can substantially reduce building energy expenditures, aligning with the needs of green building development.

[0021] Based on time-series labeling of multi-dimensional raw data and prediction of personnel movement trajectories, the system anticipates load change trends and generates differentiated priority instructions by combining high, medium, and low deviation thresholds, allocating adjustment weights to different densely populated areas. The system completes the entire closed-loop process of data collection, analysis, instruction generation, and execution without manual intervention. It can quickly adapt to dynamic changes such as personnel flow and environmental fluctuations, ensuring that air conditioning operation is always in optimal condition, thus improving the foresight and flexibility of control.

[0022] Operating parameters are collected via the air conditioning controller bus interface, supporting the connection of central air conditioning units of different brands and models. The control signals are compatible with the communication protocols of various actuators, eliminating the need for large-scale modifications to existing air conditioning systems. It can be flexibly adapted to densely populated places such as office buildings, shopping malls, and hospitals, as well as multi-zone temperature control scenarios, reducing system deployment and upgrade costs and demonstrating strong engineering practicality and promotional value. Attached Figure Description

[0023] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a schematic diagram of a machine vision-driven central air conditioning energy-saving control system according to the present invention. Figure 2 This is a structural block diagram for the evaluation and instruction generation in this invention. Detailed Implementation

[0025] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0026] Please see Figure 1-2 A machine vision-driven energy-saving control method for central air conditioning includes: S1: Obtain multi-dimensional raw data on the operation of the central air conditioning system by pre-setting data collection dimensions; S2: Preprocess the multi-dimensional raw data to obtain standardized data; perform target detection and environmental feature extraction based on the standardized data to obtain feature data of personnel distribution and indoor environment; perform fusion analysis on the feature data to obtain data analysis results; S3: Based on the data analysis results, construct an indoor load demand assessment model to obtain real-time dynamic load demand values; combine the central air conditioning operating parameters for analysis to obtain the matching deviation between load demand and air conditioning operation, and use the matching deviation to generate multi-level visual drive commands. S4: Parse the multi-level vision driving instructions to obtain the air conditioner operation adjustment parameter requirements; optimize the parameter requirements in combination with the preset energy-saving optimization mechanism to obtain energy-saving operation parameters; convert the energy-saving operation parameters to obtain the control signal of the air conditioner energy-saving operation mechanism.

[0027] As a preferred technical solution of the present invention, the specific process of obtaining multi-dimensional raw data of central air conditioning operation includes: collecting basic regional environmental data to obtain first environmental data; using a machine vision camera to capture indoor scene images, recording the activity area of ​​personnel and environmental details to obtain second visual data; collecting the operating parameters of central air conditioning through the air conditioning controller bus interface to generate third central air conditioning operation data; adding a unified timestamp to the first environmental data, second visual data, and third central air conditioning operation data to obtain multi-dimensional raw data with time-series marking.

[0028] In this embodiment, environmental sensing devices are evenly deployed in various functional areas of the room to continuously collect basic environmental data within the area, forming first environmental data. This data directly reflects the basic state of the environment in each area. Simultaneously, distributed machine vision cameras are installed at high points indoors. The installation positions are planned according to the spatial layout to ensure full coverage without blind spots, capturing images of the indoor scene in real time and recording environmental details such as personnel activity areas, movement paths, door and window opening and closing status, and changes in lighting and darkness, to obtain second visual data. The air conditioning controller bus interface is accessed through a mainstream communication protocol to collect the core operating parameters of the central air conditioning in real time, generating third central air conditioning operating data. This data directly reflects the current energy supply capacity and operating load of the air conditioning. A time synchronization protocol is used to add a unified timestamp to the three types of data to ensure that the collection time of each type of data is accurately aligned, forming multi-dimensional raw data with time-series markings.

[0029] Specifically, the acquisition of standardized data includes: performing outlier removal and normalization on the first environmental data in the multi-dimensional original data to obtain standardized environmental data; performing denoising, grayscale conversion, and size normalization operations on the second visual data to obtain standardized image data; performing unit unification conversion on the operating data of the third central air conditioning system to obtain standardized operating parameters; and integrating the standardized environmental data, standardized image data, and standardized operating parameters to obtain a structured standardized dataset.

[0030] In this embodiment, an outlier identification algorithm is used to identify and remove outliers in the first environmental data, eliminating invalid data caused by sensor malfunctions or external interference. Then, a normalization method is used to map the processed data to a unified numerical range to obtain standardized environmental data. This data eliminates outlier interference, and all indicators are within the same numerical range, facilitating subsequent multi-dimensional data fusion calculations. A filtering algorithm is used to denoise the second visual data, smoothing noise in the image while preserving edge and detail features. Subsequently, grayscale conversion is performed, transforming the color image into a single-channel grayscale image. Then, the grayscale image is scaled and normalized according to a preset size to ensure that all images have consistent resolution, resulting in standardized image data and reducing the computational complexity of subsequent target detection algorithms. For the third central air conditioning operation data, the non-standard units of equipment from different manufacturers are uniformly converted into industry standard units to obtain standardized operating parameters, eliminating differences in parameter units among equipment manufacturers and making the operating states of different brands of air conditioners comparable. Finally, through data structuring processing, the three types of standardized data are linked and integrated according to timestamps to construct a structured standardized dataset, realizing structured management of multi-source data, which facilitates rapid call and correlation analysis of subsequent algorithms.

[0031] Specifically, the target detection involves: performing frame-by-frame analysis of the standardized image data using target detection rules to identify human targets in the image and obtain the coordinates, number, and movement trajectory data of the human targets; obtaining the activity density of people based on the pixel ratio and distance conversion of the human targets; and integrating the personnel distribution information using the result data from the target detection rules.

[0032] In this embodiment, the YOLOv11 target detection algorithm is used as the core detection rule. To address the issues of missed detection and false detection caused by traditional detection algorithms when people are at different distances and have different scales in indoor scenes (such as people standing at close range, people sitting at a distance, and small targets walking through corridors), the semantic association and detail preservation of features at different scales are strengthened, while the expressive power of the feature map is improved, providing more accurate multi-scale feature support for subsequent target detection and environmental feature extraction.

[0033] Input: Receive three classic scale feature maps output by the YOLOv11 backbone network (downsampled at 1 / 8, 1 / 16, and 1 / 32 of the original image, respectively), covering complete feature information from shallow low semantic high resolution (small object details) to deep high semantic low resolution (large object overall features).

[0034] Core Enhancement Module: Dynamic Receptive Field Adjustment (DRFA) For each scale feature map, an adaptive convolution kernel generation mechanism is used to dynamically match the convolution kernel size according to the target response intensity in the feature map (e.g., a 3×3 small convolution kernel is used to preserve details when focusing on small targets in a 1 / 8 scale feature map; a 7×7 large convolution kernel is used to capture global features when focusing on large targets in a 1 / 32 scale feature map).

[0035] Deformable convolution is introduced to adaptively align features of changes in human posture (such as bending over or sitting) and occlusion (such as being blocked by tables and chairs), avoiding feature distortion caused by traditional fixed convolution kernels.

[0036] Core fusion module: Cross-scale attention-weighted fusion (CSAWF) Calculate the target confidence weight and feature response entropy value of feature maps at each scale, and assign higher weights to features with high confidence and high information content (such as the 1 / 16 scale feature containing a clear human outline has the highest weight).

[0037] By upsampling (transposed convolution), the deep low-resolution feature map is upsized to the same size as the shallow feature map, and then weighted and summed with the shallow feature map to generate three enhanced multi-scale feature maps, achieving dual enhancement of details and semantics.

[0038] Output: Enhanced feature maps at three scales, which retain the detailed texture information of small targets, enhance the semantic features of large targets, and eliminate the semantic gap between features at different scales.

[0039] It addresses the pain point of detecting multi-scale human targets in complex indoor scenarios, reduces the false negative rate of small targets, and improves the accuracy of occluded target recognition.

[0040] Enhance the generalization ability of feature maps to adapt to the scale feature differences of different indoor layouts (such as office areas, meeting rooms, and corridors) without the need for additional training of scene-specific models.

[0041] Personnel-Environment Association Sensing Layer: While accurately detecting human targets, it simultaneously extracts environmental features of the target-related area (such as ambient light around the personnel, door and window status, and equipment heat dissipation), realizing the linkage perception of personnel needs and environmental status. This provides personnel distribution information with environmental tags for subsequent data fusion analysis, improving the pertinence of load assessment and control decisions.

[0042] Structural design and working principle: Input: Enhanced feature map output from the multi-scale feature enhancement fusion layer + preprocessed normalized image data (preserving original pixel brightness, edge and other environmental information).

[0043] Two-branch parallel design: Branch 1: Personnel Detection Branch (using the core structure of the YOLOv11 detection head) Based on the enhanced feature map, the class confidence, coordinate position (bounding rectangle), and quantity statistics of human targets are predicted by the class head and the regression head, respectively. By combining temporal feature association algorithms, the coordinate changes of the same human target in consecutive frames are tracked, and the movement trajectory data of the person is output to quantify the intensity of the person's activities.

[0044] Branch Two: Environment-Related Branch (New Innovation Structure) Based on the human target candidate box output by the personnel detection branch, automatically crop the local image of the corresponding region in the standardized image (focusing on the surrounding environment of the personnel). Lightweight convolutional branches are used to extract local image illumination intensity features (pixel brightness mean and variance quantization), door and window edge features (Canny edge detection + morphological erosion and dilation to identify door and window closure status), and thermal environment correlation features (grayscale value distribution gradient, associated with wall / equipment heat dissipation areas). An attention mechanism is introduced to extract environmental features only from densely populated areas and high-activity areas, reducing unnecessary computational overhead.

[0045] The associated output module binds the target information (coordinates, quantity, trajectory, activity intensity) of the personnel detection branch to the corresponding environmental features (lighting, door and window status, thermal gradient) of the environment associated branch, generating personnel distribution information with environmental tags (e.g., there are 5 people in office area A, with moderate activity intensity, sufficient surrounding lighting, and closed doors and windows).

[0046] Output: Personnel distribution information with environmental tags, which includes the core personnel data of traditional target detection, as well as associated environmental features, directly providing linked input for subsequent environmental feature extraction and data fusion analysis modules.

[0047] This enables a one-to-one perception link between people and their environment, avoiding the blindness of environmental feature extraction and allowing subsequent fusion analysis to focus more on the environmental state corresponding to people's needs, thereby improving the accuracy of data analysis results.

[0048] The system simplifies data flow logic, eliminating the need for a separate environmental feature extraction module to traverse and analyze the entire image, thus reducing system computational complexity and improving real-time response speed.

[0049] Specifically, the environmental feature extraction involves: extracting detailed environmental features of densely populated areas based on the personnel distribution information; performing grayscale analysis on the wall temperature distribution and the heat dissipation area of ​​the central air conditioning in the standardized image data to obtain indoor thermal environment feature data; and combining the non-visual dimension environmental features in the standardized environmental data to obtain complete indoor environmental feature data.

[0050] In this embodiment, based on heat maps and regional statistical reports from personnel distribution information, densely populated areas (such as core locations in meeting rooms and office areas) are identified, and key environmental details are extracted from these areas. This involves calculating light intensity through image pixel brightness analysis (pixel brightness values ​​correspond to a preset light intensity calibration curve, unit: lux), and identifying the closure status of doors and windows using edge detection algorithms (such as the Canny algorithm) (outputting open or closed labels and the pixel width of the opening gap, indirectly reflecting ventilation volume). This focus on core areas of personnel activity ensures targeted feature extraction. Grayscale values ​​are extracted from wall areas and central air conditioning heat dissipation areas in standardized image data, establishing a mapping model between grayscale values ​​and actual temperature (calibrated using an infrared thermometer; higher grayscale values ​​correspond to higher actual temperatures). Based on the grayscale value distribution, the wall temperature gradient is analyzed (calculating the grayscale value differences at different locations on the wall and converting them into actual temperature differences, unit: ℃ / meter) and the temperature diffusion range of the heat dissipation area is analyzed (dividing the diffusion radius corresponding to different grayscale value intervals centered on the heat dissipation vent, unit: meter), resulting in indoor thermal environment characteristic data. This data quantifies the spatial distribution differences of indoor temperature, making up for the shortcomings of traditional point sensors that can only acquire temperature at a single point. It extracts non-visual environmental features such as temperature, humidity, and CO2 concentration from standardized environmental data (normalized, with values ​​ranging from [0,1]) and integrates them with environmental detail features (light intensity, door and window status) and thermal environmental feature data (temperature gradient, diffusion range) extracted from the visual dimension to supplement non-visual environmental information, ultimately forming complete indoor environmental feature data. This data is a set of multi-dimensional feature vectors, with each feature vector corresponding to the comprehensive environmental state of a region, including both visually perceptible detail features and physical indicators from traditional sensing.

[0051] Specifically, the process of fusing and analyzing the feature data includes: based on the personnel distribution information, indoor environmental feature data and standardized operating parameters, using a weighted fusion mechanism to assign weights to the corresponding data dimensions, and calculating three indicators: indoor comfort score, personnel load ratio, and central air conditioning operation adaptability; integrating the three indicators to obtain data analysis results.

[0052] In this embodiment, a hierarchical weighted fusion model is constructed, and weights are assigned according to the degree of influence of each data dimension on control decisions. Among them, personnel distribution information (number of people, density, trajectory) has the highest weight, as it directly determines the core of load demand; indoor environmental characteristic data (temperature gradient, light intensity, CO2 concentration) has the second highest weight, reflecting the impact of the environment on the load; standardized operating parameters (compressor frequency, fan speed) have the second highest weight, reflecting the current operating status of the equipment. Based on the weighted fusion model, the indoor comfort score is calculated in conjunction with the PMV (Predicted Mean Vote) comfort evaluation index. The fused parameters such as temperature and humidity, personnel activity intensity, and clothing thermal resistance are input, and the comfort value is calculated through the PMV formula, then mapped to the scoring range of [0, 100]. The higher the score, the better the comfort. This score quantifies the actual comfort experience of indoor personnel and provides comfort constraints for control decisions. The personnel load ratio is calculated according to the number of people, activity density, and human body heat dissipation standards. First, the total indoor personnel heat dissipation load (number of people × average heat dissipation power per person) is calculated, and then divided by the total indoor load (personnel heat dissipation load + equipment heat dissipation load + ... The heat transfer load of the building envelope is used to obtain the proportion of personnel load (range [0,1]). This indicator reflects the contribution of personnel heat dissipation to the indoor load and clarifies the core source of load demand. By comparing the current operating parameters of the central air conditioning with the design rated parameters and the actual load demand, the operating adaptability of the central air conditioning is calculated. For example, taking the compressor frequency as an example, the deviation rate between the actual frequency and the optimal frequency (the theoretical optimal value based on the current load demand) is calculated. Combined with the deviation rates of parameters such as fan speed and refrigerant flow, the operating adaptability is obtained by weighted summation (range [0,1], 1 indicates complete adaptability, 0 indicates complete mismatch). This indicator quantifies the degree of matching between the equipment operating status and the actual demand. The three indicators of indoor comfort score, personnel load proportion, and central air conditioning operating adaptability are linked and integrated according to time stamp to form a structured data analysis result. This result is a time-series three-dimensional indicator set. Each time node corresponds to a set of three indicator values, which comprehensively reflects the dynamic relationship between personnel demand, environmental status, and equipment operation.

[0053] Specifically, obtaining the real-time dynamic load demand value includes: constructing a dynamic load demand assessment model based on the three indicators; using the difference between the real-time temperature and humidity in the standardized environmental data and the set comfort threshold as the model correction parameter; substituting the central air conditioning operation adaptability data to calculate the basic load demand value; and dynamically correcting the basic load demand value by combining the prediction of personnel movement trends with personnel movement trajectories to obtain the real-time dynamic load demand value.

[0054] In this embodiment, three indicators—indoor comfort score, occupancy load ratio, and central air conditioning operational adaptability—are used as core input variables. A neural network algorithm is employed to construct a dynamic load demand assessment model. The model is trained and optimized using a large amount of historical data to ensure its predictive accuracy. The real-time temperature and humidity data from standardized environmental data are compared with preset human comfort temperature and humidity thresholds, and the difference is calculated as a model correction parameter to compensate for the impact of environmental changes on load demand. The three indicators and correction parameters are substituted into the dynamic load demand assessment model, combined with basic parameters such as the thermal conductivity coefficient of the building envelope and the heat dissipation of indoor equipment, to calculate the current basic load demand value. Based on occupancy trajectory data, a time-series prediction algorithm is used to predict the occupancy flow trend in the near future, determine the direction and magnitude of changes in occupancy density, and dynamically correct the basic load demand value based on the prediction results. Finally, a real-time dynamic load demand value that reflects both real-time demand and future trends is obtained.

[0055] Specifically, the process of analyzing the central air conditioning system based on its operating parameters includes: calculating the actual load supply value of the air conditioning system based on the standardized operating parameters; comparing the real-time dynamic load demand value with the actual load supply value of the air conditioning system to obtain a preliminary matching deviation; analyzing the operating adaptability data of the central air conditioning system to obtain a deviation causal tree; and combining the indoor thermal environment characteristic data to correct the preliminary matching deviation and obtain the load demand and air conditioning operation matching deviation.

[0056] In this embodiment, based on the core operating parameters in the standardized operating parameters and combined with the cooling / heating coefficient of the central air conditioning system, the actual load supply value of the air conditioning system is calculated using thermodynamic formulas, quantifying the actual cooling / heating capacity that the air conditioning system can currently provide. The difference between the real-time dynamic load demand value and the actual load supply value is calculated to obtain the preliminary matching deviation. If the demand value is greater than the supply value, it is a positive deviation; otherwise, it is a negative deviation, intuitively reflecting the supply and demand balance. In-depth analysis of the central air conditioning system's operational adaptability data is performed to identify the core factors causing the deviation, construct a deviation causal tree, and clarify the correlation and influence weight of each factor with the deviation. Combining information such as temperature gradient and heat dissipation area distribution in the indoor thermal environment characteristic data, the degree of influence of environmental factors on the deviation is judged, the preliminary matching deviation is corrected, and the deviation error caused by environmental interference is eliminated to obtain the accurate load demand and air conditioning operation matching deviation.

[0057] Specifically, the process of generating multi-level visual driving instructions using matching deviation includes: presetting high, medium, and low deviation thresholds; obtaining the instruction adjustment priority based on the level of the air conditioning operation matching deviation; assigning differentiated adjustment weights to corresponding areas based on the personnel distribution information to generate regional differentiated adjustment parameters; and integrating the instruction adjustment priority, regional differentiated adjustment parameters, and load demand to obtain multi-level visual driving instructions.

[0058] In this embodiment, based on the operating characteristics and energy-saving requirements of the central air conditioning system, a three-level deviation threshold is preset, with different levels corresponding to different instruction adjustment priorities, ensuring that the system prioritizes deviations that seriously affect energy saving and comfort. Based on the regional personnel density and comfort scores in the personnel distribution information, differentiated adjustment weights are assigned to each indoor area, with higher weights for densely populated areas with lower comfort levels, ensuring that adjustment resources are tilted towards areas with core needs. According to the weight allocation results, differentiated adjustment parameters such as temperature adjustment range and fan speed adjustment ratio are generated for each area, clarifying the specific adjustment targets for each area. The instruction adjustment priority, differentiated adjustment parameters, and real-time dynamic load demand values ​​are integrated, and multi-level visual-driven instructions containing information such as adjustment targets, adjustment range, execution priority, and regional allocation are generated according to the preset instruction format specifications.

[0059] Specifically, the process of obtaining the air conditioning operation adjustment parameter requirements includes: parsing the adjustment priority and regionally differentiated adjustment parameters in the multi-level visual drive instructions, and extracting the load correction direction in the multi-level visual drive instructions; calculating the basic adjustment range by combining the real-time dynamic load demand value; and applying a safety threshold constraint to the basic adjustment range based on the central air conditioning operation adaptability data to obtain the air conditioning operation adjustment parameter requirements.

[0060] In this embodiment, an instruction parsing mechanism is used to decode multi-level vision-driven instructions, extracting the adjustment priority, regionally differentiated adjustment parameters, and load correction direction to clarify the core objectives and directions of the adjustment. Based on the load correction direction and the difference between the real-time dynamic load demand value and the current actual load supply value, the basic adjustment range of each operating module of the air conditioner is calculated. The range of safe operating parameters for the equipment is extracted from the central air conditioner's operational adaptability data as a safety threshold to prevent equipment failure due to overload or underload. The basic adjustment range is verified and corrected using the safety threshold as a constraint, ultimately obtaining the air conditioner operation adjustment parameter requirements that balance demand and safety, satisfying both load correction requirements and ensuring safe equipment operation.

[0061] Specifically, the optimization calculation of parameter requirements includes: using a preset particle swarm optimization algorithm, taking the air conditioning operation adjustment parameter requirements as the optimization objective and the input equipment safe operation parameter range as the constraint; combining the indoor comfort score threshold, constructing an energy-saving-comfort dual-objective optimization function to obtain energy-saving operation parameters for both energy efficiency and comfort.

[0062] In this embodiment, the particle swarm optimization algorithm is selected as the core optimization algorithm. This algorithm has a fast convergence speed and strong global search capability, making it suitable for parameter optimization under multiple constraints. The air conditioning operation adjustment parameter requirements are taken as the optimization objective, clarifying the types of parameters to be optimized and the optimization direction. The safe operating parameter range of the equipment is taken as the constraint condition to ensure that the optimized parameters will not damage the equipment. Combined with the preset indoor comfort score threshold, an energy-saving and comfort dual-objective optimization function is constructed, as shown in the following formula: , , , Where X: the optimization variable vector, X=[f c ,n f ,θ v ,q r (These represent compressor frequency, fan speed, regulating valve opening, and refrigerant flow rate, respectively). ω1, ω2: Weighting coefficients, satisfying ω1+ω2=1 (ω1 is the energy-saving weight, ω2 is the comfort weight, which can be dynamically adjusted according to the scenario, such as ω1=0.6, ω2=0.4 in an office scenario). E(X): Air conditioning energy consumption function, E(X) = +k2n f +k3θ v +k4q r (k1−k4 is the energy consumption coefficient, which is determined by the equipment nameplate parameters and measured data). C(X): Indoor comfort function, derived based on the PMV evaluation model. (a, b are the fitting coefficients, and PMV(X) is the predicted average vote value, calculated from parameters such as temperature, humidity, and wind speed). X min ,X max : Safe operating boundaries for optimization variables (specified by the equipment manufacturer); C th Comfort threshold (usually set to 0.8, corresponding to the PMV range [−0.5, 0.5], to ensure human comfort).

[0063] Equipment safety constraints: X must be within the equipment's permissible operating range to avoid overload or malfunction; Comfort constraint: C(X)≥C th This ensures that energy-saving optimizations do not sacrifice core comfort experiences.

[0064] Minimizing air conditioning energy consumption is the first objective function, and maximizing indoor comfort score is the second objective function. Weighting coefficients are set to balance the importance of the two objectives. Particle swarm optimization algorithm is run to solve the dual-objective optimization function. After iterative calculation, the optimal parameter combination is selected to obtain energy-saving operating parameters that balance energy efficiency and comfort, thus achieving a balance between minimum energy consumption and optimal comfort.

[0065] Specifically, the process of obtaining the control signal for the air conditioning energy-saving operation mechanism includes: converting the energy-saving operation parameters of energy efficiency and comfort to a format suitable for the communication protocol of the central air conditioning actuator; adjusting the weights based on regional differences, decomposing the parameters into independent control parameters for the actuators in the corresponding regions, and generating the energy-saving operation control signal for the central air conditioning.

[0066] In this embodiment, the optimal energy-saving operating parameters are format-converted according to the communication protocol type of the central air conditioning actuators. The parameter data is converted into the data format specified by the protocol, and relevant fields are added as required by the protocol to ensure that the control parameters can be correctly identified and parsed by the actuators. Based on the previously generated regional differential adjustment weights, the optimal energy-saving operating parameters are decomposed by region to obtain independent control parameters for the actuators in each region, ensuring that the adjustment parameters of each region match its own needs. The decomposed independent control parameters are verified to check whether the parameters meet the operating requirements of each actuator and avoid parameter conflicts. Finally, the verified independent control parameters are combined with the protocol format to generate a complete control signal, which directly drives each actuator of the central air conditioning to operate according to the optimized parameters, ultimately achieving the dual goals of energy saving and comfort in air conditioning.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A machine vision-driven energy-saving control method for central air conditioning, characterized in that, include: S1: Obtain multi-dimensional raw data on the operation of the central air conditioning system by pre-setting data collection dimensions; S2: Preprocess the multi-dimensional raw data to obtain standardized data; perform target detection and environmental feature extraction based on the standardized data to obtain feature data of personnel distribution and indoor environment; Perform fusion analysis on feature data to obtain data analysis results; S3: Based on the data analysis results, construct an indoor load demand assessment model to obtain real-time dynamic load demand values; combine the central air conditioning operating parameters for analysis to obtain the matching deviation between load demand and air conditioning operation, and use the matching deviation to generate multi-level visual drive commands. S4: Parse the multi-level vision driving instructions to obtain the air conditioner operation adjustment parameter requirements; optimize the parameter requirements in combination with the preset energy-saving optimization mechanism to obtain energy-saving operation parameters; convert the energy-saving operation parameters to obtain the control signal of the air conditioner energy-saving operation mechanism.

2. The system according to claim 1, characterized in that, The specific process of obtaining multi-dimensional raw data of central air conditioning operation includes: collecting basic regional environmental data to obtain first environmental data; using a machine vision camera to capture indoor scene images, recording the activity area of ​​personnel and environmental details to obtain second visual data; collecting the operating parameters of central air conditioning through the air conditioning controller bus interface to generate third central air conditioning operation data; adding a unified timestamp to the first environmental data, second visual data, and third central air conditioning operation data to obtain multi-dimensional raw data with time-series marking.

3. The system according to claim 1, characterized in that, The acquisition of standardized data includes: performing outlier removal and normalization on the first environmental data in the multi-dimensional original data to obtain standardized environmental data; performing denoising, grayscale conversion and size normalization operations on the second visual data to obtain standardized image data; performing unit unification conversion on the operation data of the third central air conditioner to obtain standardized operation parameters; and integrating the standardized environmental data, standardized image data and standardized operation parameters to obtain a structured standardized dataset.

4. The system according to claim 1, characterized in that, The target detection is as follows: the standardized image data is analyzed frame by frame using target detection rules to identify human targets in the image and obtain the coordinate position, number and movement trajectory data of the human targets; the activity density of people is obtained based on the pixel ratio and distance conversion of the human targets; and the personnel distribution information is integrated using the result data of the target detection rules.

5. The system according to claim 1, characterized in that, The environmental feature extraction is as follows: based on the personnel distribution information, extract detailed environmental features of densely populated areas; Grayscale analysis is performed on the wall temperature distribution and the heat dissipation area of ​​the central air conditioning in the standardized image data to obtain indoor thermal environment characteristic data; combined with the non-visual dimension environmental features in the standardized environmental data, complete indoor environmental characteristic data is obtained.

6. The system according to claim 1, characterized in that, The specific process of fusing and analyzing the feature data includes: based on the personnel distribution information, indoor environmental feature data and standardized operating parameters, a weighted fusion mechanism is used to assign weights to the corresponding data dimensions to calculate three indicators: indoor comfort score, personnel load ratio, and central air conditioning operation adaptability; the three indicators are then integrated to obtain the data analysis results.

7. The system according to claim 1, characterized in that, The process of obtaining the real-time dynamic load demand value includes: constructing a dynamic load demand assessment model based on the three indicators; using the difference between the real-time temperature and humidity in the standardized environmental data and the set comfort threshold as the model correction parameter; substituting the central air conditioning operation adaptability data to calculate the basic load demand value; and dynamically correcting the basic load demand value by combining the prediction of personnel movement trends with personnel movement trajectories to obtain the real-time dynamic load demand value.

8. The system according to claim 1, characterized in that, The specific process of analyzing the central air conditioning system based on its operating parameters includes: calculating the actual load supply value of the air conditioning system based on the standardized operating parameters; comparing the real-time dynamic load demand value with the actual load supply value of the air conditioning system to obtain a preliminary matching deviation; analyzing the operating adaptability data of the central air conditioning system to obtain a deviation causal tree; and combining the indoor thermal environment characteristic data to correct the preliminary matching deviation and obtain the load demand and air conditioning operation matching deviation.

9. The system according to claim 1, characterized in that, The specific process of generating multi-level visual driving instructions using matching deviation includes: preset high, medium and low three-level deviation thresholds; obtaining the instruction adjustment priority according to the level of the air conditioner operation matching deviation; assigning differentiated adjustment weights to corresponding areas based on the personnel distribution information and generating regional differentiated adjustment parameters; and integrating the instruction adjustment priority, regional differentiated adjustment parameters and load demand to obtain multi-level visual driving instructions.

10. The system according to claim 1, characterized in that, The specific process for obtaining the air conditioning operation adjustment parameter requirements includes: parsing the adjustment priority and regionally differentiated adjustment parameters in the multi-level visual drive instructions, and extracting the load correction direction in the multi-level visual drive instructions; calculating the basic adjustment range by combining the real-time dynamic load demand value; and applying a safety threshold constraint to the basic adjustment range based on the central air conditioning operation adaptability data to obtain the air conditioning operation adjustment parameter requirements.

11. The system according to claim 1, characterized in that, The optimization calculation of parameter requirements includes: using a preset particle swarm optimization algorithm, taking the air conditioning operation adjustment parameter requirements as the optimization objective and the input equipment safe operation parameter range as the constraint; combining the indoor comfort score threshold, constructing an energy-saving-comfort dual-objective optimization function to obtain energy-saving operation parameters for both energy efficiency and comfort.

12. The system according to claim 1, characterized in that, The specific process of obtaining the control signal for the air conditioning energy-saving operation mechanism includes: converting the energy-saving operation parameters of energy efficiency and comfort to a format suitable for the communication protocol of the central air conditioning actuator; adjusting the weights based on regional differences, decomposing the parameters into independent control parameters for the actuators in the corresponding regions, and generating the energy-saving operation control signal for the central air conditioning.

Images