Air conditioning energy-saving regulation and control method and system based on graph neural network personnel behavior prediction

Abstract

The application discloses an air conditioner energy-saving regulation and control method and system based on a graph neural network personnel behavior prediction, and belongs to the technical field of air conditioner energy saving. The air conditioner energy-saving regulation and control method based on the graph neural network personnel behavior prediction comprises the following steps: acquiring scene basic data of a target space and personnel behavior data at a current time; wherein the personnel behavior data comprises personnel coordinates, quantity, identity data and behavior; constructing a space-person time-space heterogeneous graph according to the scene basic data and the personnel behavior data, and performing behavior prediction through a trained time-space graph neural network to obtain personnel behavior prediction results of the target space within a preset time period; obtaining a space basic load according to device data of the target space, obtaining personnel dynamic load demand according to the personnel behavior prediction results; determining air conditioner load demand according to environmental data, the space basic load and the personnel dynamic load demand, and determining an air conditioner energy-saving regulation and control strategy in the target space according to the air conditioner load demand.

Description

Technical Field

[0001] This invention belongs to the field of air conditioning energy-saving control technology, specifically relating to an air conditioning energy-saving control method and system based on graph neural network-based prediction of human behavior. Background Technology

[0002] With the acceleration of urbanization and the rapid development of the construction industry, building energy consumption has become an important component of total social energy consumption. Air conditioning systems, as the core equipment for regulating the indoor environment, account for 30% to 50% of total building energy consumption, making them a key entry point for energy conservation and emission reduction efforts. Improving the energy-saving control level of air conditioning systems can not only effectively reduce energy consumption and environmental pollutant emissions, but also lower building operating costs, thus possessing significant economic and social value.

[0003] Currently, air conditioning energy-saving control technologies mainly revolve around temperature threshold control, time-programmed control, and simple human presence detection control. Temperature threshold control sets fixed upper and lower limits for indoor temperature, activating the air conditioner when the ambient temperature exceeds the threshold. However, this method doesn't consider the dynamic changes in people's actual needs, easily leading to situations where the air conditioner runs even when no one is present, or the temperature is uncomfortable for people, making it difficult to balance energy saving and comfort. Time-programmed control plans the start and stop of the air conditioner based on preset work and rest patterns, suitable for scenarios with relatively fixed personnel activities such as offices and schools. However, for spaces with frequent personnel movement and highly unpredictable behavior, such as commercial complexes and transportation hubs, its control flexibility is insufficient, easily leading to energy waste. While simple human presence detection control can sense the presence of people through devices such as infrared and cameras and adjust the air conditioner's operation accordingly, it can only determine the "presence" of people, failing to accurately obtain detailed information such as the number of people, their distribution location, and their activities. This results in significant deviations in air conditioning load calculations and a lack of targeted control strategies.

[0004] Therefore, there is an urgent need for an air conditioning energy-saving control method and system based on graph neural network-based prediction of human behavior to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention provides an air conditioning energy-saving control method and system based on graph neural network for predicting human behavior, which solves the problems of existing air conditioning energy-saving scheduling methods not adapting to the dynamic needs of human behavior, lacking flexibility, and being insufficiently targeted.

[0006] To achieve the above objectives, this invention provides an air conditioning energy-saving control method based on graph neural network-based prediction of human behavior, comprising the following steps: Acquire basic scene data and current time personnel behavior data for the target space; among which, personnel behavior data includes: personnel coordinates, quantity, identity data, and behavior; Based on the basic scene data and personnel behavior data, a spatial-human spatiotemporal heterogeneous graph is constructed, and behavior prediction is performed through a trained spatiotemporal graph neural network to obtain the prediction results of personnel behavior in the target space within a preset time period. The basic load of the space is obtained based on the equipment data of the target space, the dynamic load demand of personnel is obtained based on the prediction results of personnel behavior, and the air conditioning load demand curve is determined based on environmental data, basic load of space and dynamic load demand of personnel. Determine the air conditioning energy-saving control strategy within the target space based on the air conditioning load demand curve.

[0007] As an embodiment of the present invention, a spatial-human spatiotemporal heterogeneous graph is constructed based on scene basic data and personnel behavior data, and behavior prediction is performed through a trained spatiotemporal graph neural network to obtain the personnel behavior prediction results of the target space within a preset time period, including: Preprocessing is performed based on the basic scene data and the current time's personnel behavior data to obtain a spatial-personnel feature dataset; Extracting data from the space-person behavior dataset yielded several triplet data sets. A spatial-human spatiotemporal heterogeneous graph is constructed based on several triplet data, and the spatial-human spatiotemporal heterogeneous graph is inferred through a trained spatiotemporal graph neural network to obtain the prediction results of human behavior.

[0008] As an embodiment of the present invention, the basic load of the space is obtained based on the acquired equipment data of the target space, the dynamic load demand of personnel is obtained based on the prediction results of personnel behavior, and the air conditioning load demand curve is determined based on environmental data, the basic load of the space, and the dynamic load demand of personnel, including: Acquire equipment data, environmental data, and current-time personnel behavior data within the target space; among which, equipment data includes: equipment quantity, equipment model, and equipment operation plan; The heat generation curve for a preset time period is generated based on the equipment data and used as the basic space load. Based on the predicted personnel behavior results and the current personnel behavior data, determine the personnel behavior data within the preset time period; Input the personnel behavior data within a preset time period into the pre-trained load demand model to obtain the dynamic load demand of personnel.

[0009] In one embodiment, the air conditioning load demand is determined based on environmental data, spatial base load, and dynamic load demand of personnel, and an air conditioning energy-saving control strategy is determined within the target space based on the air conditioning load demand, including: The preset time period is divided into multiple time periods to obtain a sequence of control time periods, and the control time period at the first position in the sequence is selected as the target time period. Obtain current environmental data and air conditioning operation data, and determine the air conditioning load demand for the target time period in the control time period sequence based on environmental data, basic space load, and dynamic load demand of personnel; Based on a preset dynamic scheduling algorithm, an air conditioning energy-saving control strategy for the target time period is generated according to air conditioning load demand and air conditioning operation data.

[0010] As one embodiment of the present invention, it also includes: Acquire the full lifecycle operation data of the air conditioning equipment within the target space; the full lifecycle operation data includes: runtime, historical start-up and shutdown counts, failure frequency, current heat exchanger cleanliness, and equipment rated energy efficiency ratio; The aging and degradation coefficient of the air conditioning equipment is calculated based on the full life cycle operation data of the air conditioning equipment, as shown below: In the formula, Indicates the aging and degradation coefficient of the equipment. , , and Both indicate that the attenuation affects the weighting coefficient. This indicates the design lifespan of the air conditioning equipment. Indicates runtime. This indicates the design rated number of start-stop cycles for the air conditioning equipment. Indicates the number of historical starts and stops. This indicates the current cleanliness of the heat exchanger. This indicates the frequency of failures within the past three months. The actual energy efficiency ratio and load correction factor of the air conditioning equipment are calculated based on the equipment aging and degradation coefficient, as shown below: In the formula, This indicates the current indoor and outdoor temperature difference. This indicates the actual energy efficiency ratio of the air conditioning equipment. Indicates the rated energy efficiency ratio of the equipment. Indicates the temperature difference adaptation coefficient. Indicates the load correction factor; The air conditioning load demand is optimized based on the load correction factor, resulting in the corrected air conditioning load demand, as shown below: In the formula, This indicates the revised air conditioning load demand. Indicates air conditioning load demand. Indicates the need for aging load replenishment. Safety redundancy factor, This indicates the actual maximum output load of the air conditioning equipment. This indicates the rated maximum output load of the air conditioning equipment.

[0011] In one embodiment, it also includes: Acquire data on personnel behavior prediction results, actual personnel behavior, and the process of implementing air conditioning energy-saving control strategies; The comprehensive evaluation coefficient of the control effect is calculated based on the predicted personnel behavior, actual personnel behavior, and process data, as shown below: In the formula, This represents the comprehensive evaluation coefficient of the regulatory effect. Indicates the assessment of bias in personnel behavior prediction. Indicates an assessment of temperature regulation efficiency. Indicates the achievement of energy conservation targets. Indicates that the comfort level meets the assessment criteria. , , and All represent preset weights. Indicates the first [number] within the preset time period Actual personnel behavior data values ​​of Wei Indicates the first [number] within the preset time period Dimension's predictive human behavior data values, This indicates the predicted completion time of the air conditioning energy-saving control strategy. This indicates the time when the actual air conditioning energy-saving control strategy was completed. This represents the predicted energy consumption of the air conditioning energy-saving control strategy. This represents the energy consumption of the actual air conditioning energy-saving control strategy. This indicates the comfort level feedback score for the personnel. This represents the total dimension of personnel behavior data; Based on the comprehensive evaluation coefficients of the regulation effect, a preset graph neural network iterative correction signal is generated, as shown below: In the formula, This represents the weight adjustment amount in the graph neural network model. Represents the iterative learning rate. This represents the gradient of the prediction loss function; when When the evaluation result of the preset graph neural network is deemed excellent, no model parameters are adjusted; when At that time, the evaluation result of the preset graph neural network was determined to be good, according to Adjust model parameters; when At that time, the evaluation result of the preset graph neural network is determined to be in need of optimization, according to Adjust the model parameters.

[0012] In addition, the present invention also provides an air conditioning energy-saving control system based on graph neural network for predicting human behavior, comprising: The data acquisition module is used to acquire basic scene data of the target space and personnel behavior data at the current time; among which, personnel behavior data includes: personnel coordinates, quantity, identity data and behavior; The behavior prediction module is used to construct a spatial-human spatiotemporal heterogeneous graph based on scene basic data and personnel behavior data, and to perform behavior prediction through a trained spatiotemporal graph neural network to obtain the personnel behavior prediction results in the target space within a preset time period. The demand determination module is used to obtain the basic load of the space based on the acquired equipment data of the target space, and to obtain the dynamic load demand of personnel based on the prediction results of personnel behavior. The strategy determination module is used to determine the air conditioning load demand based on environmental data, spatial base load, and dynamic load demand of personnel, and to determine the air conditioning energy-saving control strategy within the target space based on the air conditioning load demand.

[0013] The beneficial effects of this invention are as follows: by integrating environmental data, spatial basic load and dynamic load demand of personnel to formulate specific control strategies for air conditioning, the air conditioning load calculation is more in line with actual needs, effectively avoiding energy waste problems such as "running when no one is present" and "over-supply due to load misjudgment", and solving the problems of insufficient flexibility and poor scenario adaptability of traditional control strategies.

[0014] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0015] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a flowchart illustrating the air conditioning energy-saving control method based on graph neural network for predicting human behavior, as described in this invention. Figure 2 This is a schematic diagram of the modules of the air conditioning energy-saving control system based on graph neural network for predicting human behavior according to the present invention. Detailed Implementation

[0016] like Figure 1 As shown, this invention provides an air conditioning energy-saving control method based on graph neural network-based prediction of human behavior, comprising the following steps: Acquire basic scene data and current time personnel behavior data for the target space; among which, personnel behavior data includes: personnel coordinates, quantity, identity data, and behavior; Based on the basic scene data and personnel behavior data, a spatial-human spatiotemporal heterogeneous graph is constructed, and behavior prediction is performed through a trained spatiotemporal graph neural network to obtain the prediction results of personnel behavior in the target space within a preset time period. The basic load of the space is obtained based on the equipment data of the target space, the dynamic load demand of personnel is obtained based on the prediction results of personnel behavior, and the air conditioning load demand curve is determined based on environmental data, basic load of space and dynamic load demand of personnel. Determine the air conditioning energy-saving control strategy within the target space based on the air conditioning load demand curve.

[0017] The working principle of the above technical solution is as follows: In actual use, multiple sensors are used to achieve accurate and compliant acquisition of personnel behavior data. Specifically, UWB positioning base stations (to obtain the current coordinates of personnel) and high-definition smart cameras (equipped with human posture recognition algorithms and privacy desensitization modules to collect the number of people and their behavior status such as staying, moving, and gathering, while blurring sensitive information such as faces) are deployed in key areas of the target space. Combined with image processing methods, personnel identity data (such as age group, gender, etc.) is obtained, and basic scene data (including space area, device location, etc.) is collected simultaneously. Based on the above basic scene data and personnel behavior data, a spatial node and personnel node structure is constructed, with spatiotemporal relationships between nodes (such as the movement trajectory of personnel in different spaces, and the number of people in the same space). The spatial-human spatiotemporal heterogeneous graph is constructed with the aggregation duration as the edge. This heterogeneous graph is input into a spatiotemporal graph neural network that has been pre-trained with a large amount of historical spatiotemporal data. The network's spatiotemporal attention mechanism is used to mine the dynamic correlation between human behavior and the spatial environment, thereby outputting prediction results such as the change in the number of people, coordinate distribution, and main behavior types in the target space within a preset time period (e.g., 3 hours). Subsequently, equipment data such as air conditioning equipment parameters and power of other electrical equipment in the target space are collected to obtain the basic load of the space. Combined with the human behavior prediction results, the dynamic load demand of people is obtained, as well as environmental data such as indoor temperature and humidity, to determine the air conditioning load demand in the future preset time period. Finally, based on the load demand, an air conditioning energy-saving control strategy is formulated, which includes the timing of air conditioning start-up and shutdown, and the adjustment of operating power.

[0018] The beneficial effects of the above technical solution are as follows: By integrating environmental data, spatial basic load and dynamic load demand of personnel, the specific control strategy of air conditioning is formulated, making the air conditioning load calculation more in line with actual needs. This effectively avoids energy waste problems such as "running when no one is present" and "over-supply due to load misjudgment", and solves the problems of insufficient flexibility and poor scenario adaptability of traditional control strategies.

[0019] In one embodiment, a spatial-human spatiotemporal heterogeneous graph is constructed based on scene baseline data and human behavior data, and behavior prediction is performed using a trained spatiotemporal graph neural network to obtain the predicted human behavior results in the target space within a preset time period, including: Preprocessing is performed based on the basic scene data and the current time's personnel behavior data to obtain a spatial-personnel feature dataset; Extracting data from the space-person behavior dataset yielded several triplet data sets. A spatial-human spatiotemporal heterogeneous graph is constructed based on several triplet data, and the spatial-human spatiotemporal heterogeneous graph is inferred through a trained spatiotemporal graph neural network to obtain the prediction results of human behavior.

[0020] The working principle and beneficial effects of the above technical solution are as follows: Preprocessing (including outlier removal, data format standardization, and spatiotemporal dimension alignment) of the target space's basic scene data and current-time personnel behavior data yields a structured space-person feature dataset. Based on this dataset, several triples describing space, personnel, and spatiotemporal relationships are extracted. These triples adopt a structure of <space node, spatiotemporal relationship, personnel node>. A space-person spatiotemporal heterogeneous graph is constructed based on these triples, with functional partitions and sensor deployment points of the target space designated as space nodes, personnel as personnel nodes, and spatiotemporal relationships in the triples as edges connecting the nodes, assigning attributes such as time intervals and behavior types to the edges. The spatiotemporal graph neural network requires a complete training process: collecting historical scene data, personnel behavior data, and corresponding environmental data of the target space, dividing them into training and validation sets. The model is prepared with a validation set and a test set, and the data is standardized in accordance with the preprocessing stage. Then, a spatiotemporal attention-based graph neural network architecture is used, which includes a node embedding layer (converting the discrete features of spatial and personnel nodes into high-dimensional vectors), a spatiotemporal convolutional layer (capturing the spatial correlation and temporal evolution of nodes), an attention layer (dynamically allocating weights to different nodes and edges, focusing on key spatiotemporal correlation information), and a fully connected output layer (outputting the personnel behavior prediction results). The historical spatiotemporal heterogeneous graph in the training set is input into the model, and the error between the predicted behavior and the actual behavior is used as the loss function (using the mean squared error loss function). The Adam optimizer is used to iteratively update the model parameters, setting the number of iterations, learning rate, and other hyperparameters until the model's prediction accuracy on the validation set reaches a preset threshold and tends to converge, finally obtaining a well-trained spatiotemporal graph neural network that meets the accuracy requirements. The constructed spatial-person spatiotemporal heterogeneous graph of the current time is input into the well-trained graph neural network for inference to obtain prediction results such as the change in the number of people, spatial distribution, and behavior types in the target space within a preset time period.

[0021] In one embodiment, the space base load is obtained based on the acquired equipment data of the target space, the dynamic load demand of personnel is obtained based on the personnel behavior prediction results, and the air conditioning load demand curve is determined based on environmental data, the space base load, and the dynamic load demand of personnel, including: Acquire equipment data, environmental data, and current-time personnel behavior data within the target space; among which, equipment data includes: equipment quantity, equipment model, and equipment operation plan; The heat generation curve for a preset time period is generated based on the equipment data and used as the basic space load. The predicted personnel behavior results are input into the pre-trained load demand model to obtain the dynamic load demand of personnel.

[0022] The working principle and beneficial effects of the above technical solution are as follows: Equipment data includes: number of equipment, equipment model (used to determine the heat dissipation coefficient based on operating power to dissipate heat to the outside during operation), and operating plan (e.g., intermittent operation from 8:30 to 18:00 on weekdays); then, the start and end times of operation and operating power of each equipment within a preset time period are specified through the operating plan; then, the equipment model is used to calculate the heat dissipation of the equipment to the target space within the preset time period; then, the heat generation curve within the preset time period is generated by summarizing according to the time scale, and this curve is directly used as the basic load of the space; In addition, when determining the dynamic load demand of personnel, the behavior data of personnel at each time point in the preset time period is determined based on the prediction results of personnel behavior. The behavior data of personnel at each time point is input into the pre-trained load demand model. The model maps the corresponding relationship of personnel number, behavior type, age, gender, etc., and outputs the dynamic load demand of personnel (e.g., if a person is running in the target space, the most comfortable environmental temperature, humidity, etc. are determined according to their age group, gender, etc.).

[0023] In one embodiment, the air conditioning load demand is determined based on environmental data, spatial base load, and dynamic load demand of personnel, and an air conditioning energy-saving control strategy is determined within the target space based on the air conditioning load demand, including: The preset time period is divided into multiple time periods to obtain a sequence of control time periods, and the control time period at the first position in the sequence is selected as the target time period. Obtain current environmental data and air conditioning operation data, and determine the air conditioning load demand for the target time period in the control time period sequence based on environmental data, basic space load, and dynamic load demand of personnel; Based on a preset dynamic scheduling algorithm, an air conditioning energy-saving control strategy for the target time period is generated according to air conditioning load demand and air conditioning operation data.

[0024] The working principle and beneficial effects of the above technical solution are as follows: After acquiring environmental data (indoor temperature and humidity), spatial baseline load (generated by summarizing equipment heat dissipation), and dynamic load requirements of personnel, the preset time period (e.g., 1 hour) is first divided into multiple continuous control time periods according to a preset time granularity (e.g., 15 minutes / segment), forming a control time period sequence; the first time period of the sequence is selected as the target time period to ensure the timeliness of control; then, the current environmental data (e.g., current indoor temperature and humidity) and air conditioning operation data (e.g., current operating power and air outlet temperature) are collected, combined with the acquired spatial baseline load and personnel dynamic load requirements. The dynamic load demand of personnel determines the air conditioning load demand for the target time period (i.e., the air conditioning energy supply demand to meet personnel comfort and match equipment heat dissipation). Specifically, the process involves comparing the current indoor temperature and humidity with the comfort temperature and humidity thresholds corresponding to the dynamic load demand of personnel, calculating the temperature and humidity deviation, and simultaneously calculating the matching degree between the space's basic load and the current indoor thermal environment. With achieving the comfort temperature and humidity standard as the core objective, the space's basic load and the dynamic load demand of personnel are coupled, and corrected using the temperature and humidity deviation values ​​to obtain the initial load range that meets comfort requirements. Finally, the air conditioning load demand is determined based on the volume of the target space, such as: The algorithm considers factors such as the required cooling or heating capacity and humidity control for the target space from the first time point to the second time point. A dynamic scheduling algorithm, specifically Model Predictive Control (MPC), is preferred. This algorithm accurately processes time-series load data and adapts to rolling optimization logic. Its core objective function, designed with energy saving as its core, is: min∑(air conditioner operating power × operating time). Constraints include: indoor temperature and humidity must be within the comfort range corresponding to the dynamic load demand of the occupants (e.g., 24-26℃ in summer, humidity 40%-60%), air conditioner operating power must not exceed the equipment's rated power limit, and air conditioner start-stop frequency must not exceed the equipment's safety threshold. Based on this objective function, the MPC algorithm uses historical load data, current air conditioning operating status, and load demand for the target time period to predict air conditioning energy consumption and comfort compliance under different control parameters, generating the optimal combination of control parameters (such as operating power and air outlet mode) to form an air conditioning energy-saving control strategy for the target time period. Subsequently, through a rolling optimization mechanism, after completing the control of the current target time period, the completed time period is discarded, and the subsequent time periods in the sequence are progressively used as new target time periods. The above process of "obtaining real-time data of the current time - determining load demand - generating control strategy" is repeated to achieve dynamic and continuous control of the entire preset time period.

[0025] In one embodiment, it also includes: Acquire the full lifecycle operation data of the air conditioning equipment within the target space; the full lifecycle operation data includes: runtime, historical start-up and shutdown counts, failure frequency, current heat exchanger cleanliness, and equipment rated energy efficiency ratio; The aging and degradation coefficient of the air conditioning equipment is calculated based on the full life cycle operation data of the air conditioning equipment, as shown below: In the formula, Indicates the aging and degradation coefficient of the equipment. , , and Both indicate that the attenuation affects the weighting coefficient. This indicates the design lifespan of the air conditioning equipment. Indicates runtime. This indicates the design rated number of start-stop cycles for the air conditioning equipment. Indicates the number of historical starts and stops. This indicates the current cleanliness of the heat exchanger. This indicates the frequency of failures within the past three months. The actual energy efficiency ratio and load correction factor of the air conditioning equipment are calculated based on the equipment aging and degradation coefficient, as shown below: In the formula, This indicates the current indoor and outdoor temperature difference. This indicates the actual energy efficiency ratio of the air conditioning equipment. Indicates the rated energy efficiency ratio of the equipment. Indicates the temperature difference adaptation coefficient. Indicates the load correction factor; The air conditioning load demand is optimized based on the load correction factor, resulting in the corrected air conditioning load demand, as shown below: In the formula, This indicates the revised air conditioning load demand. Indicates air conditioning load demand. Indicates the need for aging load replenishment. Safety redundancy factor, This indicates the actual maximum output load of the air conditioning equipment. This indicates the rated maximum output load of the air conditioning equipment.

[0026] The working principle of the above technical solution is as follows: After determining the air conditioning load demand, a preset dynamic scheduling algorithm is used to determine the air conditioning energy-saving scheduling strategy based on the load demand. However, the energy-saving scheduling strategy obtained at this time is based on the air conditioning equipment operating at its rated energy efficiency. As the air conditioning equipment ages, its output capacity will not reach the rated state, causing the air conditioning equipment to operate according to the energy-saving scheduling strategy and fail to meet the required demand, resulting in failure to meet the comfort requirements of personnel. Therefore, this technical solution, combined with the aging state of the air conditioning equipment, corrects the air conditioning load demand. By acquiring the operating data of the air conditioning throughout its entire life cycle, the aging state of the equipment is quantified from four dimensions: "duration loss, mechanical loss, cleanliness loss, and failure loss". Then, by calculating the aging attenuation coefficient, the multi-dimensional aging factors are integrated into a single quantitative index, thereby accurately reflecting the degree of attenuation of the equipment's actual operating capacity relative to its rated state. Finally, the actual energy efficiency ratio of the equipment is corrected based on the aging coefficient to avoid the "discrepancy between theoretical energy efficiency and actual energy efficiency" caused by equipment aging. The load calculation deviation is then addressed; the air conditioning load demand is dynamically corrected using a load correction coefficient and aging compensation load to ensure that the load demand does not exceed the actual output capacity of the equipment after aging, while maintaining reasonable safety redundancy; finally, based on the corrected air conditioning load demand, the core control parameters such as the compressor are determined through a dynamic scheduling algorithm, so that the control strategy can meet the load demand and adapt to the operating characteristics of the equipment after aging, avoiding equipment overload or energy waste.

[0027] The beneficial effects of the above technical solution are as follows: After determining the air conditioning load demand, the air conditioning load demand is corrected based on the aging status of the air conditioning equipment. Then, a preset dynamic scheduling algorithm is used to determine the air conditioning energy-saving scheduling strategy based on the corrected air conditioning load demand. By correcting the aging of the air conditioning load demand, the output of the air conditioning equipment can meet the air conditioning load demand while maintaining the minimum energy consumption.

[0028] In one embodiment, it also includes: Acquire data on personnel behavior prediction results, actual personnel behavior, and the process of implementing air conditioning energy-saving control strategies; The comprehensive evaluation coefficient of the control effect is calculated based on the predicted personnel behavior, actual personnel behavior, and process data, as shown below: In the formula, This represents the comprehensive evaluation coefficient of the regulatory effect. Indicates the assessment of bias in personnel behavior prediction. Indicates an assessment of temperature regulation efficiency. Indicates the achievement of energy conservation targets. Indicates that the comfort level meets the assessment criteria. , , and All represent preset weights. Indicates the first [number] within the preset time period Actual personnel behavior data values ​​of Wei Indicates the first [number] within the preset time period Dimension's predictive human behavior data values, This indicates the predicted completion time of the air conditioning energy-saving control strategy. This indicates the time when the actual air conditioning energy-saving control strategy was completed. This represents the predicted energy consumption of the air conditioning energy-saving control strategy. This represents the energy consumption of the actual air conditioning energy-saving control strategy. This indicates the comfort level feedback score for the personnel. This represents the total dimension of personnel behavior data; Based on the comprehensive evaluation coefficients of the regulation effect, a preset graph neural network iterative correction signal is generated, as shown below: In the formula, This represents the weight adjustment amount in the graph neural network model. Represents the iterative learning rate. This represents the gradient of the prediction loss function; when When the evaluation result of the preset graph neural network is deemed excellent, no model parameters are adjusted; when At that time, the evaluation result of the preset graph neural network was determined to be good, according to Adjust model parameters; when At that time, the evaluation result of the preset graph neural network is determined to be in need of optimization, according to Adjust the model parameters.

[0029] The working principle and beneficial effects of the above technical solution are as follows: By collecting multi-dimensional feedback data (human behavior prediction deviation, temperature regulation efficiency, energy consumption compliance rate, and comfort) during the strategy formulation and operation process, a comprehensive evaluation coefficient of the control effect is calculated to fully quantify the actual execution effect of the air conditioning energy-saving scheduling strategy; then, based on the evaluation coefficient and the prediction loss gradient, a model weight correction signal is generated—the better the effect, the smaller the correction magnitude; the worse the effect, the larger the correction magnitude, and the correction only applies to "subsequent prediction tasks"; finally, a visual evaluation report and iteration suggestions are output to form a data closed loop, allowing the graph neural network prediction model to be continuously optimized with actual operating data, and the accuracy of the subsequently generated strategies gradually improves.

[0030] like Figure 2 As shown, the present invention also provides an air conditioning energy-saving control system based on graph neural network for predicting human behavior, comprising: The data acquisition module is used to acquire basic scene data of the target space and personnel behavior data at the current time; among which, personnel behavior data includes: personnel coordinates, quantity, identity data and behavior; The behavior prediction module is used to construct a spatial-human spatiotemporal heterogeneous graph based on scene basic data and personnel behavior data, and to perform behavior prediction through a trained spatiotemporal graph neural network to obtain the personnel behavior prediction results in the target space within a preset time period. The demand determination module is used to obtain the basic load of the space based on the acquired equipment data of the target space, and to obtain the dynamic load demand of personnel based on the prediction results of personnel behavior. The strategy determination module is used to determine the air conditioning load demand based on environmental data, spatial base load, and dynamic load demand of personnel, and to determine the air conditioning energy-saving control strategy within the target space based on the air conditioning load demand.

[0031] This technical solution serves as the system technical solution for the aforementioned air conditioning energy-saving control method based on graph neural network human behavior prediction. Its working principle and beneficial effects are the same as those of the method technical solution, and will not be elaborated further here.

[0032] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. An air conditioning energy-saving control method based on graph neural network for predicting human behavior, characterized in that, Includes the following steps: Acquire basic scene data and current time personnel behavior data for the target space; among which, personnel behavior data includes: personnel coordinates, quantity, identity data, and behavior; Based on the basic scene data and personnel behavior data, a spatial-human spatiotemporal heterogeneous graph is constructed, and behavior prediction is performed through a trained spatiotemporal graph neural network to obtain the prediction results of personnel behavior in the target space within a preset time period. The basic load of the space is obtained based on the equipment data of the target space, and the dynamic load demand of personnel is obtained based on the prediction results of personnel behavior. The air conditioning load demand is determined based on environmental data, spatial basic load, and dynamic load demand of personnel, and an air conditioning energy-saving control strategy is determined within the target space based on the air conditioning load demand.

2. The air conditioning energy-saving control method based on graph neural network-based human behavior prediction according to claim 1, characterized in that, Based on the scene's basic data and personnel behavior data, a spatial-human spatiotemporal heterogeneous graph is constructed. Then, a trained spatiotemporal graph neural network is used for behavior prediction, yielding the predicted personnel behavior results for the target space within a preset time period, including: Preprocessing is performed based on the basic scene data and the current time's personnel behavior data to obtain a spatial-personnel feature dataset; Extracting data from the space-person behavior dataset yielded several triplet data sets. A spatial-human spatiotemporal heterogeneous graph is constructed based on several triplet data, and the spatial-human spatiotemporal heterogeneous graph is inferred through a trained spatiotemporal graph neural network to obtain the prediction results of human behavior.

3. The air conditioning energy-saving control method based on graph neural network-based human behavior prediction according to claim 1, characterized in that, Based on the acquired equipment data of the target space, the basic load of the space is obtained; based on the predicted personnel behavior, the dynamic load demand of personnel is obtained; and based on environmental data, the basic load of the space, and the dynamic load demand of personnel, the air conditioning load demand curve is determined, including: Acquire equipment data, environmental data, and current-time personnel behavior data within the target space; among which, equipment data includes: equipment quantity, equipment model, and equipment operation plan; The heat generation curve for a preset time period is generated based on the equipment data and used as the basic space load. The predicted personnel behavior results are input into the pre-trained load demand model to obtain the dynamic load demand of personnel.

4. The air conditioning energy-saving control method based on graph neural network human behavior prediction according to claim 1, characterized in that, Based on environmental data, spatial baseline load, and dynamic occupant load demand, the air conditioning load demand is determined, and based on this demand, an energy-saving control strategy for air conditioning within the target space is determined, including: The preset time period is divided into multiple time periods to obtain a sequence of control time periods, and the control time period at the first position in the sequence is selected as the target time period. Obtain current environmental data and air conditioning operation data, and determine the air conditioning load demand for the target time period in the control time period sequence based on environmental data, basic space load, and dynamic load demand of personnel; Based on a preset dynamic scheduling algorithm, an air conditioning energy-saving control strategy for the target time period is generated according to air conditioning load demand and air conditioning operation data.

5. The air conditioning energy-saving control method based on graph neural network human behavior prediction according to claim 1, characterized in that, Also includes: Acquire the full lifecycle operation data of the air conditioning equipment within the target space; the full lifecycle operation data includes: runtime, historical start-up and shutdown counts, failure frequency, current heat exchanger cleanliness, and equipment rated energy efficiency ratio; The aging and degradation coefficient of the air conditioning equipment is calculated based on the full life cycle operation data of the air conditioning equipment, as shown below: In the formula, Indicates the aging and degradation coefficient of the equipment. , , and Both indicate that the attenuation affects the weighting coefficient. This indicates the design lifespan of the air conditioning equipment. Indicates runtime. This indicates the design rated number of start-stop cycles for the air conditioning equipment. Indicates the number of historical starts and stops. This indicates the current cleanliness of the heat exchanger. This indicates the frequency of failures within the past three months. The actual energy efficiency ratio and load correction factor of the air conditioning equipment are calculated based on the equipment aging and degradation coefficient, as shown below: In the formula, This indicates the current indoor and outdoor temperature difference. This indicates the actual energy efficiency ratio of the air conditioning equipment. Indicates the rated energy efficiency ratio of the equipment. Indicates the temperature difference adaptation coefficient. Indicates the load correction factor; The air conditioning load demand is optimized based on the load correction factor, resulting in the corrected air conditioning load demand, as shown below: In the formula, This indicates the revised air conditioning load demand. Indicates air conditioning load demand. Indicates the need for aging load replenishment. Safety redundancy factor, This indicates the actual maximum output load of the air conditioning equipment. This indicates the rated maximum output load of the air conditioning equipment.

6. The air conditioning energy-saving control method based on graph neural network human behavior prediction according to claim 1, characterized in that, Also includes: Acquire data on personnel behavior prediction results, actual personnel behavior, and the process of implementing air conditioning energy-saving control strategies; The comprehensive evaluation coefficient of the control effect is calculated based on the predicted personnel behavior, actual personnel behavior, and process data, as shown below: In the formula, This represents the comprehensive evaluation coefficient of the regulatory effect. Indicates the assessment of bias in personnel behavior prediction. Indicates an assessment of temperature regulation efficiency. Indicates the achievement of energy conservation targets. Indicates that the comfort level meets the assessment criteria. , , and All represent preset weights. Indicates the first [number] within the preset time period Actual personnel behavior data values ​​of Wei Indicates the first [number] within the preset time period Dimension's predictive human behavior data values, This indicates the predicted completion time of the air conditioning energy-saving control strategy. This indicates the time when the actual air conditioning energy-saving control strategy was completed. This represents the predicted energy consumption of the air conditioning energy-saving control strategy. This represents the energy consumption of the actual air conditioning energy-saving control strategy. This indicates the comfort level feedback score for the personnel. This represents the total dimension of personnel behavior data; Based on the comprehensive evaluation coefficients of the regulation effect, a preset graph neural network iterative correction signal is generated, as shown below: In the formula, This represents the weight adjustment amount in the graph neural network model. Represents the iterative learning rate. This represents the gradient of the prediction loss function; when When the evaluation result of the preset graph neural network is deemed excellent, no model parameters are adjusted; when At that time, the evaluation result of the preset graph neural network was determined to be good, according to Adjust model parameters; when At that time, the evaluation result of the preset graph neural network is determined to be in need of optimization, according to Adjust the model parameters.

7. An air conditioning energy-saving control system based on graph neural network for predicting human behavior, characterized in that, include: The data acquisition module is used to acquire basic scene data of the target space and personnel behavior data at the current time; among which, personnel behavior data includes: personnel coordinates, quantity, identity data and behavior; The behavior prediction module is used to construct a spatial-human spatiotemporal heterogeneous graph based on scene basic data and personnel behavior data, and to perform behavior prediction through a trained spatiotemporal graph neural network to obtain the personnel behavior prediction results in the target space within a preset time period. The demand determination module is used to obtain the basic load of the space based on the acquired equipment data of the target space, and to obtain the dynamic load demand of personnel based on the prediction results of personnel behavior. The strategy determination module is used to determine the air conditioning load demand based on environmental data, spatial base load, and dynamic load demand of personnel, and to determine the air conditioning energy-saving control strategy within the target space based on the air conditioning load demand.

Images