Decorative air supply air conditioner AI energy-saving regulation and control system based on XGBoost prediction

Abstract

The invention discloses a decoration air supply air conditioner AI energy-saving regulation and control system based on XGBoost prediction, and relates to the technical field of building environment control, the system comprises a vertical decoration column air pipe structure, a plurality of layers of built-in layered air pipes are arranged along the height of a column on the basis of a vertical hollow steel structure column matched with the building space layout, an air outlet of each layer faces a corresponding subarea, and the vertical decoration column air pipe structure is provided with a plurality of layers of built-in layered air pipes; the column body is covered with a ventilation decorative net which is detachably connected in a buckled mode, and each layer of air pipe is independently provided with an adjusting component and communicated with an air conditioner host. Simulation data and historical collected data are fused through the collaborative prediction module, thermal load and thermal comfort indexes are pre-judged in advance, a reliable basis is provided for regulation and control decision making, the AI regulation and control decision making module is combined with layered air duct characteristics, differential regulation and control strategies are generated, air supply of an air conditioner is accurately matched with dynamic thermal loads of all areas, and the regulation and control efficiency is improved. The thermal comfort requirement of people is met, and energy waste caused by excessive refrigeration or heating is avoided.

Description

Technical Field

[0001] This invention relates to the field of building environment control technology, specifically to an AI energy-saving control system for decorative air conditioning based on XGBoost prediction. Background Technology

[0002] In the field of architecture, large-space, high-density buildings such as opera houses and theaters have extremely high requirements for the control of the internal microenvironment. As a key facility for controlling the internal environment, the air conditioning system needs to accurately meet the thermal comfort needs of people and the functional needs of the building. Vertical decorative column air supply schemes, due to their unique advantages, can combine architectural aesthetics with ventilation needs and have gradually become a commonly used ventilation method in such buildings. It not only helps to improve the overall aesthetics of the building, but also optimizes the indoor airflow organization to a certain extent, providing people with a relatively comfortable indoor environment.

[0003] However, existing vertical decorative column duct structures and related air conditioning control systems have many problems. First, there is a lack of standardized design in the vertical decorative column duct structure. The column dimensions, duct layering methods, and ventilation component parameters have not been standardized. This leads to repeated adjustments to the geometric model and boundary conditions during CFD simulation modeling, which not only complicates the modeling process but also makes it difficult to guarantee the accuracy of the simulation, thus affecting the accurate prediction of the indoor thermal environment. Second, existing air conditioning control systems fail to fully consider the characteristics of the duct layering structure. Their control logic mainly relies on fixed procedures or manual operation. This control method cannot adapt to dynamic heat load changes caused by factors such as personnel movement and equipment heat dissipation in a timely manner, making it difficult to achieve energy-saving goals while meeting the thermal comfort needs of personnel. In addition, there is a lack of effective coordination mechanism between structural design, simulation analysis, load prediction, and control execution. The system response is lagging and cannot respond to changes in the thermal environment in advance, resulting in an unstable indoor thermal environment that affects the comfort of personnel and the normal use of the building. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI-powered energy-saving control system for decorative air conditioning based on XGBoost prediction. This system provides a unified parameter basis for CFD simulation through the standardized design of vertical decorative column duct structures, simplifying the simulation modeling process and improving simulation accuracy. Simultaneously, the collaborative prediction module integrates simulation data with historical data to train the prediction model, achieving accurate prediction of heat load and thermal comfort indicators. The AI ​​control decision module formulates differentiated control strategies based on the characteristics of layered ducts, ensuring precise matching of air conditioning supply with the heat load requirements of each area. The feedback correction module dynamically adjusts the prediction model parameters by comparing actual monitoring values ​​with predicted values, ensuring the system's prediction accuracy and control stability. This effectively solves the problems existing in existing technologies, achieving synergistic optimization of thermal comfort and energy saving, and is suitable for the control of air conditioning systems in large-space buildings such as large opera houses and theaters.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an AI energy-saving control system for decorative air conditioning based on XGBoost prediction, the system comprising: Vertical decorative column duct structure: Based on the vertical hollow steel structure column adapted to the building space layout, multiple layers of built-in layered ducts are set along the column height. The air outlet of each layer faces the corresponding zone. The column is covered with a detachable buckle-connected permeable decorative net. Each layer of duct is equipped with an independent adjustment component and is connected to the air conditioning unit. Data acquisition module: Based on the layered duct layout and building functional zoning, environmental parameter sensors, duct operation parameter sensors and personnel status acquisition devices are deployed in the corresponding areas. Thermal comfort indicators are calculated according to preset logic, and all data is transmitted to the system backend wirelessly. Collaborative prediction module: Based on the standardized parameters of the vertical decorative column duct structure, a CFD model is generated and meshed. After loading preset boundary conditions, a turbulence model is used to complete the simulation. The simulation results are fused with historical data to train the XGBoost prediction model. Real-time data is input and predicted values ​​of future heat load and thermal comfort index are output. AI control and decision-making module: With thermal comfort and energy saving as the goals, a multi-objective optimization logic is constructed. Constraints on thermal comfort, duct operation and structural safety are set. After solving the multi-objective optimization algorithm, the optimal control parameters of each layer of duct are generated by combining the functional differences of the layered ducts. Execution module: Through servo motor drive components, frequency converter control components and angle control motor, it controls the tiered duct adjustment components, air conditioning unit and air outlet guide vanes respectively. After receiving the optimal adjustment parameters, it performs linkage actions to realize the synchronous adjustment of air volume, air temperature and air direction. Feedback correction module: Compares the actual data with the predicted data after adjustment and calculates the deviation. Based on the deviation threshold, it maintains the model parameters, dynamically corrects the feature weights of the XGBoost model, or updates the CFD boundary conditions and retrains the prediction model.

[0006] Furthermore, the data acquisition module, based on the layered duct layout and building functional zoning, evenly distributes temperature and humidity sensors, wind speed sensors, and CO2 concentration sensors within the corresponding functional zones of each layered duct to collect environmental parameters. Airflow sensors and temperature sensors are respectively placed at the inlet and outlet sections of each layered duct to collect duct operating parameters. Video acquisition devices with AI image recognition capabilities are deployed at personnel flow nodes to obtain personnel density and flow rate. The data processing logic, based on the collected temperature, humidity, and wind speed parameters, generates thermal comfort indices in real time according to a preset thermal comfort index calculation logic. All acquisition devices and processing logic are connected via a wireless transmission protocol to transmit environmental parameters, operating parameters, personnel status parameters, and thermal comfort indices to the system backend.

[0007] Furthermore, the collaborative prediction module receives standardized parameters of the vertical decorative column duct structure, automatically generates a three-dimensional CFD geometric model, and then uses an unstructured mesh to divide the model into grids. The mesh is refined for the interior of the layered ducts, near the permeable decorative mesh, and in areas with high personnel activity. For non-critical areas outside the columns, a conventional mesh density is used to balance simulation accuracy and computational efficiency. Subsequently, preset boundary conditions are loaded, including the permeability resistance coefficient of the permeable decorative mesh, the reference air supply temperature and airflow of each layered duct, human metabolic heat dissipation parameters, and heat dissipation parameters of stage lighting and other equipment. The RNG k-ε turbulence model is used for simulation calculations, outputting the simulation results of heat load distribution, temperature field distribution, wind speed field distribution, and thermal comfort index distribution for each functional zone. Then, a dataset is constructed... The dataset includes simulated data from CFD simulation calculations and actual data collected during historical operations. After preprocessing, the data is divided into training and validation sets. Using environmental parameters, personnel status parameters, and duct operation parameters as input features, and heat load, PMV value, and PPD value as output labels, an XGBoost prediction model is constructed. The learning rate, tree depth, and regularization coefficient parameters of the model are optimized through optimization algorithms to determine the optimal hyperparameter combination with the goal of minimizing the prediction error of the validation set. Finally, the current data transmitted from the data acquisition module is received in real time, converted into a format that matches the model's input features, and then input into the trained XGBoost model. The model outputs the predicted heat load, PMV, and PPD values ​​for each functional area within a preset future time period.

[0008] Furthermore, the collaborative prediction module uses the RNG k-ε turbulence model for simulation calculations, and its turbulent kinetic energy k transport equation is calculated as follows: The dissipation rate ε is calculated using the transport equation as follows: ,in, This refers to air density, specifically the air density under standard atmospheric pressure. It refers to the time step in CFD simulation. yes The directional airflow velocity component corresponds to the airflow velocity in different areas within the opera house. yes The spatial coordinates of the directions correspond to the three-dimensional spatial layout of the opera house. It is the viscosity coefficient of air molecules. It is the turbulent viscosity coefficient. , yes and The turbulent Prandtl number, It is a turbulent kinetic energy generation term, driven by the airflow velocity gradient and heat dissipation from personnel / equipment.

[0009] Furthermore, the collaborative prediction module optimizes the model's learning rate, tree depth, and regularization coefficient parameters using an optimization algorithm. The algorithm formula is as follows: ,in, It is the first The total error of the model in each iteration. This is the sample size. The mean squared error loss function is calculated using the following formula: , For the true value, For the front Tree prediction value, This is the first Tree pairs of samples The predicted contribution It is a regularization term, and its calculation formula is: , It is a complexity control parameter used to penalize splitting of tree nodes. The number of leaf nodes. The weights are those of the leaf nodes.

[0010] Furthermore, the AI ​​control and decision-making module constructs a multi-objective optimization function. The input variables of the function include the air volume and air temperature of each layer of ducts, and the objective items include thermal comfort compliance and air conditioning energy consumption. The priority of the two objective items is adjusted by weighting coefficients. Then, three types of constraints are set: thermal comfort constraints, duct operation constraints, and structural safety constraints. Subsequently, an improved particle swarm optimization algorithm is used to solve the multi-objective optimization function. The convergence speed and optimal solution accuracy of the solution process are improved by dynamically adjusting the inertia weight and learning factor. Finally, based on the air volume and air temperature of each layer of ducts obtained from the optimization solution, and combined with the functional zoning characteristics of the corresponding layer of ducts, a differentiated control strategy is generated. For the bottom layer ducts corresponding to the near-personnel area, the air temperature is preferentially adjusted to the target value. For the middle layer ducts corresponding to the densely populated area, the air volume is preferentially adjusted to the target value. For the upper layer ducts corresponding to the building area, the air volume and air temperature are simultaneously adjusted to the target values. Finally, the optimal control parameters of the opening of the adjustment components of each layer of ducts, the air supply power of the air conditioning unit, and the air outlet guide angle are output.

[0011] Furthermore, the AI ​​control decision module employs an improved particle swarm optimization algorithm to solve the multi-objective optimization function. First, it defines the control parameter combination for each particle corresponding to each layer of the opera house's ventilation ducts, sets the population size, and clarifies the constraint boundaries of the particle positions. Then, it dynamically adjusts the inertia weight, initially setting it to 0.8-0.9, gradually decreasing it to 0.4-0.5 as iterations progress. Simultaneously, it dynamically adjusts the learning factor, increasing the self-awareness learning factor in the early stages and the social awareness learning factor in the later stages. In each iteration, the particle velocity is first updated based on the current inertia weight and learning factor, then the particle position is adjusted based on the velocity, and feasible region correction is performed for particles exceeding the constraint range. Subsequently, the fitness of each particle is calculated, and the individual optimal position of each particle and the population's optimal position are updated. When the iteration reaches its maximum number of iterations or the fitness of the population's optimal position changes by less than 0.01 for 3-5 consecutive generations, the iteration stops, and the control parameters corresponding to the population's optimal position are output.

[0012] Furthermore, the execution module includes drive control logic and execution components. After receiving the optimal control parameters output by the AI ​​control decision module, the drive control logic converts the parameters into control signals for the execution components. A servo motor drive component is configured for the adjustment component of each layer of ductwork, and the adjustment component is driven to move according to the opening control signal of the adjustment component to achieve precise control of the air supply volume. A frequency converter control component is configured for the air conditioning unit, and the cooling / heating power and fan speed of the unit are adjusted according to the air supply temperature and air supply volume requirements. Electric guide vanes are set at the air outlet of each layer of ductwork, and an angle control motor is configured to adjust the vane angle according to the guide angle control signal so that the air supply direction covers the corresponding functional area.

[0013] Furthermore, the feedback correction module receives the adjusted actual data transmitted by the data acquisition module in real time, compares it with the predicted data output by the collaborative prediction module parameter by parameter, and calculates the absolute and relative deviations of each parameter. Then, a deviation threshold is set. When the deviation is less than the first threshold, the parameters of the current XGBoost prediction model are maintained. When the deviation is greater than the first threshold but less than the second threshold, the feature weights of the model are dynamically corrected. When the deviation is greater than the second threshold, the recalibration of the CFD simulation boundary conditions is triggered. The ventilation resistance of the ventilation decorative net and the heat dissipation parameters of the equipment under the current state are collected to update the simulation boundary conditions. At the same time, the latest collected actual data is called to supplement the dataset and the XGBoost prediction model is retrained.

[0014] Furthermore, the feedback correction module calculates the absolute and relative deviations of each parameter, and the formula for calculating the absolute deviation is as follows: The formula for calculating the relative deviation is: ,in, It is the absolute deviation of a certain parameter. It is the relative deviation of a certain parameter. These are the actual monitored values ​​of the parameters. These are the predicted values ​​of the parameters.

[0015] Compared with existing technologies, this AI energy-saving control system for decorative air conditioning based on XGBoost prediction has the following advantages: I. This invention integrates simulation data and historical data through a collaborative prediction module to predict heat load and thermal comfort indicators in advance, providing a reliable basis for control decisions. Furthermore, the AI ​​control decision module combines the characteristics of layered ducts to generate differentiated control strategies, enabling precise matching of air conditioning supply with the dynamic heat load of each area. This satisfies the thermal comfort needs of personnel while avoiding energy waste caused by excessive cooling or heating. The feedback correction module dynamically adjusts the prediction model parameters or updates the simulation boundary conditions by comparing actual monitoring values ​​with predicted values ​​in real time, ensuring that the system is always in the optimal operating state. This effectively solves the problems of slow response and difficulty in balancing thermal comfort and energy saving in existing systems.

[0016] Second, this invention standardizes the vertical decorative column duct structure design, unifies the column size, duct layering method and ventilation component parameters, provides a unified parameter basis for CFD simulation, avoids the complex process of repeatedly adjusting geometric models and boundary conditions, and significantly improves the efficiency and accuracy of simulation modeling. The ventilation decorative mesh adopts a detachable snap-fit ​​connection design, and the layered ducts are independently configured with adjustment components, which reduces the operational complexity and maintenance cost of system maintenance.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 This is a block diagram of an AI energy-saving control system for decorative air conditioning based on XGBoost prediction. Figure 2 This is a flowchart of an AI-powered energy-saving control system for decorative air conditioning based on XGBoost prediction. Detailed Implementation

[0020] 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 below.

[0021] Example 1 Based on the architectural layout of the opera house, a vertical hollow steel structure column adapted to the overall design is used as the basic carrier. Three layers of built-in layered air ducts are set along the height of the column. The air outlets of the first layer of air ducts face the audience area of ​​the stalls, the air outlets of the middle layer of air ducts correspond to the stage and side stage areas, and the air outlets of the upper layer of air ducts cover the audience area of ​​the balcony. The exterior of the column is covered with a detachable buckle-connected permeable decorative net to ensure aesthetics and easy maintenance. Each layer of layered air ducts is independently equipped with adjustment components and is connected to the central air conditioning unit of the opera house through a dedicated pipe to ensure a stable airflow supply.

[0022] The data acquisition module is designed according to a three-layer duct layout and the three functional zones of the opera house: the stalls, the stage, and the balcony. Temperature and humidity sensors, wind speed sensors, and CO2 concentration sensors are evenly distributed in each zone to collect environmental parameters in real time. Air volume sensors and temperature sensors are installed at the inlet and outlet of each duct layer to accurately capture duct operating parameters. Video acquisition devices with AI image recognition capabilities are deployed at personnel flow nodes such as the opera house entrance, audience aisles, and stage side doors to obtain personnel density and flow rate in different areas. Based on the collected temperature, humidity, and wind speed parameters, thermal comfort indicators are generated in real time according to the preset thermal comfort index calculation logic. All collected environmental parameters, operating parameters, personnel status parameters, and thermal comfort indicators are synchronously transmitted to the system backend via a wireless transmission protocol.

[0023] The collaborative prediction module receives standardized parameters of the vertical decorative column duct structure and automatically generates a 3D CFD geometric model of the opera house's duct system. An unstructured mesh is used to divide the model, with mesh refinement applied to the interior of the layered ducts, the vicinity of the permeable decorative mesh, and densely populated areas such as the audience seating and stage. A conventional mesh density is used for non-critical areas outside the columns to balance simulation accuracy and computational efficiency. Preset boundary conditions are applied, including the ventilation resistance coefficient of the permeable decorative mesh, the reference air supply temperature and airflow of each layered duct, human metabolic heat dissipation parameters, and heat dissipation parameters of stage lighting and sound equipment. The RNG k-ε turbulence model is used for simulation calculation, and its turbulent kinetic energy k-transport equation is calculated as follows: The dissipation rate ε is calculated using the transport equation as follows: ,in, This refers to air density, specifically the air density under standard atmospheric pressure. It refers to the time step in CFD simulation. yes The directional airflow velocity component corresponds to the airflow velocity in different areas within the opera house. yes The spatial coordinates of the directions correspond to the three-dimensional spatial layout of the opera house. It is the viscosity coefficient of air molecules. It is the turbulent viscosity coefficient. , yes and The turbulent Prandtl number, This is a turbulent kinetic energy generation term, driven by airflow velocity gradient and personnel / equipment heat dissipation. It outputs simulation results of heat load distribution, temperature field distribution, wind speed field distribution, and thermal comfort index distribution for each functional zone. A dataset is constructed, containing simulated data from CFD simulation output and actual data collected during the opera house's historical operation. After preprocessing, it is divided into training and validation sets. Environmental parameters, personnel status parameters, and duct operation parameters are used as input features, and heat load, PMV value, and PPD value are used as output labels. An XGBoost prediction model structure is constructed, and the learning rate, tree depth, and regularization coefficient parameters of the model are optimized using an optimization algorithm. The algorithm formula is as follows: ,in, It is the first The total error of the model in each iteration. This is the sample size. The mean squared error loss function is calculated using the following formula: , For the true value, For the front Tree prediction value, This is the first Tree pairs of samples The predicted contribution It is a regularization term, and its calculation formula is: , yes , The number of leaf nodes. The optimal hyperparameter combination is determined by minimizing the prediction error of the validation set, with the leaf node weights as the objective. During the optimization process, the mean squared error loss function is used to calculate the error, and the regularization term is used to control the model complexity. The current data transmitted by the data acquisition module is received in real time, converted into a format that matches the model input features, and then input into the trained XGBoost model. The output is the predicted heat load, PMV, and PPD values ​​of each functional area in the future preset time period.

[0024] The AI-powered control and decision-making module constructs a multi-objective optimization function. Input variables are the air volume and temperature of each duct layer. Objectives include thermal comfort compliance and air conditioning energy consumption. The priority of these two objectives is adjusted using weighting coefficients. Considering the high density of people during opera performances, the weight of thermal comfort compliance is appropriately increased. Three types of constraints are set: thermal comfort constraints ensure that the PMV and PPD values ​​of each area are within a reasonable range; duct operation constraints limit the adjustment range of air volume and temperature; and structural safety constraints prevent the duct operating pressure from exceeding the design threshold. An improved particle swarm optimization algorithm is used to solve the problem. Solve the multi-objective optimization function, dynamically adjust the inertia weight and learning factor to improve the solution efficiency and the accuracy of the optimal solution. Based on the optimization results, generate differentiated control strategies according to the functional zoning characteristics of each layer of air ducts: the bottom layer air ducts correspond to the pool seating area, and prioritize adjusting the supply air temperature to the target value; the middle layer air ducts correspond to the stage and side stage areas with high personnel density, and prioritize adjusting the supply air volume to the target value; the upper layer air ducts correspond to the balcony seating area, and simultaneously adjust the supply air volume and supply air temperature to the target values. Finally, output the optimal control parameters for the opening of the adjustment components of each layer of air ducts, the supply air power of the air conditioning unit, and the air outlet guide angle.

[0025] After receiving the optimal control parameters output by the AI ​​control decision module, the execution module's drive control logic converts them into control signals recognizable by the execution components. Servo motor drive components are configured for the adjustment components of each layer of ductwork. The adjustment components are driven to move according to the opening control signal of the adjustment components to achieve precise control of the air supply volume. Variable frequency control components are configured for the air conditioning unit. Based on the air supply temperature and air supply volume requirements, the cooling / heating power and fan speed of the unit are adjusted. Electric guide vanes are installed at the air outlet of each layer of ductwork, and angle control motors are configured to adjust the vane angle according to the guide angle control signal to ensure that the air supply direction accurately covers the corresponding functional zones and meets the airflow requirements of different areas.

[0026] The feedback correction module receives the actual data after regulation transmitted by the data acquisition module in real time, compares it parameter by parameter with the predicted data output by the collaborative prediction module, and calculates the absolute and relative deviations of each parameter using the absolute deviation calculation formula and the relative deviation calculation formula. The absolute deviation calculation formula is as follows: The formula for calculating the relative deviation is: ,in, It is the absolute deviation of a certain parameter. It is the relative deviation of a certain parameter. These are the actual monitored values ​​of the parameters. The predicted values ​​of the parameters are set with a deviation threshold. When the deviation is less than the first threshold, the parameters of the current XGBoost prediction model are maintained. When the deviation is greater than the first threshold but less than the second threshold, the feature weights of the model are dynamically corrected. When the deviation is greater than the second threshold, the recalibration of the CFD simulation boundary conditions is triggered. The ventilation resistance of the ventilation decorative net, the heat dissipation parameters of stage equipment and human body under the current state are collected to update the simulation boundary conditions. At the same time, the latest collected actual data is called to supplement the dataset and the XGBoost prediction model is retrained.

[0027] Example 2 Based on the architectural layout of the atrium of a large commercial complex, vertical hollow steel structure columns that match the decoration style of the atrium are selected. Four layers of built-in layered air ducts are set along the height of the columns. The air outlets of the first layer of air ducts face the shopping area on the first floor of the atrium, the second layer of air ducts correspond to the second-floor catering area, the third layer of air ducts cover the third-floor office area, and the fourth layer of air ducts are directed to the public leisure area at the top of the atrium. The exterior of the columns is equipped with detachable clip-on permeable decorative nets to balance the decorative effect and future maintenance. Each layer of layered air ducts is equipped with independent adjustment components and is connected to the central air conditioning unit of the commercial complex through ventilation ducts to ensure airflow supply to each area.

[0028] The data acquisition module is designed according to a four-layer duct layout and functional zoning of each floor in the atrium. Temperature and humidity sensors, wind speed sensors, and CO2 concentration sensors are evenly distributed in the shopping area on the first floor, the dining area on the second floor, the office area on the third floor, and the leisure area on the fourth floor. The density of CO2 concentration sensors is appropriately increased in the dining area on the second floor to accurately collect environmental parameters in each area. Air volume sensors and temperature sensors are installed at the inlet and outlet sections of each duct to collect duct operation parameters in real time. Video acquisition devices with AI image recognition capabilities are deployed at personnel flow nodes such as escalator entrances and exits in the atrium, entrances and exits in the dining area, and passageways in the office area to obtain personnel density and flow rate in each area. Based on the collected temperature, humidity, and wind speed parameters, thermal comfort indicators are generated in real time according to the preset thermal comfort index calculation logic. All environmental parameters, operation parameters, personnel status parameters, and thermal comfort indicators are uniformly transmitted to the system backend via a wireless transmission protocol.

[0029] The collaborative prediction module receives standardized parameters of the vertical decorative column duct structure and automatically generates a 3D CFD geometric model of the atrium duct system in the commercial complex. It uses an unstructured mesh to divide the model, refining the mesh in densely populated areas such as the interior of the layered ducts, near the permeable decorative mesh, the dining area, and the ground-floor shopping area. For non-critical areas outside the columns, a conventional mesh density is used to balance simulation accuracy and computational efficiency. Preset boundary conditions are applied, including the ventilation resistance coefficient of the permeable decorative mesh, the reference air supply temperature and airflow of each layered duct, human metabolic heat dissipation parameters, and heat dissipation parameters of cooking equipment in the dining area and electrical appliances in the office area. The RNG k-ε turbulence model is used for simulation calculations, outputting simulation results of heat load distribution, temperature field distribution, wind speed field distribution, and thermal comfort index distribution for each functional zone. A dataset is constructed and integrated with the CFD simulation. The output simulated data and the actual collected data from the historical operation of the commercial complex are preprocessed and divided into training and validation sets. Using environmental parameters, personnel status parameters, and duct operation parameters as input features, and heat load, PMV value, and PPD value as output labels, an XGBoost prediction model structure is constructed. The learning rate, tree depth, and regularization coefficient parameters of the model are optimized through optimization algorithms. The optimal hyperparameter combination is determined with the goal of minimizing the prediction error of the validation set. The mean squared error loss function is used to calculate the error during the optimization process. The regularization term is combined to avoid model overfitting. The current data from the data acquisition module is received in real time, converted into a format that matches the input features of the model, and then input into the trained XGBoost model. The model outputs the predicted heat load, PMV, and PPD values ​​for each functional area of ​​each floor in the future preset time period.

[0030] The AI-powered control and decision-making module constructs a multi-objective optimization function. Input variables include the air volume and temperature of each duct layer. Objectives include thermal comfort compliance and air conditioning energy consumption. Priorities are adjusted using weighting coefficients. Due to the high heat dissipation from equipment in the dining area, the weight for thermal comfort compliance is appropriately increased. Three types of constraints are set: thermal comfort constraints limit the range of PMV and PPD indicators for each area; duct operation constraints define the adjustment limits of air volume and temperature; and structural safety constraints ensure that the duct operating pressure does not exceed design standards. An improved particle swarm optimization algorithm is used to solve the multi-objective optimization function, dynamically adjusting the inertia weight and learning factor. Initially, the inertia weight is set to... The initial value was 0.8-0.9, gradually decreasing to 0.4-0.5 in the later stages. The self-awareness learning factor was increased in the early stages, and the social awareness learning factor was increased in the later stages to improve the convergence speed and the accuracy of the optimal solution. Based on the optimization results and the functional characteristics of each layer of ductwork, control strategies were generated: the first-layer ductwork corresponds to the shopping area, prioritizing the adjustment of the air supply temperature; the second-layer ductwork corresponds to the densely populated catering area, prioritizing the adjustment of the air supply volume; the third-layer ductwork corresponds to the office area, simultaneously adjusting the air supply volume and temperature; the fourth-layer ductwork corresponds to the leisure area, adjusting according to a balance between thermal comfort and energy saving. The final output was the optimal control parameters for the opening of the adjustment components of each layer of ductwork, the air supply power of the air conditioning unit, and the guide angle of the air outlet.

[0031] After receiving the optimal control parameters from the AI ​​control decision module, the execution module's drive control logic converts them into control signals for the execution components. Each layer of the ductwork's adjustment components are equipped with servo motor drive components, which operate according to the opening control signal to achieve precise control of the air supply volume. The air conditioning unit is equipped with a frequency converter control component, which adjusts the cooling / heating power and fan speed according to the air supply temperature and air supply volume requirements. Each ductwork outlet is equipped with electric guide vanes and an angle control motor, which adjusts the vane angle according to the guide angle control signal to ensure that the air supply direction covers the corresponding functional areas and meets the airflow requirements of different scenarios such as shopping, dining, and office.

[0032] The feedback correction module receives the actual data after adjustment transmitted by the data acquisition module in real time, compares it parameter by parameter with the predicted data of the collaborative prediction module, calculates the absolute and relative deviations of each parameter using the absolute deviation calculation formula and the relative deviation calculation formula, sets deviation thresholds, maintains the current XGBoost model parameters when the deviation is less than the first threshold, dynamically corrects the model feature weights when the deviation is greater than the first threshold but less than the second threshold, and recalibrates the CFD simulation boundary conditions when the deviation is greater than the second threshold. It also collects the current ventilation resistance of the ventilation decorative net, the heat dissipation parameters of catering equipment and office appliances to update the boundary conditions, and supplements the latest actual data into the dataset to retrain the XGBoost prediction model.

[0033] 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. An AI-powered energy-saving control system for decorative air conditioning based on XGBoost prediction, characterized in that, The system includes: Vertical decorative column duct structure: Based on the vertical hollow steel structure column adapted to the building space layout, multiple layers of built-in layered ducts are set along the column height. The air outlet of each layer faces the corresponding zone. The column is covered with a detachable buckle-connected permeable decorative net. Each layer of duct is equipped with an independent adjustment component and is connected to the air conditioning unit. Data acquisition module: Based on the layered duct layout and building functional zoning, environmental parameter sensors, duct operation parameter sensors and personnel status acquisition devices are deployed in the corresponding areas. Thermal comfort indicators are calculated according to preset logic, and all data is transmitted to the system backend wirelessly. Collaborative prediction module: Based on the standardized parameters of the vertical decorative column duct structure, a CFD model is generated and meshed. After loading preset boundary conditions, a turbulence model is used to complete the simulation. The simulation results are fused with historical data to train the XGBoost prediction model. Real-time data is input and predicted values ​​of future heat load and thermal comfort index are output. AI control and decision-making module: With thermal comfort and energy saving as the goals, a multi-objective optimization logic is constructed. Constraints on thermal comfort, duct operation and structural safety are set. After solving the multi-objective optimization algorithm, the optimal control parameters of each layer of duct are generated by combining the functional differences of the layered ducts. Execution module: Through servo motor drive components, frequency converter control components and angle control motor, it controls the tiered duct adjustment components, air conditioning unit and air outlet guide vanes respectively. After receiving the optimal adjustment parameters, it performs linkage actions to realize the synchronous adjustment of air volume, air temperature and air direction. Feedback correction module: Compares the actual data with the predicted data after adjustment and calculates the deviation. Based on the deviation threshold, it maintains the model parameters, dynamically corrects the feature weights of the XGBoost model, or updates the CFD boundary conditions and retrains the prediction model.

2. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The data acquisition module is designed according to the layered duct layout and building functional zoning. Temperature and humidity sensors, wind speed sensors, and CO2 concentration sensors are evenly distributed within the functional zones corresponding to each layered duct to collect environmental parameters. Air volume sensors and temperature sensors are respectively placed at the inlet and outlet sections of each layered duct to collect duct operating parameters. Video acquisition devices with AI image recognition capabilities are deployed at personnel flow nodes to obtain personnel density and flow rate. The data processing logic generates thermal comfort indices in real time based on the collected temperature, humidity, and wind speed parameters and according to the preset thermal comfort index calculation logic. All acquisition devices and processing logic are connected via a wireless transmission protocol to transmit environmental parameters, operating parameters, personnel status parameters, and thermal comfort indices to the system backend.

3. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The collaborative prediction module receives standardized parameters of the vertical decorative column duct structure, automatically generates a 3D CFD geometric model, and then uses an unstructured mesh to divide the model into grids. The mesh is refined for the interior of the layered ducts, near the permeable decorative mesh, and in areas with high personnel activity. For non-critical areas outside the columns, a conventional mesh density is used to balance simulation accuracy and computational efficiency. Subsequently, preset boundary conditions are loaded, including the ventilation resistance coefficient of the permeable decorative mesh, the reference air supply temperature and airflow of each layered duct, human metabolic heat dissipation parameters, and heat dissipation parameters of stage lighting and other equipment. The RNG k-ε turbulence model is used for simulation calculations, outputting the simulation results of heat load distribution, temperature field distribution, wind speed field distribution, and thermal comfort index distribution for each functional zone. Then, a dataset is constructed, which includes... The XGBoost prediction model is constructed using simulated data from CFD simulation calculations and actual data collected during historical operations. After preprocessing, the data is divided into training and validation sets. Environmental parameters, personnel status parameters, and duct operation parameters are used as input features, and heat load, PMV value, and PPD value are used as output labels. The learning rate, tree depth, and regularization coefficient parameters of the model are optimized through optimization algorithms to determine the optimal hyperparameter combination with the goal of minimizing the prediction error of the validation set. Finally, the current data transmitted from the data acquisition module is received in real time, converted into a format that matches the model's input features, and then input into the trained XGBoost model. The model outputs the predicted values ​​of heat load, PMV, and PPD for each functional area in the future preset time period.

4. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 3, characterized in that, The collaborative prediction module uses the RNG k-ε turbulence model for simulation calculations, and its turbulent kinetic energy k transport equation is calculated as follows: The dissipation rate ε is calculated using the transport equation as follows: ,in, This refers to air density, specifically the air density under standard atmospheric pressure. It refers to the time step in CFD simulation. yes The directional airflow velocity component corresponds to the airflow velocity in different areas within the opera house. yes The spatial coordinates of the directions correspond to the three-dimensional spatial layout of the opera house. It is the viscosity coefficient of air molecules. It is the turbulent viscosity coefficient. , yes and The turbulent Prandtl number, It is a turbulent kinetic energy generation term, driven by the airflow velocity gradient and heat dissipation from personnel / equipment.

5. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 3, characterized in that, The collaborative prediction module optimizes the model's learning rate, tree depth, and regularization coefficient parameters using an optimization algorithm. The algorithm formula is as follows: ,in, It is the first The total error of the model in each iteration. This is the sample size. The mean squared error loss function is calculated using the following formula: , For the true value, For the front Tree prediction value, This is the first Tree pairs of samples The predicted contribution It is a regularization term, and its calculation formula is: , It is a complexity control parameter used to penalize splitting of tree nodes. The number of leaf nodes. The weights are those of the leaf nodes.

6. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The AI-driven control and decision-making module constructs a multi-objective optimization function. The input variables of the function include the air volume and air temperature of each layer of ducts, and the objective items include thermal comfort achievement and air conditioning energy consumption. The priority of the two objective items is adjusted by weighting coefficients. Then, three types of constraints are set: thermal comfort constraints, duct operation constraints, and structural safety constraints. Subsequently, an improved particle swarm optimization algorithm is used to solve the multi-objective optimization function. The convergence speed and optimal solution accuracy of the solution process are improved by dynamically adjusting the inertia weight and learning factor. Finally, based on the air volume and air temperature of each layer of ducts obtained from the optimization solution, and combined with the functional zoning characteristics of the corresponding layer of ducts, a differentiated control strategy is generated. For the bottom layer ducts corresponding to the near-personnel area, the air temperature is preferentially adjusted to the target value. For the middle layer ducts corresponding to the densely populated area, the air volume is preferentially adjusted to the target value. For the upper layer ducts corresponding to the building area, the air volume and air temperature are adjusted to the target values ​​simultaneously. Finally, the optimal control parameters of the opening of the adjustment components of each layer of ducts, the air supply power of the air conditioning unit, and the air outlet guide angle are output.

7. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The AI ​​control and decision-making module employs an improved particle swarm optimization algorithm to solve a multi-objective optimization function. First, it defines the control parameter combination for each particle corresponding to the various duct layers of the opera house, sets the population size, and clarifies the constraint boundaries of particle positions. Then, it dynamically adjusts the inertia weight, initially setting it to 0.8-0.9, gradually decreasing it to 0.4-0.5 as iterations progress. Simultaneously, it dynamically adjusts the learning factor, increasing the self-awareness learning factor in the early stages and the social awareness learning factor in the later stages. In each iteration, the particle velocity is first updated based on the current inertia weight and learning factor, then the particle position is adjusted based on the velocity, and feasible region corrections are performed on particles exceeding the constraint range. Subsequently, the fitness of each particle is calculated, and the individual optimal position of each particle and the population's optimal position are updated. Iteration stops when the maximum number of iterations is reached or the fitness of the population's optimal position changes by less than 0.01 for 3-5 consecutive generations, and the control parameters corresponding to the population's optimal position are output.

8. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The execution module includes drive control logic and execution components. After receiving the optimal control parameters output by the AI ​​control decision module, the drive control logic converts the parameters into control signals for the execution components. A servo motor drive component is configured for the adjustment component of each layer of ductwork, and the adjustment component is driven to move according to the opening control signal of the adjustment component to achieve precise control of the air supply volume. A frequency converter control component is configured for the air conditioning unit, and the cooling / heating power and fan speed of the unit are adjusted according to the air supply temperature and air supply volume requirements. Electric guide vanes are set at the air outlet of each layer of ductwork, and an angle control motor is configured to adjust the vane angle according to the guide angle control signal so that the air supply direction covers the corresponding functional area.

9. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 1, characterized in that, The feedback correction module receives the actual data after adjustment transmitted by the data acquisition module in real time, compares it with the predicted data output by the collaborative prediction module, and calculates the absolute and relative deviations of each parameter. Then, it sets a deviation threshold. When the deviation is less than the first threshold, the parameters of the current XGBoost prediction model are maintained. When the deviation is greater than the first threshold but less than the second threshold, the feature weights of the model are dynamically corrected. When the deviation is greater than the second threshold, the recalibration of the CFD simulation boundary conditions is triggered. The ventilation resistance of the ventilation decorative net and the heat dissipation parameters of the equipment under the current state are collected to update the simulation boundary conditions. At the same time, the latest collected actual data is called to supplement the dataset and the XGBoost prediction model is retrained.

10. The decorative air conditioning AI energy-saving control system based on XGBoost prediction according to claim 9, characterized in that, The feedback correction module calculates the absolute and relative deviations of each parameter, and the formula for calculating the absolute deviation is as follows: The formula for calculating the relative deviation is: ,in, It is the absolute deviation of a certain parameter. It is the relative deviation of a certain parameter. These are the actual monitored values ​​of the parameters. These are the predicted values ​​of the parameters.

Images