HVAC energy-saving self-adaptive control system and method based on AI optimization

Abstract

The present application relates to the technical field of digital temperature control, in particular to an HVAC energy-saving adaptive control system and method based on AI optimization, which comprises an internal zone heat load analysis module, an external heat penetration quantification module, a wind-water linkage cold quantity optimization module, a cold-electricity conversion efficiency measurement module and an actuator driving reconstruction module. In the present application, by analyzing the superposition process of indoor air specific enthalpy and human body heat dissipation, the disturbance value caused by building envelope thermal inertia is calculated, the energy efficiency attenuation amplitude under variable frequency working condition is quantified, the execution mechanism control loop parameters are reconstructed by using reinforcement learning, the heat load prediction accuracy is effectively improved, the matching degree of cold quantity supply and actual demand is strengthened, the equipment operating frequency is optimized to suppress efficiency decline, the real-time monitoring and dynamic compensation of refrigeration system energy efficiency ratio are realized, mechanical wear and tear and hydraulic imbalance fluctuation are greatly reduced, air conditioning operation stability is enhanced, and the system can achieve efficient energy-saving adjustment goal under all working conditions.

Description

Technical Field

[0001] This invention relates to the field of digital temperature control technology, and in particular to an AI-optimized energy-saving adaptive control system and method for HVAC systems. Background Technology

[0002] Digital temperature control technology mainly involves a system that uses temperature sensors to collect ambient temperature data and combines it with a controller to regulate heating or cooling equipment. Its core components include temperature signal acquisition, setpoint comparison, control command generation, and actuator control. It is widely used in building environment regulation, industrial process control, and equipment operation management. By constructing a temperature feedback loop, it achieves stable maintenance of the target temperature and combines humidity and airflow parameters to form a multi-variable collaborative regulation system. Traditional HVAC energy-saving adaptive control systems, on the other hand, are control systems that regulate the temperature, humidity, and airflow within a building. Their technical focus is on adjusting the operating status of air conditioning equipment under different load conditions and changes in the external environment. They use pre-set upper and lower temperature limits, acquire real-time temperature data through indoor temperature sensors, compare it with the setpoints, and control the compressor start / stop frequency or adjust valve openings based on the deviation. They also switch fan operating speeds based on time period settings and manually adjust operating parameters according to outdoor temperature or seasonal changes. In some scenarios, they manually set the supply and return air temperatures using empirical rules to complete system operation control.

[0003] Traditional HVAC systems rely solely on single-temperature feedback within the building. Because they ignore the coupled effects of human metabolic heat dissipation and environmental humidity on perceived thermal balance, indoor heat load calculations exhibit significant lags and struggle to adapt to unsteady load fluctuations caused by occupant movement. Furthermore, they lack refined quantification methods for outdoor weather disturbances and the thermal inertia effect of the building envelope, resulting in a mismatch between cooling supply and actual demand in both time and space. In addition, the actuators switch between start and stop based on fixed thresholds, failing to avoid mechanical losses and hydraulic imbalances under variable frequency operation, leading to frequent energy waste and increased equipment wear. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an AI-optimized adaptive energy-saving control system and method for HVAC systems.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: the AI-optimized HVAC energy-saving adaptive control system includes: The internal zone thermal load analysis module measures the specific enthalpy of indoor air by temperature and humidity transmitters, combines it with human infrared pyroelectric signals, calculates the heat dissipation of personnel activities, performs unsteady superposition of sensible heat and latent heat, simulates the heat transfer lag and the heat storage and release process of the wall, and generates indoor thermal load intensity index. The external heat penetration quantification module obtains the external atmospheric temperature and solar radiation illuminance, calls the indoor heat load intensity index, calculates the solar radiation absorption and temperature difference heat transfer heat flux density, analyzes the external disturbance and thermal inertia effect of the building envelope, and generates external disturbance interference values. The air-water linkage cooling capacity optimization module calls the external disturbance interference value, monitors the operating parameters of the chiller unit, uses AI neural network to optimize the shaft power under load in multivariable nonlinear space, derives the compressor operating frequency set value, and establishes the compressor operating frequency regulation quantity. The refrigeration-to-electricity conversion efficiency measurement module calls the compressor operating frequency control quantity, analyzes the ratio of real-time electrical power of the chiller unit to theoretical cooling capacity output, quantifies the deviation of the refrigeration performance coefficient under variable frequency operation, evaluates the compressor power loss, and establishes the energy utilization efficiency ratio. The actuator-driven reconfiguration module calls the energy utilization efficiency ratio, uses AI reinforcement learning and combines monitoring and control to optimize the control loop of the terminal equipment, adjusts the response of water valve opening and fan speed, and generates a set of dynamic control strategies for HVAC.

[0006] As a further aspect of the present invention, the indoor heat load intensity index includes sensible heat and latent heat coupling weight, heat equivalent of personnel activity, and space heat flow lag coefficient; the external disturbance interference value includes the thermal capacity delay of the building envelope, the solar heat gain equivalent factor, and the cold and hot boundary offset; the compressor operating frequency control value includes the variable frequency compressor reference speed setting value, the cooling capacity step compensation step size, and the equipment response time constant limit value; the energy utilization efficiency ratio includes the refrigeration performance coefficient decay degree, the partial load energy efficiency ratio deviation term, and the extreme point of power consumption refrigeration conversion; and the HVAC dynamic control strategy set includes the proportional-integral-differential parameter reconstruction matrix, the chilled water valve opening step size, and the compressor variable frequency start-stop dead zone limit.

[0007] As a further aspect of the present invention, the internal zone thermal load analysis module includes: The air enthalpy potential calculation module measures the indoor air specific enthalpy through temperature and humidity transmitters, identifies the indoor air state point, substitutes it into the enthalpy-humidity diagram equation of humid air to obtain the total enthalpy energy per unit mass of air, and outputs the environmental specific enthalpy evolution parameters. The cluster human heat source evolution submodule measures the intensity and coverage area of ​​infrared pyroelectricity of the human body in space based on the environmental enthalpy evolution parameter, and combines the experience of heat dissipation by human metabolism to deduce the equivalent heat dissipation of multiple people in different activity states, and outputs dynamic human heat source tensor. The unsteady-state aggregation submodule of heat load establishes a three-dimensional spatial fluid dynamics hybrid model based on the environmental specific enthalpy evolution parameter and the dynamic human body heat source tensor, calculates the thermodynamic stratification interface changes caused by the sinking of cold air and the rising of hot air, obtains the total heat load accumulation rate, and generates an indoor heat load intensity index.

[0008] As a further aspect of the present invention, the external thermal penetration quantization module includes: The meteorological thermal pressure gradient deconstruction submodule identifies the dry-bulb temperature and normal direct solar radiation intensity of the outdoor environment based on the indoor thermal load intensity index, and deconstructs the meteorological thermal pressure intensity from multiple directions using sky diffuse radiation, outputting the meteorological boundary thermal driving force. The thermal resistance delay evolution module of the building envelope measures the thermal conductivity and heat storage coefficient of the building exterior wall and glass curtain wall according to the meteorological boundary thermal driving force, establishes the finite difference heat transfer equation, solves the heat flow fluctuation curve and hysteresis phase angle formed by solar radiation and temperature difference on the inner surface of the wall, and outputs the heat capacity decay hysteresis constant. The internal and external heat flow convergence calculation module uses the heat capacity attenuation lag constant to call the spatial heat flow lag coefficient in the indoor heat load intensity index, and combines the meteorological boundary thermal driving force and the heat capacity attenuation lag constant to calculate the increase in sensible heat entering the room through the building envelope, quantify the offset of cooling demand caused by the convergence of internal and external heat flows, and output the external disturbance interference value.

[0009] As a further aspect of the present invention, the air-water linkage cooling capacity optimization module includes: The wind and water thermal coupling submodule calculates the frictional heat compensation value generated by the chilled water pump and the air conditioning unit blower when overcoming the pipe network resistance by collecting the pressure difference, flow rate and temperature gradient at both ends of the water and air paths based on the external disturbance interference values, and outputs the mechanical side heat source additional amount. The medium-level heat transfer calculation module analyzes the convective heat transfer between chilled water and air in the surface cooler based on the additional heat source on the mechanical side, calculates the heat transfer effectiveness of cold energy transfer across the medium using the logarithmic average temperature difference method, and outputs the cascade heat transfer resistance factor. The cooling capacity and compressor optimization submodule combines the mechanical side heat source additional quantity and the cascade heat transfer impedance factor, and introduces the particle swarm optimization algorithm to establish a multi-dimensional mapping surface of evaporation temperature, condensation temperature and compressor speed, identify the point where the total energy consumption of the equipment tends to the optimal speed operating point, and generate the compressor operating frequency regulation quantity.

[0010] As a further aspect of the present invention, the cold-to-electric conversion efficiency measurement module includes: The instantaneous cooling-to-electricity ratio tracking submodule measures the real-time cooling capacity produced by the evaporator and the active power of the power grid consumed by the compressor based on the variable frequency compressor reference speed setting value in the compressor operating frequency control quantity, solves the instantaneous cooling-to-electricity conversion rate, and outputs the cooling coefficient of the air conditioner operating condition. The skewed operation loss diagnosis submodule identifies the isentropic compression power value of the compressor under the differentiated volumetric efficiency based on the cooling coefficient of the air conditioner operating conditions, compares the real-time power consumption with the theoretical isentropic compression power value, separates mechanical friction power consumption and internal leakage dissipation, and outputs the compressor deviation loss amount. The energy efficiency degradation deviation submodule compares the cooling coefficient under the air conditioner's operating condition with the performance curve under the design nominal condition, introduces the compressor's deviation loss, analyzes the overall efficiency drop under variable frequency fluctuation conditions, locks the energy efficiency speed range, and outputs the energy utilization efficiency ratio.

[0011] As a further aspect of the present invention, the isentropic compression power value refers to the isentropic compression power value calculated by determining the volumetric efficiency curve of the compressor based on the reference speed setting of the variable frequency compressor, and combining the real-time pressure and real-time temperature at the compressor intake port. The overall efficiency drop refers to the percentage by which the coefficient of performance (COP) of the air conditioner deviates from the performance curve under the design nominal state, and the percentage is weighted and summed with the compressor's deviation in consumption to obtain the overall efficiency drop.

[0012] As a further aspect of the present invention, the actuator drive reconfiguration module includes: The control loop damping optimization submodule uses the attenuation of the cooling performance coefficient in the energy utilization efficiency ratio as a penalty term in the reinforcement learning algorithm, and re-derives the optimal damping ratio of the variable air volume terminal and the PID control loop of the variable frequency water pump through monitoring and control, and outputs the loop response gain parameter. The valve-air linkage dead zone reshaping submodule analyzes the flow characteristic curve of the chilled water regulating valve and the total pressure air volume characteristic of the blower, redefines the insensitive zone range of the actuator based on the real-time load rate of the air conditioner, and outputs the mechanical anti-vibration dead zone boundary. The drive strategy assembly submodule combines the loop response gain parameters and the mechanical anti-vibration dead zone limit, and sets the overshoot ramp function for the compressor to transition from the current Hertz to the target frequency through monitoring and control. It packages and compiles the function into a servo control instruction set readable by the lower-level machine, and generates a dynamic control strategy set for HVAC.

[0013] As a further aspect of the present invention, the re-delineation of the insensitive zone range of the actuator operation refers to extracting the flow gain slope corresponding to the current load from the flow characteristic curve of the chilled water regulating valve according to the changing trend of the real-time load rate of the air conditioner, and determining the supply air pressure fluctuation deviation value in combination with the total pressure air volume characteristics of the supply fan, and setting the product of the flow gain slope and the supply air pressure fluctuation deviation value as the insensitive zone range.

[0014] The AI-optimized HVAC energy-saving adaptive control method is executed based on the aforementioned AI-optimized HVAC energy-saving adaptive control system, and includes the following steps: S1: Collect spatial thermal state parameters through air temperature and humidity sensors, use enthalpy-humidity diagram theory to obtain the total heat state of the air, perform spatial integration of human metabolic heat production based on infrared pyroelectric array, evolve and calculate the spatiotemporal unsteady-state accumulation of cold and heat sources in the building, and generate indoor heat load intensity index. S2: Based on the indoor heat load intensity index, input the outdoor meteorological boundary parameters and the thermal conductivity constant of the building wall, calculate the transient heat flux density driven by solar radiation and temperature difference through the finite difference method, analyze the peak reduction and delay effect of the wall heat capacity on the heat transfer of the inner surface, quantify the additional cooling load that penetrates into the room, and generate external disturbance interference values. S3: Based on the external disturbance values, the operating state points corresponding to the total energy consumption of the equipment are solved by using the thermodynamic and fluid dynamic equations of the chiller unit, water pump and fan group, and AI optimization algorithm. The nonlinear mapping point between the cooling capacity of the evaporator and the Hertz of the compressor is calculated, and the compressor operating frequency control quantity is generated. S4: Based on the compressor operating frequency control, identify the output cooling capacity and power consumption value of the cold source side, use the reverse Carnot cycle deviation method to analyze the isentropic efficiency reduction and mechanical wear power consumption surge caused by speed change, verify the real-time decay degree of the cooling coefficient during air conditioner inverter operation, and output the energy utilization efficiency ratio. S5: Based on the energy utilization efficiency ratio, by monitoring and controlling the efficiency decay section, reinforcement learning is used to reorganize the proportional, integral, and differential parameters of the underlying actuator, dynamically modify the dead zone of the chilled water valve and the response step slope of the blower, eliminate the oscillation loss of the air conditioning network caused by fluid coupling, and generate a set of dynamic control strategies for HVAC.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by acquiring multi-source parameters and integrating them into a layered thermodynamic physical model, precise separation of indoor sensible and latent heat and dynamic mapping of occupant heat dissipation tensor are achieved, significantly improving the timeliness and accuracy of heat load prediction. By combining the analysis of meteorological thermal pressure gradient and wall heat storage delay law, a compensation mechanism for the penetration of external disturbances into the internal environment is established, effectively resisting the interference of severe environmental fluctuations on system stability. The nonlinear mapping law between equipment shaft power and operating frequency is captured by neural network, and the deviation of energy efficiency attenuation caused by speed drift is corrected in real time, enhancing the cooling output and electrical power conversion efficiency. By reshaping the loop damping through reinforcement learning and defining the insensitive dead zone of the actuator, the oscillation of the pipeline network caused by fluid coupling is eliminated, mechanical wear is reduced, and the system is ensured to always be in the high-efficiency operating range under partial load. Attached Figure Description

[0016] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the internal zone thermal load analysis module in this invention; Figure 3 This is a flowchart of the external heat penetration quantization module in this invention; Figure 4 This is a flowchart of the cooling capacity optimization module in the air-water linkage of this invention; Figure 5 This is a flowchart of the cold-to-electric conversion efficiency measurement module in this invention; Figure 6 This is a flowchart of the actuator-driven reconfiguration module in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] Please see Figure 1 This invention provides a technical solution: an AI-optimized HVAC energy-saving adaptive control system includes: The internal zone thermal load analysis module measures the indoor air specific enthalpy through temperature and humidity transmitters, and calculates the heat dissipated by human activities by combining spatial infrared human body pyroelectric signals. It performs unsteady superposition calculations on the sensible and latent heat of the building's interior space, deduces the hysteresis effect of heat transfer and the heat storage and release process of the walls, and generates an indoor thermal load intensity index. The external heat penetration quantification module obtains the external atmospheric temperature and solar radiation irradiance, calls the indoor heat load intensity index, combines the thermal heat transfer coefficient of the building envelope, calculates the heat flux density of solar radiation heat absorption and indoor-outdoor temperature difference heat transfer, analyzes the delayed decay effect caused by the thermal inertia of the envelope, calculates the impact tendency of heat penetration from the outside to the inside on the system's cooling demand, and generates external disturbance interference values. The air-water linkage cooling capacity optimization module calls external disturbance values, monitors the operating hertz of the chiller compressor, the temperature difference between the chilled water supply and return water, the static pressure of the blower and the position of the variable air volume terminal valve, calculates the enthalpy difference balance between the heat carried away by the chilled water circulation and the cooling capacity released by the air supply system, and uses an AI neural network optimization algorithm in a multivariable nonlinear space to solve the shaft power operating point under the current load, derives the corresponding equipment speed setpoint, and establishes the compressor operating frequency control quantity; The refrigeration-electric conversion efficiency measurement module calls the compressor operating frequency regulation quantity to monitor and control it and identify the real-time input power reading of the chiller unit. Based on the refrigeration principle, it calculates the ratio of the theoretically output cooling capacity to the real-time power consumption, quantifies the deviation of the unit's refrigeration performance coefficient under variable frequency operation, identifies the power consumption loss of the compressor in the target frequency range, and establishes the energy utilization efficiency ratio. The actuator-driven reconfiguration module calls the energy utilization efficiency ratio, introduces an AI reinforcement learning model for the deviation of the cooling performance coefficient, monitors and controls the proportional, integral and differential coefficients of the control loop of the reconfigured terminal equipment, calibrates the response step limits of the water valve opening and the fan speed, offsets the equipment operation oscillation caused by mechanical wear or hydraulic imbalance, and generates a set of dynamic control strategies for HVAC.

[0020] Indoor heat load intensity indicators include sensible heat and latent heat coupling weights, heat equivalent of human activity, and space heat flow lag coefficient. External disturbance values ​​include building envelope heat capacity delay, solar heat gain equivalent factor, and cold / hot boundary offset. Compressor operating frequency control values ​​include variable frequency compressor base speed setpoint, cooling capacity step compensation step size, and equipment response time constant limit. Energy efficiency ratios include cooling performance coefficient decay, partial load energy efficiency ratio deviation, and extreme point of power consumption cooling conversion. HVAC dynamic control strategy set includes proportional-integral-differential parameter reconstruction matrix, chilled water valve opening step size, and compressor variable frequency start / stop dead zone limit.

[0021] Please see Figure 2 The internal zone thermal load analysis module includes: The air enthalpy potential calculation module measures the indoor air specific enthalpy through temperature and humidity transmitters, identifies the indoor air state point, substitutes it into the enthalpy-humidity diagram equation of humid air to obtain the total enthalpy energy per unit mass of air, and outputs the environmental specific enthalpy evolution parameters. Temperature and humidity transmitters are driven to acquire dry-bulb temperature and relative humidity parameters of spatial grid nodes at a frequency of 100Hz. The raw data matrix is ​​input into the preprocessing channel of a gate array chip, where a median filtering algorithm is executed to filter out abnormal spikes, generating a smooth temperature and humidity dataset. The standard atmospheric pressure wet air property database in the built-in memory is retrieved, using dry-bulb temperature and relative humidity as the joint search key, and a bilinear interpolation algorithm is used to locate the target air state point. The specific heat capacity of dry air, the specific heat capacity of water vapor, and the latent heat of vaporization of water at 0℃ are extracted for the corresponding state point. These constants are substituted into the enthalpy-humidity diagram of the wet air state equation, and the sensible heat of dry air and the latent heat of water vapor are scalarly summed to calculate the total enthalpy energy per unit mass of air. Parallel computation is performed by traversing spatial nodes with a grid step size of 0.5m to establish a tensor matrix containing three-dimensional coordinates and enthalpy energy. Difference operations are performed on the continuous tensor matrix along the time dimension to analyze the thermodynamic state transition gradient, generating environmental specific enthalpy evolution parameters reflecting thermodynamic evolution characteristics. The data processing is executed independently within the floating-point unit, locking the absolute error of the specific enthalpy calculation within 0.2 kJ / kg, and completing the dynamic tracking and parameter output of the air specific enthalpy.

[0022] The cluster human heat source evolution submodule measures the intensity and coverage area of ​​infrared pyroelectricity of the human body in space based on the environmental enthalpy evolution parameter. Combined with the experience of heat dissipation by human metabolism, it infers the equivalent heat dissipation of multiple people under different activity states and outputs a dynamic human heat source tensor. The system receives environmental enthalpy evolution parameters and synchronously triggers an infrared pyroelectric sensor matrix. It captures the infrared radiation flux of moving targets in space, converts it into an analog voltage signal of 0 to 5V, and outputs a digital pyroelectric infrared image via an analog-to-digital converter. A recognition threshold of 0.8V is set, background thermal noise is filtered out, and pixel sets of connected regions with infrared signal intensity exceeding the boundary are extracted. The infrared pyroelectric intensity of the target is quantified, and the physical coverage area is calculated based on the total number of pixels in the connected regions. The centroid coordinates of the coverage area are differentiated over 5 consecutive seconds to determine the target's movement speed. Activity states are categorized based on movement speed: below 0.1 m / s is classified as a sedentary state; between 0.1 m / s and 1.0 m / s as a light activity state; and above 1.0 m / s as a vigorous activity state. The system calls upon the human metabolic heat mapping matrix to match the corresponding sensible and latent heat dissipation baseline values ​​for each activity state. A matrix multiplication is performed on the total number of identified individuals, activity state category identifiers, and heat dissipation baseline values ​​to deduce the equivalent heat dissipation of multiple individuals in differentiated activity states. The time derivative of the environmental enthalpy evolution parameter is integrated into the equivalent heat dissipation of the cluster, and compensation correction is performed on transient heat flow fluctuations to generate a multi-dimensional feature set containing spatial coordinates, timestamps, and heat generation power, and output a dynamic human body heat source tensor.

[0023] The unsteady aggregation submodule of heat load establishes a three-dimensional spatial fluid dynamics hybrid model based on the environmental specific enthalpy evolution parameter and the dynamic human body heat source tensor. It calculates the thermodynamic stratification interface changes caused by cold air sinking and hot air rising, obtains the total heat load accumulation rate, and generates indoor heat load intensity index. A three-dimensional spatial fluid dynamics hybrid model is established in a digital twin engine by receiving the environmental enthalpy evolution parameter and the dynamic human body heat source tensor. The finite volume method is used to divide the three-dimensional physical space into a hexahedral orthogonal computational grid with a side length of 0.1m. The environmental enthalpy evolution parameter is used as the background energy field loading condition for the flow field, and the dynamic human body heat source tensor is injected into the corresponding coordinate grid as a discrete volume heat source term. The Navier-Stokes partial differential equations are invoked, combined with the k-epsilon turbulence model, to solve for the mass conservation, momentum conservation, and energy conservation matrices. A Bousines approximation term is introduced in the vertical direction of the momentum equation to solve for the buoyancy driving force caused by the change in fluid density due to temperature differences. Iterative calculations are performed until the energy transfer residual is below 0.0001, indicating that the flow field has reached transient convergence. Temperature distribution slice data at various altitude levels are extracted, and abrupt change zones with an absolute temperature gradient greater than 1.5℃ / m are identified. The thermodynamic stratification interface changes caused by the sinking of cold air and the rising of hot air are calculated, and the physical elevation of the stratification interface is located. The total enthalpy change of the upper and lower regions of the layered interface is volume-integrated along the time step, and the integral term of the heating power of the dynamic human body heat source tensor is superimposed to obtain the total heat load accumulation rate. The peak value of the accumulation rate is normalized according to the heat flow distribution weighting coefficient to generate an indoor heat load intensity index.

[0024] Please see Figure 3The external heat penetration quantization module includes: The meteorological thermal pressure gradient deconstruction submodule is based on the indoor thermal load intensity index, identifies the dry-bulb temperature and normal direct solar radiation illuminance of the outdoor environment, and deconstructs the meteorological thermal pressure intensity from multiple directions using sky diffuse radiation, outputting the meteorological boundary thermal driving force. The system receives indoor thermal load intensity indices and acquires outdoor environmental parameters via a 1-minute cycle through a meteorological station data bus. It collects outdoor dry-bulb temperature sequences and normal direct solar irradiance captured by an all-weather direct radiation meter, using the Laida criterion to clean up radiation data that deviates from the mean. A sky-scattered radiation deconstruction algorithm is introduced, taking normal direct solar irradiance, latitude and longitude coordinates, real-time solar altitude angle, and azimuth angle as inputs. It calculates the difference between the total horizontal radiation and the horizontal projection of direct radiation, separating the isotropic sky-scattered radiation component. Based on the building's orientation and azimuth angle, the normal direct solar irradiance is decomposed into equivalent direct radiation intensities on four vertical facades using the cosine projection theorem. Sky-scattered radiation is distributed to each facade according to the shape factor ratio, deconstructing multi-directional meteorological thermal pressure intensity. A linear combination calculation is performed between the meteorological thermal pressure intensity of each facade and the temperature difference driving force between the outdoor dry-bulb temperature and the indoor set temperature. The weighting parameter of radiation driving force is set to 0.6, and the weighting parameter of temperature difference driving force is set to 0.4. The weighted summation yields the comprehensive thermodynamic gradient of each boundary, and the dataset containing the direction vector and heat flux scalar is encapsulated to output the meteorological boundary thermal driving force.

[0025] The thermal resistance delay evolution module of the building envelope measures the thermal conductivity and heat storage coefficient of the building exterior wall and glass curtain wall based on the meteorological boundary thermal driving force, establishes the finite difference heat transfer equation, solves the heat flow fluctuation curve and hysteresis phase angle formed by solar radiation and temperature difference on the inner surface of the wall, and outputs the heat capacity decay hysteresis constant. The meteorological boundary thermal driving force is obtained, and the material list of the building envelope is retrieved from the Building Information Modeling (BIM) database to obtain the physical and thermal properties of the building's exterior walls and glass curtain walls. The thermal conductivity, density, and specific heat capacity parameters of each structural layer of the exterior wall are extracted, and their product is calculated to obtain the heat storage coefficient of each layer. A one-dimensional discretized spatial grid is established along the thickness direction of the exterior envelope, with a grid step size of 5 mm, and a finite difference heat transfer equation is established. The Crank-Nicholson implicit difference scheme is adopted, using the meteorological boundary thermal driving force as the time-varying boundary condition for the outer surface and the set indoor temperature as the boundary condition for the inner surface. With a step size of 60 s, the implicit difference iterative matrix solution is performed on the heat conduction process, outputting the temperature matrix of each depth node. The temperature sequence of the innermost grid node is extracted, and combined with the inner surface heat transfer coefficient, the dynamic value of the heat flux density released from the inner surface to the interior is calculated. A comparison calculation is performed between the peak times of the meteorological boundary thermal driving force wave and the peak times of the heat flux density wave on the inner surface. The difference between the two on the time axis is calculated, and the heat flux fluctuation curves and hysteresis phase angles formed by solar radiation and temperature difference on the inner surface of the wall are plotted. The peak hysteresis time of the composite exterior wall is extracted and converted into a quantitative parameter characterizing thermal inertia, and the heat capacity decay hysteresis constant is output.

[0026] The internal and external heat flow convergence calculation module uses the heat capacity decay lag constant to call the spatial heat flow lag coefficient in the indoor heat load intensity index. It combines the meteorological boundary thermal driving force and the heat capacity decay lag constant to calculate the increase in sensible heat entering the room through the building envelope, quantifies the offset of cooling demand caused by the convergence of internal and external heat flows, and outputs the external disturbance interference value. Obtain the heat capacity attenuation lag constant and call the spatial heat flow lag coefficient in the indoor heat load intensity index. Shift the time series vector of the meteorological boundary thermal driving force backward by the corresponding time length of the heat capacity attenuation lag constant to align the external thermal shock with its actual penetration to the inner surface thermal effect in the time dimension. Multiply the shifted meteorological boundary thermal driving force amplitude by the corresponding comprehensive heat transfer coefficient and area of ​​the building envelope, and introduce the spatial heat flow lag coefficient to perform attenuation correction calculations to estimate the increase in sensible heat entering the room through the building envelope after double filtering. Perform scalar summation on the calculated sensible heat increments of the building envelope in each orientation at the current moment to obtain the increased indoor heat load caused by external climate fluctuations. Superimpose this sensible heat increment onto the baseline steady-state cooling demand base to quantify the cooling demand offset caused by the convergence of internal and external heat flows. This offset calculation objectively reflects the net interference caused by external heat source fluctuations on the cooling capacity after considering thermal inertia. Encapsulate the cooling demand offset into a standard transmission message and output the external disturbance interference value. After applying the interference compensation algorithm, the prediction error rate of load fluctuations is stably converged to within 3.1%, and the heat flow convergence calculation is completed.

[0027] Please see Figure 4 The feng shui-linked cooling capacity optimization module includes: The wind and water thermal coupling submodule calculates the frictional heat compensation value generated by the chilled water pump and the air conditioning unit blower when overcoming the pipe network resistance by collecting the pressure difference, flow rate and temperature gradient at both ends of the water and air paths based on the external disturbance interference values, and outputs the mechanical side heat source additional amount. External disturbance values ​​are received, and differential pressure signals at 10Hz are acquired from both ends of the water and air paths via differential pressure transmitters and Pitot tube wind pressure sensors installed in the pipeline network. Ultrasonic flow meters are used to collect the volumetric flow rate of the chilled water main, and hot-wire anemometers are used to measure the wind speed in the main air supply duct and convert it into air supply flow rate. Temperature gradient data is obtained using platinum resistance thermometers, and the rated efficiency curves of the pumps and fans are read from their nameplates. Based on Bernoulli's equation, the acquired differential pressure and actual flow rate are multiplied to calculate the fluid mechanical work consumed in overcoming pipeline resistance. The fluid mechanical work is divided by the overall mechanical efficiency of the motor at the current speed to obtain the motor shaft power. The difference between the shaft power and the fluid mechanical work is calculated. Based on the law of conservation of energy, the energy lost due to viscous shear friction and impeller mechanical friction is completely converted into heat from the temperature rise of the fluid medium. The frictional heat compensation value generated by the chilled water pump and the air conditioning unit's blower when overcoming pipeline resistance is analyzed. The external disturbance values ​​and the accumulated friction heat compensation values ​​from both the wind and water paths are combined using a matrix and added to a scalar to construct a data structure that reflects the heat dissipation effect of the equipment's mechanical components, and outputs the additional heat source quantity on the mechanical side.

[0028] The medium-level heat transfer calculation module analyzes the convective heat transfer between chilled water and air in the surface cooler based on the additional heat source on the mechanical side, calculates the heat transfer effectiveness of cold energy transfer across the medium using the logarithmic average temperature difference method, and outputs the cascade heat transfer resistance factor. The additional heat source on the mechanical side is obtained, and a mathematical analytical model of convective heat transfer is established for the surface cooler inside the air conditioning unit. Four temperature gradient parameters—inlet water temperature, outlet water temperature, inlet air temperature, and outlet air temperature—are extracted from the surface cooler, and the maximum and minimum temperature differences between the hot and cold fluids at both ends of the surface cooler are calculated. Using the core formula of the logarithmic average temperature difference method, the difference between the maximum and minimum temperature differences is divided by the ratio of their natural logarithms to calculate the actual average heat transfer temperature difference for cross-medium heat transfer. The polynomial fitting equation of the standard heat transfer area of ​​the surface cooler and the comprehensive heat transfer coefficient under the current airflow rate is retrieved to solve for the theoretical maximum heat transfer under the current conditions. The actual sensible heat absorbed by the surface cooler is compared with the theoretical maximum heat transfer by division to measure the heat transfer effectiveness parameters for cross-medium heat transfer. The baseline heat transfer efficiency constant is set to 0.8. The penalty ratio term for the deviation of the actual heat transfer efficiency from the baseline value is calculated. The penalty ratio term is coupled with the additional thermal resistance of the mechanical heat source and the additional thermal resistance to perform a coupled multiplication operation. The attenuation resistance matrix of the cold energy transferred from the water side to the wind side is quantified, and the cascade heat transfer impedance factor is output to complete the heat transfer efficiency deconstruction and impedance parameter output.

[0029] The cooling capacity and compressor optimization submodule combines the mechanical side heat source additional quantity and the cascade heat transfer impedance factor, introduces the particle swarm optimization algorithm to establish a multi-dimensional mapping surface of evaporation temperature, condensation temperature and compressor speed, identifies the equipment's total energy consumption to tend to the optimal speed operating point, and generates compressor operating frequency control quantity. The additional heat source on the mechanical side and the cascade heat transfer impedance factor are obtained and loaded into the particle swarm optimization algorithm engine, with the goal of minimizing total power consumption. The particle swarm size parameter is initialized to 50, the maximum number of iterations is set to 100, the inertia weight factor is linearly decreased from 0.9 to 0.4, and both the individual learning factor and the social learning factor are set to 2.0. Evaporation temperature, condensation temperature, and compressor speed are defined as the coordinate axes of the three-dimensional search space. In a single iteration, the particle's current position coordinates are substituted into the total energy consumption simulation equation including the pre-factor to calculate the total power consumption fitness value at that coordinate point. The individual's current fitness is compared with the historical best fitness to refresh the individual's optimal position; the global optimal position is refreshed by comparing the optimal fitness of all particles. The particle optimization direction and step size are continuously adjusted through the velocity update formula to establish a multi-dimensional mapping surface of evaporation temperature, condensation temperature, and compressor speed. When the iteration reaches the preset threshold or the global optimal fitness value changes by less than 0.1% for 10 consecutive iterations, the optimization process is frozen, the global optimal particle coordinates after convergence are extracted, the total energy consumption of the equipment tends to the optimal speed operating point, the inverter output Hertz parameter corresponding to the optimal operating point is extracted, and the compressor operating frequency control quantity is generated.

[0030] Please see Figure 5 The cold-to-electric conversion efficiency measurement module includes: The instantaneous cooling-to-electricity ratio tracking submodule measures the real-time cooling capacity produced by the evaporator and the active power of the grid consumed by the compressor based on the variable frequency compressor reference speed set value in the compressor operating frequency regulation, solves the instantaneous cooling-to-electricity conversion rate, and outputs the cooling coefficient of the air conditioner operating condition. The system reads the compressor's operating frequency control value and the reference speed setpoint of the variable frequency compressor. It polls the high-frequency electromagnetic flowmeter and the dual platinum resistance thermometer on the evaporator water-side pipeline to capture the real-time chilled water mass flow rate and the temperature difference between the evaporator inlet and outlet water. The system multiplies the mass flow rate, water constant pressure specific heat capacity, and inlet / outlet water temperature difference to measure the real-time cooling capacity produced by the evaporator. It collects the effective voltage, effective current, and power factor using power meters installed on the power supply bus. The system measures the active power consumed by the compressor using a three-phase active power formula. The system divides the real-time cooling capacity by the consumed active power to calculate the instantaneous cooling-to-electricity conversion rate. It performs arithmetic averaging and noise reduction on 60 instantaneous cooling-to-electricity conversion rate sample sequences collected within one minute, filtering out outliers caused by voltage transient fluctuations. The smoothed cooling-to-electricity conversion rate mean is encapsulated into a floating-point data packet to construct a steady-state cooling performance coefficient index array. This process simultaneously updates the performance coefficient register state, continuously outputting the cooling coefficient under the air conditioning operating conditions.

[0031] The skewed operation loss diagnosis submodule identifies the isentropic compression power value of the compressor under different volumetric efficiencies based on the cooling coefficient of the air conditioner. It compares the real-time power consumption with the theoretical isentropic compression power value, separates the mechanical friction power consumption and internal leakage dissipation, and outputs the compressor deviation loss. The isentropic compression power value refers to the volumetric efficiency curve of the compressor determined based on the reference speed setting of the variable frequency compressor, combined with the real-time pressure and temperature at the compressor intake port, to calculate the isentropic compression power value. The system obtains the coefficient of performance (COP) of the air conditioner under operating conditions and analyzes the isentropic compression power of the compressor under differentiated volumetric efficiencies. It extracts the baseline speed setpoint, calls the unit controller's speed-volume efficiency mapping curve matrix, and retrieves the theoretical volumetric efficiency corresponding to the set speed. Using absolute pressure sensors and temperature probes at the suction and discharge ports, it collects real-time suction pressure, suction temperature, and discharge pressure. It retrieves the refrigerant thermodynamic property parameter library and calculates the isentropic compression power using the adiabatic isentropic compression equation, combined with the suction and discharge pressure ratio, adiabatic index, and theoretical volumetric efficiency. It reads the measured real-time active power consumption of the compressor, subtracts the constant copper and iron losses of the motor's stator and rotor, and obtains the mechanical shaft power. It compares the real-time active power consumption with the theoretical isentropic compression power, performs a subtraction operation to solve for the deviation, analyzes the internal composition ratio of the deviation, and separates the mechanical friction power consumption and internal leakage dissipation components by combining speed and lubricating oil temperature difference parameters. It normalizes and encodes the separated refrigerant high-pressure backflow internal leakage dissipation scaling value, packages it to generate loss characteristic parameters, and outputs the compressor deviation loss amount.

[0032] The sub-module for energy efficiency degradation deviation compares the cooling coefficient under air conditioning operating conditions with the performance curve under the design nominal state, introduces the compressor deviation loss, analyzes the overall efficiency drop under variable frequency fluctuation conditions, locks the energy efficiency speed range, and outputs the energy utilization efficiency ratio. Overall performance drop refers to the percentage by which the coefficient of performance (COP) of the air conditioner deviates from the performance curve under the design nominal condition. The percentage is then weighted and summed with the compressor's deviation in losses to obtain the overall performance drop. The system obtains the coefficient of performance (COP) for the air conditioning system under operating conditions. It retrieves a 3D performance surface data table generated from tests conducted under standard operating conditions, using real-time condenser inlet water temperature and evaporator outlet water temperature as dual independent variables. A bicubic spline interpolation algorithm is used to calculate the theoretical nominal COP for the current operating condition. The obtained COP for the air conditioning system under operating conditions is divided by the theoretical nominal COP, and this ratio is subtracted from 1 to calculate the percentage deviation of the COP from the design nominal state. An internal leakage dissipation percentage from the compressor's deviation loss is introduced, with a weighting factor of 0.7 for the percentage deviation and 0.3 for the internal leakage dissipation scale value. A weighted sum is performed on the percentage and the compressor's deviation loss to analyze the overall efficiency drop under variable frequency fluctuation conditions. The overall efficiency drop across historical speed ranges is sorted in descending order, and continuous speed frequency bands with drop rates below 5% are extracted. This energy efficiency speed range is then locked and set as a hard constraint boundary for variable frequency control. The overall efficiency drop is then reciprocated and normalized to output the energy utilization efficiency ratio.

[0033] Please see Figure 6 The actuator-driven reconfiguration module includes: The control loop damping optimization submodule uses the attenuation of the cooling performance coefficient in the energy utilization efficiency ratio as a penalty term in the reinforcement learning algorithm, and re-derives the optimal damping ratio of the variable air volume terminal and the PID control loop of the variable frequency water pump through monitoring and control, and outputs the loop response gain parameter. The degradation rate of the coefficient of performance (COP) within the energy utilization efficiency ratio is extracted, and a deep deterministic strategy gradient reinforcement learning algorithm model is activated. Tensor concatenation is performed on the state vectors of real-time cooling load, ambient temperature and humidity deviation, and valve opening to construct the input layer state feature tensor. The model's action space output is defined as the proportional gain and integral gain of the variable air volume (VAV) terminal control loop, and the optimal damping ratio of the variable frequency pump control loop. In the reinforcement learning reward function definition, the COP degradation rate is defined as the core penalty term. The model network includes an input layer, a fully connected hidden layer using a modified linear unit activation function, and an output layer using a hyperbolic tangent activation function. Multi-scene action simulation is performed in a virtual simulation environment through monitoring and control. Historical state transition samples are extracted and loaded into the experience replay pool. Gradient descent is used to update the network weight matrix using time difference error. After the optimization iteration terminates, the action vector parameter combination that makes the global cumulative reward value converge to the maximum value is extracted, and the optimal proportional gain, integral gain, and damping ratio are analyzed to output the loop response gain parameter.

[0034] The valve-air linkage dead zone reshaping submodule analyzes the flow characteristic curve of the chilled water regulating valve and the total pressure air volume characteristic of the blower, redefines the insensitive zone range of the actuator based on the real-time load rate of the air conditioner, and outputs the mechanical anti-vibration dead zone boundary. The scope of the insensitive zone for the actuator is redefined by extracting the flow gain slope corresponding to the current load from the flow characteristic curve of the chilled water regulating valve based on the real-time load rate change trend of the air conditioner, and determining the supply air pressure fluctuation deviation value in combination with the total pressure air volume characteristics of the supply fan. The product of the flow gain slope and the supply air pressure fluctuation deviation value is set as the insensitive zone. The flow characteristic curve of the chilled water regulating valve and the total pressure and air volume characteristics of the blower are analyzed. A pre-set polynomial relating the valve opening to the flow rate is retrieved from memory. Based on the real-time load rate change trend of the air conditioning system, the flow characteristic curve of the chilled water regulating valve is differentiated to extract the flow gain slope corresponding to the current load operating point. The quadratic function characteristic equation of the total pressure and air volume of the blower at the current speed is retrieved. Combined with the intersection point of the duct network resistance curve, the extreme value of the allowable supply air pressure fluctuation deviation under this operating condition is determined. The extracted flow gain slope and the supply air pressure fluctuation deviation value are numerically multiplied to calculate the physical scale of the dead zone width. Using the current control reference point as the central axis, the aforementioned physical scale is extended equidistantly to both sides to redefine the range of the actuator's insensitive action zone. This range is entered into the control command interception filter to determine the range to which the newly generated control command change belongs. When the calculated control command change falls within the insensitive zone, the command increment is forcibly cleared, the transmission link of frequently adjusted control signals is cut off, and the mechanical anti-vibration dead zone boundary is output.

[0035] The drive strategy set assembly submodule combines the loop response gain parameter and the mechanical anti-vibration dead zone limit, and sets the compressor's overshoot ramp function from the current Hertz to the target frequency through monitoring and control. It is then packaged and compiled into a servo control instruction set readable by the lower-level machine to generate a dynamic control strategy set for HVAC. The receiver loop response gain parameter and mechanical anti-vibration dead zone limit are used to read the compressor target frequency output from the front-end module. The actual output Hertz of the inverter is obtained through the sensor bus, and the difference between the actual output Hertz and the target frequency is compared. The control algorithm module compares the difference with the dead zone limit threshold to determine the action to be executed. The overshoot-free ramp function model is retrieved, and the maximum frequency change rate of the inverter is set to 1.5Hz / s. Based on the difference amplitude and the preset ramp rate, the discretized frequency setpoint sequence of the transition from the current Hertz to the target frequency is calculated. The time-frequency setpoint sequence and the loop response gain parameter execution data are packaged, hexadecimal encoded according to the industrial communication protocol message format, and a servo control instruction set readable by the lower-level machine is compiled. The packaged message is sent to the communication register between the programmable logic controller and the inverter to generate a dynamic control strategy set for HVAC, closing the control link. After introducing dead zone reshaping and smoothing strategy control, the abnormal oscillation action of the actuator is eliminated, the steady-state response time is reduced by 22%, and adaptive optimization closed-loop execution is completed.

[0036] The AI-optimized HVAC energy-saving adaptive control method is implemented based on the aforementioned AI-optimized HVAC energy-saving adaptive control system and includes the following steps: S1: Collect spatial thermal state parameters through air temperature and humidity sensors, use enthalpy-humidity diagram theory to obtain the total heat state of the air, perform spatial integration of human metabolic heat production based on infrared pyroelectric array, evolve and calculate the spatiotemporal unsteady-state accumulation of cold and heat sources in the building, and generate indoor heat load intensity index. S2: Based on the indoor heat load intensity index, input the outdoor meteorological boundary parameters and the thermal conductivity constant of the building wall, solve the transient heat flux density driven by solar radiation and temperature difference through the finite difference method, analyze the peak reduction and delay effect of the wall heat capacity on the heat transfer of the inner surface, quantify the additional cooling load that penetrates into the room, and generate external disturbance interference values. S3: Based on external disturbance values, the system uses the thermodynamic and fluid dynamic equations of the chiller unit, water pump, and fan group, and employs an AI optimization algorithm to solve the operating state point corresponding to the total energy consumption of the equipment, calculates the nonlinear mapping point between the evaporator cooling capacity and the compressor Hertz, and generates the compressor operating frequency control quantity. S4: Based on the compressor operating frequency regulation, identify the output cooling capacity and power consumption value on the cold source side, use the reverse Carnot cycle deviation method to analyze the isentropic efficiency reduction and mechanical wear power consumption surge caused by speed change, verify the real-time decay of the cooling coefficient during air conditioner inverter operation, and output the energy utilization efficiency ratio. S5: Based on the energy utilization efficiency ratio, through monitoring and control of the efficiency decay section, reinforcement learning is used to reorganize the proportional, integral, and differential parameters of the underlying actuator, dynamically modify the dead zone of the chilled water valve and the response step slope of the blower, eliminate the oscillation loss of the air conditioning network caused by fluid coupling, and generate a set of dynamic control strategies for HVAC.

[0037] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. An AI-optimized adaptive energy-saving control system for HVAC systems, characterized in that, The system includes: The internal zone thermal load analysis module measures the specific enthalpy of indoor air by temperature and humidity transmitters, combines it with human infrared pyroelectric signals, calculates the heat dissipation of personnel activities, performs unsteady superposition of sensible heat and latent heat, simulates the heat transfer lag and the heat storage and release process of the wall, and generates indoor thermal load intensity index. The external heat penetration quantification module obtains the external atmospheric temperature and solar radiation illuminance, calls the indoor heat load intensity index, calculates the solar radiation absorption and temperature difference heat transfer heat flux density, analyzes the external disturbance and thermal inertia effect of the building envelope, and generates external disturbance interference values. The air-water linkage cooling capacity optimization module calls the external disturbance interference value, monitors the operating parameters of the chiller unit, uses AI neural network to optimize the shaft power under load in multivariable nonlinear space, derives the compressor operating frequency set value, and establishes the compressor operating frequency regulation quantity. The refrigeration-to-electricity conversion efficiency measurement module calls the compressor operating frequency control quantity, analyzes the ratio of real-time electrical power of the chiller unit to theoretical cooling capacity output, quantifies the deviation of the refrigeration performance coefficient under variable frequency operation, evaluates the compressor power loss, and establishes the energy utilization efficiency ratio. The actuator-driven reconfiguration module calls the energy utilization efficiency ratio, uses AI reinforcement learning and combines monitoring and control to optimize the control loop of the terminal equipment, adjusts the response of water valve opening and fan speed, and generates a set of dynamic control strategies for HVAC.

2. The AI-optimized HVAC energy-saving adaptive control system according to claim 1, characterized in that, The indoor heat load intensity index includes sensible heat and latent heat coupling weight, heat equivalent of personnel activity, and space heat flow lag coefficient. The external disturbance interference values ​​include the heat capacity delay of the building envelope, the solar heat gain equivalent factor, and the cold and hot boundary offset. The compressor operating frequency control values ​​include the variable frequency compressor reference speed set value, the cooling capacity step compensation step size, and the equipment response time constant limit value. The energy utilization efficiency ratio includes the refrigeration performance coefficient decay degree, the partial load energy efficiency ratio deviation term, and the extreme point of power consumption refrigeration conversion. The HVAC dynamic control strategy set includes the proportional-integral-differential parameter reconstruction matrix, the chilled water valve opening step size, and the compressor variable frequency start-stop dead zone limit.

3. The AI-optimized HVAC energy-saving adaptive control system according to claim 1, characterized in that, The internal zone thermal load analysis module includes: The air enthalpy potential calculation module measures the indoor air specific enthalpy through temperature and humidity transmitters, identifies the indoor air state point, substitutes it into the enthalpy-humidity diagram equation of humid air to obtain the total enthalpy energy per unit mass of air, and outputs the environmental specific enthalpy evolution parameters. The cluster human heat source evolution submodule measures the intensity and coverage area of ​​infrared pyroelectricity of the human body in space based on the environmental enthalpy evolution parameter, and combines the experience of heat dissipation by human metabolism to deduce the equivalent heat dissipation of multiple people in different activity states, and outputs dynamic human heat source tensor. The unsteady-state aggregation submodule of heat load establishes a three-dimensional spatial fluid dynamics hybrid model based on the environmental specific enthalpy evolution parameter and the dynamic human body heat source tensor, calculates the thermodynamic stratification interface changes caused by the sinking of cold air and the rising of hot air, obtains the total heat load accumulation rate, and generates an indoor heat load intensity index.

4. The AI-optimized HVAC energy-saving adaptive control system according to claim 3, characterized in that, The external heat penetration quantization module includes: The meteorological thermal pressure gradient deconstruction submodule identifies the dry-bulb temperature and normal direct solar radiation intensity of the outdoor environment based on the indoor thermal load intensity index, and deconstructs the meteorological thermal pressure intensity from multiple directions using sky diffuse radiation, outputting the meteorological boundary thermal driving force. The thermal resistance delay evolution module of the building envelope measures the thermal conductivity and heat storage coefficient of the building exterior wall and glass curtain wall according to the meteorological boundary thermal driving force, establishes the finite difference heat transfer equation, solves the heat flow fluctuation curve and hysteresis phase angle formed by solar radiation and temperature difference on the inner surface of the wall, and outputs the heat capacity decay hysteresis constant. The internal and external heat flow convergence calculation module uses the heat capacity attenuation lag constant to call the spatial heat flow lag coefficient in the indoor heat load intensity index, and combines the meteorological boundary thermal driving force and the heat capacity attenuation lag constant to calculate the increase in sensible heat entering the room through the building envelope, quantify the offset of cooling demand caused by the convergence of internal and external heat flows, and output the external disturbance interference value.

5. The AI-optimized HVAC energy-saving adaptive control system according to claim 4, characterized in that, The air-water linkage cooling capacity optimization module includes: The wind and water thermal coupling submodule calculates the frictional heat compensation value generated by the chilled water pump and the air conditioning unit blower when overcoming the pipe network resistance by collecting the pressure difference, flow rate and temperature gradient at both ends of the water and air paths based on the external disturbance interference values, and outputs the mechanical side heat source additional amount. The medium-level heat transfer calculation module analyzes the convective heat transfer between chilled water and air in the surface cooler based on the additional heat source on the mechanical side, calculates the heat transfer effectiveness of cold energy transfer across the medium using the logarithmic average temperature difference method, and outputs the cascade heat transfer resistance factor. The cooling capacity and compressor optimization submodule combines the mechanical side heat source additional quantity and the cascade heat transfer impedance factor, and introduces the particle swarm optimization algorithm to establish a multi-dimensional mapping surface of evaporation temperature, condensation temperature and compressor speed, identify the point where the total energy consumption of the equipment tends to the optimal speed operating point, and generate the compressor operating frequency regulation quantity.

6. The AI-optimized HVAC energy-saving adaptive control system according to claim 5, characterized in that, The cryogenic conversion efficiency measurement module includes: The instantaneous cooling-to-electricity ratio tracking submodule measures the real-time cooling capacity produced by the evaporator and the active power of the power grid consumed by the compressor based on the variable frequency compressor reference speed setting value in the compressor operating frequency control quantity, solves the instantaneous cooling-to-electricity conversion rate, and outputs the cooling coefficient of the air conditioner operating condition. The skewed operation loss diagnosis submodule identifies the isentropic compression power value of the compressor under the differentiated volumetric efficiency based on the cooling coefficient of the air conditioner operating conditions, compares the real-time power consumption with the theoretical isentropic compression power value, separates mechanical friction power consumption and internal leakage dissipation, and outputs the compressor deviation loss amount. The energy efficiency degradation deviation submodule compares the cooling coefficient under the air conditioner's operating condition with the performance curve under the design nominal condition, introduces the compressor's deviation loss, analyzes the overall efficiency drop under variable frequency fluctuation conditions, locks the energy efficiency speed range, and outputs the energy utilization efficiency ratio.

7. The AI-optimized HVAC energy-saving adaptive control system according to claim 6, characterized in that, The isentropic compression power value refers to the isentropic compression power value calculated based on the volumetric efficiency curve of the compressor determined by the reference speed setting of the variable frequency compressor, combined with the real-time pressure and real-time temperature of the compressor intake port. The overall efficiency drop refers to the percentage by which the coefficient of performance (COP) of the air conditioner deviates from the performance curve under the design nominal state, and the percentage is weighted and summed with the compressor's deviation in consumption to obtain the overall efficiency drop.

8. The AI-optimized HVAC energy-saving adaptive control system according to claim 6, characterized in that, The actuator-driven reconfiguration module includes: The control loop damping optimization submodule uses the attenuation of the cooling performance coefficient in the energy utilization efficiency ratio as a penalty term in the reinforcement learning algorithm, and re-derives the optimal damping ratio of the variable air volume terminal and the PID control loop of the variable frequency water pump through monitoring and control, and outputs the loop response gain parameter. The valve-air linkage dead zone reshaping submodule analyzes the flow characteristic curve of the chilled water regulating valve and the total pressure air volume characteristic of the blower, redefines the insensitive zone range of the actuator based on the real-time load rate of the air conditioner, and outputs the mechanical anti-vibration dead zone boundary. The drive strategy assembly submodule combines the loop response gain parameters and the mechanical anti-vibration dead zone limit, and sets the overshoot ramp function for the compressor to transition from the current Hertz to the target frequency through monitoring and control. It packages and compiles the function into a servo control instruction set readable by the lower-level machine, and generates a dynamic control strategy set for HVAC.

9. The AI-optimized HVAC energy-saving adaptive control system according to claim 8, characterized in that, The re-delineation of the insensitive zone range for the actuator's operation refers to extracting the flow gain slope corresponding to the current load from the flow characteristic curve of the chilled water regulating valve based on the changing trend of the real-time air conditioning load rate, and determining the supply air pressure fluctuation deviation value in conjunction with the total pressure air volume characteristics of the supply fan. The product of the flow gain slope and the supply air pressure fluctuation deviation value is then set as the insensitive zone range.

10. An AI-optimized adaptive energy-saving control method for HVAC systems, characterized in that, The method is used to implement the AI-optimized HVAC energy-saving adaptive control system according to any one of claims 1-9, and includes the following steps: S1: Collect spatial thermal state parameters through air temperature and humidity sensors, use enthalpy-humidity diagram theory to obtain the total heat state of the air, perform spatial integration of human metabolic heat production based on infrared pyroelectric array, evolve and calculate the spatiotemporal unsteady-state accumulation of cold and heat sources in the building, and generate indoor heat load intensity index. S2: Based on the indoor heat load intensity index, input the outdoor meteorological boundary parameters and the thermal conductivity constant of the building wall, calculate the transient heat flux density driven by solar radiation and temperature difference through the finite difference method, analyze the peak reduction and delay effect of the wall heat capacity on the heat transfer of the inner surface, quantify the additional cooling load that penetrates into the room, and generate external disturbance interference values. S3: Based on the external disturbance values, the operating state points corresponding to the total energy consumption of the equipment are solved by using the thermodynamic and fluid dynamic equations of the chiller unit, water pump and fan group, and AI optimization algorithm. The nonlinear mapping point between the cooling capacity of the evaporator and the Hertz of the compressor is calculated, and the compressor operating frequency control quantity is generated. S4: Based on the compressor operating frequency control, identify the output cooling capacity and power consumption value of the cold source side, use the reverse Carnot cycle deviation method to analyze the isentropic efficiency reduction and mechanical wear power consumption surge caused by speed change, verify the real-time decay degree of the cooling coefficient during air conditioner inverter operation, and output the energy utilization efficiency ratio. S5: Based on the energy utilization efficiency ratio, by monitoring and controlling the efficiency decay section, reinforcement learning is used to reorganize the proportional, integral, and differential parameters of the underlying actuator, dynamically modify the dead zone of the chilled water valve and the response step slope of the blower, eliminate the oscillation loss of the air conditioning network caused by fluid coupling, and generate a set of dynamic control strategies for HVAC.

Images