Radiation floor-central air conditioner integrated cooperative control system and method

Abstract

The invention relates to the technical field of building environment control, and provides a radiation floor-central air conditioner integrated cooperative control system and method.The radiation floor-central air conditioner integrated cooperative control method comprises a dynamic power distribution method and a feedforward-feedback mixed anti-condensation control method, and dynamic power distribution comprises the step of obtaining multi-dimensional environment state data of a to-be-monitored area; constructing a building thermodynamic model, and identifying the total load of the building in real time by using a Kalman filtering algorithm; solving by taking minimization of operation cost, maximization of thermal comfort and system response stability as targets to obtain an optimal power distribution proportion of the radiation floor system and the central air-conditioning system; and optimal load sharing is solved and updated in real time between the radiation floor and the central air conditioner through a dynamic power distribution mechanism, so that the two tail ends operate cooperatively according to sensible heat / latent heat and load change characteristics, and energy superposition and offset are avoided. And meanwhile, feedforward-feedback combined dew point control is adopted, so that the available capability and comfortable stability of radiant cooling are improved on the premise of ensuring no dew formation.

Description

Technical Field

[0001] This disclosure relates to the field of building environment control technology, specifically to an integrated collaborative control system and method based on radiant floor and central air conditioning. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Energy conservation and consumption reduction in the building sector have become an important direction for achieving sustainable social development. Heating, ventilation, and air conditioning (HVAC) systems account for a large proportion of building energy consumption, and improving their energy efficiency is crucial for reducing overall building energy consumption. In recent years, composite air conditioning systems combining radiant heating / cooling systems with convection central air conditioning systems have been increasingly applied and promoted. Radiant floor systems offer advantages such as uniform energy supply, good thermal comfort, quiet operation, and the ability to utilize high-temperature cold sources or low-temperature heat sources, resulting in a high energy efficiency ratio, making them suitable for handling steady-state sensible heat loads in buildings. Central air conditioning systems, on the other hand, offer fast response, strong dehumidification capabilities, and flexible fresh air handling, demonstrating significant advantages in dealing with latent heat loads and rapidly changing load conditions. Therefore, achieving efficient synergistic operation between radiant floor systems and central air conditioning systems, and fully leveraging their respective advantages, is one of the key technical issues that urgently needs to be addressed in the building HVAC field.

[0004] In current engineering practice, radiant floor systems and central air conditioning systems are typically designed and controlled as two independent subsystems, each adjusted independently based on its own temperature or operating parameters, lacking a unified and coordinated control strategy. This separate control mode easily leads to the superposition or even cancellation of cooling and heating outputs between the systems, resulting in energy waste and potentially causing problems such as localized overcooling and decreased thermal comfort, making it difficult to achieve optimal overall system energy efficiency.

[0005] Furthermore, under radiant cooling conditions, condensation easily occurs when the floor surface temperature is lower than the indoor air dew point, severely affecting operational safety and reliability. Existing anti-condensation measures mostly employ increasing the supply water temperature or simple start-stop control strategies. These passive and conservative control methods not only limit the cooling capacity of the radiant floor system but also easily lead to fluctuations in the indoor thermal environment, making it difficult to balance safety and comfort. At the same time, existing control systems are generally based on fixed-parameter PID control. This type of feedback-based control system is essentially a passive response and is ill-suited to the large hysteresis, strong coupling, and nonlinear characteristics of radiant-air conditioning composite systems. The overall adaptability and economy of operation still need improvement. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes an integrated collaborative control system and method based on radiant floor heating and central air conditioning. Based on dynamic power distribution and adaptive control, it achieves integrated collaborative control of radiant floor heating and central air conditioning. Through load distribution optimization, condensation risk avoidance, and system adaptive and predictive control, it effectively improves the economy, comfort, and safety of system operation.

[0007] To achieve the above objectives, the present disclosure adopts the following technical solution: One or more embodiments provide an integrated collaborative control method for radiant floor-to-central air conditioning, including a dynamic power allocation method and a feedforward-feedback hybrid anti-condensation control method. The dynamic power allocation method includes the following steps: Acquire multi-dimensional environmental status data of the area to be monitored; A building thermodynamics model was constructed, and the total building load was identified in real time using the Kalman filter algorithm. To minimize operating costs, maximize thermal comfort, and improve system response stability, a multi-objective optimization function is constructed. ; The sequential quadratic programming algorithm is used to solve the multi-objective optimization function in real time, and the optimal power allocation ratio between the radiant floor system and the central air conditioning system is obtained. Based on the optimal power allocation ratio between the radiant floor system and the central air conditioning system obtained by solving the problem, power allocation is performed based on the logic of spatial vertical gradient partitioning.

[0008] One or more embodiments provide an integrated collaborative control system based on radiant floor-to-central air conditioning, including: A distributed multi-mode sensor network is deployed at multiple vertical gradient levels in the building space to collect multi-dimensional environmental status data of the building space. The intelligent control unit establishes a communication connection with the distributed multi-mode sensor network and is configured to use the above-mentioned integrated collaborative control method based on radiant floor-central air conditioning. An integrated actuator is used to execute control commands to regulate the coordinated operation of the radiant floor system and the central air conditioning system.

[0009] Compared with the prior art, the beneficial effects of this disclosure are as follows: This invention employs a dynamic power allocation mechanism to solve and update the optimal load sharing between the radiant floor and central air conditioning system in real time, enabling the two types of terminals to operate collaboratively according to sensible / latent heat and load change characteristics, thus avoiding energy superposition and cancellation. Simultaneously, it utilizes a feedforward-feedback combined dew point (anti-condensation) control system. Feedforward predicts dew point risks and adjusts them in advance, while feedback corrects deviations in floor surface temperature and humidity, improving the availability and comfort stability of radiant cooling while ensuring no condensation.

[0010] This invention introduces a real-time load identification mechanism based on a building thermodynamics model and Kalman filtering algorithm, enabling dynamic power allocation to be based on accurate load estimation and avoiding the bias problems caused by traditional empirical allocation methods. The multi-objective optimization function J integrates operating costs, thermal comfort, and system response stability into the optimization framework, making the power allocation results of the radiant floor system and central air conditioning system more in line with the optimal overall performance requirements. The application of a sequential quadratic programming algorithm improves the solution efficiency and stability of the method under complex constraints, enabling real-time updates of power allocation as environmental conditions change. The power allocation method based on spatial vertical gradient zoning effectively matches the adjustment advantages of radiant and convective energy supply at different heights, reducing uneven heating and cooling, and improving the overall energy efficiency and indoor thermal environment quality of the system.

[0011] The advantages of this disclosure, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0012] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute a limitation thereof.

[0013] Figure 1 This is a schematic diagram of the overall hardware architecture topology of the integrated collaborative control system based on radiant floor and central air conditioning in Embodiment 1 of this disclosure; Figure 2 This is a schematic diagram illustrating the internal functional module structure and data flow relationship of the intelligent control unit in Embodiment 1 of this disclosure; Figure 3 This is a logic flowchart of the dynamic power allocation strategy based on multi-objective optimization according to Embodiment 1 of this disclosure; Figure 4 This is a block diagram of the signal processing and execution logic of the feedforward-feedback hybrid anti-condensation control strategy in Embodiment 1 of this disclosure; Figure 5 This is a schematic diagram of the model structure based on a long short-term memory neural network for dew point temperature prediction according to Embodiment 1 of this disclosure; Figure 6 This is a schematic diagram showing the comparison of indoor temperature fluctuations and total system energy consumption between a system based on an integrated radiant floor-central air conditioning control system and a system using a traditional independent control mode under typical summer cooling conditions, as described in Embodiment 1 of this disclosure.

[0014] Among them: 100, distributed multi-mode sensor network; 110, sensor array; 120, temperature and humidity transmitter; 130, carbon dioxide concentration sensor; 200, intelligent control unit; 300, integrated actuator; 310, electric three-way proportional-integral regulating valve; 320, variable frequency circulating pump set; 330, variable frequency compressor; 340, electronic expansion valve; 350, DC brushless variable speed fan. Detailed Implementation

[0015] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0017] It should be noted that the terminology used herein is for descriptive purposes only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0018] Example 1 In one or more of the technical solutions disclosed in the implementation methods, such as Figures 1 to 6 As shown, the integrated collaborative control system based on radiant floor and central air conditioning includes: The distributed multi-mode sensor network 100 is deployed at multiple vertical gradient levels in the building space to collect multi-dimensional environmental status data of the building space. The intelligent control unit 200 establishes a communication connection with the distributed multi-mode sensor network 100 and is configured to execute dynamic power allocation and feedforward-feedback hybrid anti-condensation control strategies based on the collected environmental status data, and adaptively calculates control commands for controlling the cooling and heating equipment in the building space. The integrated actuator 300 is used to execute control commands to regulate the coordinated operation of the radiant floor system and the central air conditioning system; In this embodiment, as described above, the system uses a distributed multi-mode sensor network 100 to perceive the building space environment in real time. The distributed multi-mode sensor network 100 is deployed layer by layer along the building's height, covering the near-ground layer, the human activity height layer, and the ceiling area, forming an environmental perception system that reflects the indoor vertical temperature and humidity gradient characteristics. The intelligent control unit 200 fuses and analyzes the environmental status data from the distributed multi-mode sensor network 100, and constructs a coordinated control logic for radiant and convective energy supply based on the building's thermal characteristics and the load sharing relationship between the radiant floor system and the central air conditioning system. Simultaneously, the intelligent control unit 200 introduces a feedforward-feedback hybrid anti-condensation control strategy: by predicting the trend of environmental humidity changes and dew point temperature, it adjusts the energy supply level of the radiant floor system in advance before the floor surface temperature approaches the dew point threshold; and by combining feedback information on floor surface temperature and air conditions, it corrects control commands in real time, thereby maintaining the continuous and stable operation of the radiant cooling system while ensuring that condensation does not occur on the floor surface. The final control commands are executed by the integrated actuator 300 to synchronously adjust the water supply temperature and flow rate of the radiant floor system and the air supply parameters of the central air conditioning system, so as to achieve coordinated operation of the two systems.

[0019] In the above implementation, by deploying a distributed multi-mode sensor network 100 at multiple vertical gradient levels in the building space, the system can comprehensively and precisely grasp the spatial distribution characteristics of the indoor thermal and humidity environment, providing a more accurate data foundation for collaborative control. The intelligent control unit 200 implements dynamic power allocation based on multi-dimensional environmental state data, enabling the radiant floor system and the central air conditioning system to complement each other in terms of load sharing. This avoids the problem of superposition or mutual cancellation of cold and heat outputs in traditional separate control, thereby improving the overall energy efficiency of the system. Compared to simple passive feedback control, the feedforward-feedback hybrid anti-condensation control strategy can identify condensation risks in advance and make proactive adjustments, effectively suppressing condensation without significantly increasing the supply water temperature, balancing operational safety and indoor thermal comfort. Furthermore, this collaborative control method enhances the system's adaptability to load changes and environmental disturbances, reduces energy consumption fluctuations, and improves the stability and economy of the composite air conditioning system during long-term operation.

[0020] In some embodiments, the distributed multi-mode sensor network 100 may include sensors deployed at multiple vertical gradient levels and key physical interfaces in the building space to collect environmental status data in real time, including floor surface temperature distribution matrix, indoor vertical space temperature and humidity field, building envelope inner surface temperature, personnel activity density, carbon dioxide concentration, etc. In the specific deployment and implementation of the distributed multi-mode sensor network 100, in order to obtain high-precision, multi-dimensional building environment field data, this embodiment adopts a three-dimensional sensor deployment scheme in a typical open office building space. Specifically, a one-dimensional or two-dimensional sensor array 110 composed of multiple high-precision NTC thermistors is pre-embedded in the concrete filling layer or leveling layer of the floor, about 5 to 10 millimeters below the decorative surface layer.

[0021] Furthermore, the collected data is stored through a floor surface temperature distribution matrix. The row coordinates of the matrix correspond to the X-axis direction of the building space, and the column coordinates correspond to the Y-axis direction of the building space. Each element value in the matrix precisely corresponds to the measured floor surface temperature value at the corresponding XY coordinate position, forming a temperature distribution data grid with a spatial resolution of no less than 2 meters × 2 meters, which can accurately capture the spatial heterogeneity characteristics of floor surface temperature.

[0022] Optionally, the sensors in the sensor array 110 can be selected with an accuracy class of ±0.1℃. The measuring points are arranged according to the spatial layout with a certain grid spacing, such as 2 meters × 2 meters, to collect data in real time and construct a high-resolution temperature distribution matrix of the floor surface, thereby accurately capturing the highest point, lowest point and average value of the surface temperature.

[0023] In the vertical space dimension of the room, temperature and humidity transmitters 120 are set sequentially according to different heights. After the temperature and humidity transmitters 120 collect data, the temperature and humidity field of the room is represented in the form of a three-dimensional matrix, forming the temperature and humidity field of the room. The first dimension of the matrix is ​​the height level, such as 0.1m, 1.1m, 2.5m. The second dimension is the horizontal area division, which is divided into several sub-areas according to the sensor placement position. The third dimension is the parameter type, such as temperature value and relative humidity value. The temperature and humidity field matrix can be represented as TH(x,y,z), where x∈{height level}, y∈{horizontal sub-area}, z∈{temperature, humidity}. The matrix elements are the measured parameter values ​​at the corresponding positions. This matrix can intuitively reflect the differences in the distribution of indoor temperature and humidity in the vertical and horizontal directions.

[0024] In one specific implementation, this embodiment places composite temperature and humidity transmitters 120 with integrated temperature and humidity sensing functions at three typical height levels: the human foot and ankle area at approximately 0.1 meters above the ground, the head and breathing area of ​​a seated person at approximately 1.1 meters, and the upper air stagnation area at approximately 2.5 meters.

[0025] The composite temperature and humidity transmitter 120 can be connected via RS-485 industrial bus in a daisy chain manner, and the collected data is aggregated to the data acquisition gateway in Modbus-RTU protocol format to monitor the vertical temperature gradient and humidity stratification of indoor space in real time.

[0026] Furthermore, in order to accurately quantify the dynamic load indoors, carbon dioxide concentration sensors 130 based on the non-dispersive infrared (NDIR) absorption principle were deployed in densely populated areas such as main entrances and meeting rooms, and the real-time number of people indoors was inverted by the rate of change of CO2 concentration.

[0027] Optionally, a passive infrared (PIR) pyroelectric array sensor can be deployed to detect the activity and distribution of people.

[0028] The raw data collected by all the aforementioned sensor nodes are initially aggregated and digitized via the nearest I / O module or wireless gateway, and finally transmitted to the intelligent control unit 200.

[0029] In some embodiments, the intelligent control unit 200 is used to receive multi-dimensional environmental state data, perform calculations through an adaptive algorithm engine, and output a set of coordinated control commands for the radiant floor system and the central air conditioning system. like Figure 2 As shown, the intelligent control unit 200 includes a data acquisition interface, a data cleaning and preprocessing module, used to receive multi-dimensional environmental status data and preprocess the received data; The intelligent control unit 200 includes an identification module, a dynamic power distribution module, and an anti-condensation safety control module; The identification module is used to identify the total heat and moisture load of a building based on the constructed building thermodynamic model; The dynamic power allocation module is configured to execute a dynamic power allocation strategy, which constructs a multi-objective optimization function and solves it through a multi-objective optimization solver, dynamically decomposing the total heat and humidity load into the base load power component undertaken by the radiant floor system and the peak-shaving power component undertaken by the central air conditioning system. The anti-condensation safety control module is configured to dynamically adjust the anti-condensation safety margin based on the dew point change rate, and combine a feedforward-feedback hybrid anti-condensation control strategy with long short-term memory network dew point trend prediction and real-time monitoring to perform anti-condensation regulation. The intelligent control unit 200 is the core of the entire collaborative control system, and its hardware carrier can be an industrial-grade embedded computer based on a high-end ARM Cortex-A series processor. This unit possesses powerful edge computing capabilities, and its internal components include an operating system supporting real-time task scheduling and a runtime environment for various algorithm engines. The unit has a rich array of hardware interface modules, receiving data from the distributed multi-mode sensor network 100 via Ethernet or serial ports, and communicating with the building's local weather station or public meteorological service API via standard building automation network protocols BACnet / IP or LonWorks to obtain real-time key external environmental data such as outdoor dry-bulb temperature, relative humidity, wind speed and direction, and total solar irradiance. The intelligent control unit 200's non-volatile memory stores an offline-trained building thermodynamics model, a long short-term memory neural network model, and a multi-objective optimization solver including a sequential quadratic programming (SQP) algorithm library. This control unit is configured to perform a complete loop calculation within a fixed, relatively short control cycle.

[0030] The intelligent control unit 200, based on real-time operating conditions, scientifically and dynamically allocates the building's total heating and cooling load to the radiant floor system and the central air conditioning system. The power of the two systems are respectively... In this embodiment, a multi-objective optimization model is constructed and solved at the software level of the intelligent control unit 200 to achieve this function.

[0031] A further technical solution involves constructing a building thermodynamic model and using a Kalman filter algorithm to identify the total building load in real time, including the following steps: Step 101: Abstract the building space into a multi-order resistive-capacitive network, and establish a state-space model as the building thermodynamic model. The specific formula of the model is as follows: Equations of state: ; Observation equation: ; in, The state vector includes indoor air node temperature, building envelope node temperature, and floor node temperature. The observation vector is represented by the indoor air temperature, floor surface temperature, etc. collected by the sensor network. The input vector is represented by outdoor temperature, solar radiation intensity, and heat generated by indoor people and equipment. A is the state matrix, B is the input matrix, C is the observation matrix, and D is the direct transmission matrix. and These are process noise and observation noise, respectively.

[0032] By using the prediction and update iteration process of the Kalman filter, the optimal estimate of the state vector x is obtained, and the filter residuals are attributed to unmodeled disturbances, thereby identifying the building's real-time total heat and humidity load.

[0033] Step 102: For the constructed building thermodynamic model, obtain the current environmental state data, and identify the total sensible heat load acting on the building space at the current moment using the Kalman filter algorithm. With total latent heat load It includes the following steps: Step 1021: Initialize the Kalman filter parameters, including the initial estimates of the state vector. (0∣0), initial error covariance matrix P(0∣0), process noise covariance matrix Q, and observation noise covariance matrix R; Among them, Q and R are determined through offline calibration based on sensor accuracy and system disturbance characteristics; Step 1022: Based on the current input vector and the state estimate from the previous time step, predict the prior error covariance matrix using the state equation. and the prior estimate of the state vector at the current moment. : ; ; in, for The prior estimate of the state vector at time step [time]. The prior error covariance matrix; Step 1023: Obtain the observation vector at the current time. Based on the predicted prior error covariance matrix, the Kalman gain is calculated. : ; Step 1024: Correct the prior estimate of the state vector based on the Kalman gain to obtain the posterior estimate of the state vector. and posterior error covariance matrix : ; ; in, It is the identity matrix; Step 1025: Calculate the observation residuals By integrating the components representing unmodeled thermal disturbances in the residuals and combining them with the building heat balance equation, the total sensible heat load is obtained. and total latent heat load : ; ; ; in, The total heat and humidity load, and Let be the specific heat capacity and mass of the i-th state node, respectively. Let be the estimated rate of change of the i-th state variable. Let J be the heat loss of the j-th heat transfer path in the building envelope. This represents the percentage of sensible heat load.

[0034] The total sensible heat load acting on the building space at the current moment in step 102 With total latent heat load The identification method mainly includes setting the initial state estimate, error covariance matrix, and process / observation noise covariance, and calibrating according to sensor accuracy; using Kalman filtering to combine input and observation values ​​to obtain the optimal state estimate; calculating the difference between the observation value and the model prediction value, which is the unmodeled thermal and humidity disturbance; the total heat load is the thermal change of each state node plus the residual correction amount, the latent heat load is calculated through the indoor humidity change rate, and the sensible heat load is equal to the total heat load minus the latent heat load.

[0035] At the start of each dynamic power allocation control cycle, the system initiates an online heat load observer based on a Kalman filter algorithm. This observer embeds a simplified building thermodynamics model, such as a second- or third-order RC network model describing heat exchange between the building space and the building envelope. The model utilizes indoor air temperature, floor surface temperature, and inner surface temperature of the building envelope from a distributed multi-mode sensor network 100 as state observations, while taking outdoor temperature, solar radiation intensity, and estimated heat dissipation from people and equipment based on CO2 concentration and activity levels as inputs. Through the recursive prediction and correction process of the Kalman filter, the system not only obtains smooth optimal estimates of state variables such as indoor temperature, but more importantly, it attributes the residuals between the observed values ​​and the model predictions to unmodeled disturbances, thereby accurately identifying the total sensible heat load acting on the building space in real time. With total latent heat load The sum of these two load components constitutes the total load demand. .

[0036] Furthermore, the constructed multi-objective optimization function aims to minimize operating costs, maximize thermal comfort, and enhance system response stability. as follows: ; in, Let be the value of the comprehensive objective function to be minimized; The power allocated to the radiant floor system, The power allocated to the central air conditioning system; First item Let be the system operating cost function, and be an economic indicator. The calculation formula is as follows: ; in, Real-time electricity pricing, which is the real-time time-of-use pricing obtained from the power grid or energy management system, makes the system tend to reduce energy consumption during peak electricity price periods; and These are the instantaneous energy efficiency ratios of the radiant floor system and the central air conditioning system under the current operating conditions, respectively. and These two values ​​are not fixed constants, but are obtained by looking up tables or calculating based on real-time parameters from the device performance curves or performance models stored in the controller. The calculation formula is as follows: ; in, The water temperature for the radiant floor. The return water temperature, The cooling / heating intensity per unit area of ​​radiant floor. The fitting coefficients were obtained by fitting the equipment using the least squares method from the equipment manufacturer's performance data sheet; a physical correction term was also introduced. It is used to correct energy efficiency degradation when the risk of condensation is approaching. This is a correction factor, with a value ranging from 0.8 to 1.0. This is the dew point temperature. The final energy efficiency value is... .

[0037] ; in, The evaporator evaporation temperature. The condenser condensing temperature. For air conditioning air volume, For fitting coefficients; physical correction terms , This is a correction factor, with a value ranging from 0.7 to 1.0. The air conditioning supply temperature, Set the temperature for the air conditioner. Final energy efficiency rating. .

[0038] Second item The comfort evaluation function represents the user's thermal comfort evaluation index. This embodiment uses Fanger's PMV-PPD model, and the calculation formula is as follows: ; in, The predicted average thermal perception vote is calculated based on current and predicted environmental parameters. The target value for comfort is set to 0 for a neutral feeling, but it can also be dynamically set based on user adjustment data. To take into account air temperature Mean radiant temperature relative humidity of air air velocity Clothing quantity of personnel and metabolic rate The comfort index consists of six basic parameters, and the specific calculation steps are as follows: Calculate skin temperature With the surface temperature of clothing Initial value: ; ; Calculate the heat generated by the human body (H) and the heat dissipated through evaporation (E): ; ; Calculate the radiative heat loss R and the convective heat loss C: ; ; in, This is the area coefficient for clothing. ; Let be the convective heat transfer coefficient, when hour, ;when hour, .

[0039] Iterative computation and until thermal equilibrium is achieved. : ; ; in, It facilitates radiative heat exchange between the skin and clothing.

[0040] Calculate the final PMV value: ; in, To reduce heat loss through diffusion and evaporation. The amount of heat dissipated through perspiration is the amount that is felt. This refers to the amount of heat dissipated through respiration and evaporation.

[0041] Third item The system response stability penalty function is used to suppress the adjustment amplitude and frequency of the actuator. Its purpose is to prevent the control system from excessively pursuing instantaneous optimality, leading to overly frequent and drastic actuator actions, thereby causing equipment wear, increased energy consumption, and system oscillation. In this embodiment, the system response stability penalty function is defined as a function related to the change in power distribution command between two consecutive control cycles, as follows: ; in, Here, t represents the penalty coefficient, and t represents the discrete time step.

[0042] This is a set of weighted coefficients that are dynamically adjusted based on the building's operating mode and environmental conditions. These weights are not fixed but can be dynamically adjusted according to the building's operating mode or user settings. For example, during office hours, the comfort weight w2 should be assigned a higher value; while during off-peak hours at night, the economy weight w1 should dominate.

[0043] The multi-objective optimization function is optimized and solved, with constraints including: 1) Energy balance constraint: the sum of the power borne by the two subsystems must be equal to the currently identified total heat and moisture load demand; 2) Latent heat load constraint, the power P allocated to the central air conditioning system ac It must be at least equal to the total latent heat load Q lat .

[0044] 3) Physical operational boundary constraints, and The value must be within the range of the minimum and maximum power that the respective subsystem equipment can provide. The constraint conditions are expressed in the following formula: ; in, For the building's real-time total load demand, These are the upper and lower limits of the physical operating power for radiant floor systems and central air conditioning systems.

[0045] Furthermore, a sequential quadratic programming algorithm is used to solve the multi-objective optimization function in real time, obtaining the optimal power allocation ratio between the radiant floor system and the central air conditioning system. The solution method for the multi-objective optimization function includes the following steps: Step 21, Initialization: Initialize the initial iteration point, Lagrange multipliers, maximum number of iterations, and initial values ​​of the gradient matrix of the constraint function; Set optimization variable vector Initial iteration point Set convergence precision and Maximum number of iterations Lagrange multiplier vectors and initial values ​​of the gradient matrix of the constraint function This step outputs the current iteration point and initial dual / gradient information, providing a baseline for subsequent construction of QP subproblems.

[0046] Step 22: Construct a QP subproblem: Calculate the gradient of the objective function at the current iteration point. With Hessian matrix Calculate the constraint function values ​​for each constraint condition. The gradient is used to construct a QP subproblem by approximating the objective function in the second order and constraining it to the first order linearization, which serves as a solvable local approximate optimization model for this round. At the current iteration point At, for the objective function Taylor expansion to second-order terms, for constraint functions Expanding to first-order terms, construct QP subproblems: ; in: The iteration step size vector, For Hessian matrix, For the objective function in gradient at; For constraint functions in The gradient at that point.

[0047] Step 23, Solution and Update: Solve the QP subproblem using the effective set method to obtain the optimal step size. Then, the step size coefficient is determined through a line search. The iteration points are updated, and the Lagrange multipliers are updated according to the KKT conditions. This will enable the next iteration to more accurately characterize the activity and feasibility of constraints.

[0048] The formula for calculating the new iteration point is: ; in, Set the step size coefficients for the line search; update the Lagrange multipliers. The multiplier values ​​are adjusted based on the KKT conditions to gradually improve the constraint satisfaction. Step 24, Convergence Determination and Output: Calculate the change in the objective function and the constraint violation, and determine whether the convergence condition is met. If convergence is achieved, the optimal solution is obtained; otherwise, proceed to step 22 to execute the next iteration.

[0049] Calculate the change in the objective function and the amount of constraint violation ;like and The iteration converges, and the optimal solution is output. ;like If convergence has not yet occurred, output the current best feasible solution and record a convergence warning message.

[0050] Sequential Quadratic Programming (SQP) is an efficient iterative algorithm for handling nonlinear constrained optimization problems. In each control cycle, the SQP algorithm transforms the nonlinear programming problem into a series of quadratic programming subproblems for solution, ultimately finding the current optimal power allocation solution quickly and reliably. In typical operating scenarios, the algorithm's solution naturally reflects the coordinated strategy of base load and peak load: stable, gradually changing sensible heat loads (base load) are primarily allocated to the more energy-efficient radiant floor system, while all latent heat loads and transient, drastic sensible heat loads caused by personnel activity and changes in solar radiation (peak load) are allocated to the faster-responding central air conditioning system. For example, when a large number of people suddenly flood into a quiet office for a meeting, the load observer will quickly identify a sharp increase in sensible and latent heat loads. At this time, the optimization algorithm will immediately instruct the central air conditioning system to significantly increase its power output to quickly cool and dehumidify, while simultaneously gradually increasing the power command for the radiant floor system, which has high thermal inertia. This avoids thermal discomfort and over-adjustment caused by response lag in traditional control modes.

[0051] In this embodiment, a multi-objective optimization function is adopted, simultaneously considering operating costs, thermal comfort (PMV), and system response stability. Sequential quadratic programming (SQP) is used for optimal power allocation. The operating cost term comprehensively considers electricity price periods, equipment energy efficiency degradation, and maintenance factors. The thermal comfort term uses a dynamic PMV model to provide real-time feedback on indoor thermal environment deviations. The response stability term introduces power change rate constraints to suppress system oscillations. These three factors are weighted and fused to form the objective function, with the weighting coefficients adaptively adjusted based on time period, season, and user preferences.

[0052] Furthermore, based on the optimal power allocation ratio between the radiant floor system and the central air conditioning system obtained from the solution, power allocation is performed based on the logic of spatial vertical gradient zoning, including the following process: Step 201: Divide the building's interior vertical space into the ground-affected zone, the core human activity zone, and the upper stagnation zone according to the sensor layout; Step 202: When the building load is in a steady state or slowly changing condition, prioritize scheduling the radiant floor system to bear the basic load required to maintain the thermal comfort of the core human activity area; Step 203: When the building load experiences transient high-frequency fluctuations or a large amount of latent heat load, the central air conditioning system is prioritized for scheduling, so that it can handle the hot and humid air in the upper stagnation zone and eliminate indoor peak loads.

[0053] A specific judgment scheme, the quantitative judgment method for building load conditions is as follows: The sensible heat load change rate r is calculated based on real-time indoor temperature, relative humidity, and total system heat supply. Q (t), latent heat load percentage η L (t) Two quantitative indicators are used to determine the sensible heat load change rate threshold; the threshold r is set. Q,th A value of 5% / min can be used, with the latent heat load percentage threshold η. L,th A latent heat load of ≥30% can be defined as a large latent heat load. When r Q (t)≤r Q,th And η L (t)<η L,th When the building load is determined to be in a steady state or a slowly changing condition, step 202, the radiant floor priority scheduling strategy, is executed. When r Q (t)>r Q,th or η L (t)≥η L,th If the building load is determined to be experiencing transient high-frequency fluctuations or a large amount of latent heat load, step 203, the central air conditioning priority scheduling strategy, will be executed.

[0054] This allocation strategy also incorporates the concept of zoned control based on indoor vertical temperature gradients, prioritizing the thermal environment quality of the main activity areas of the human body while ensuring overall comfort.

[0055] In some embodiments, the anti-condensation safety control module includes a feedforward-feedback hybrid anti-condensation control strategy comprising a feedback control loop and a feedforward control loop; Feedback control loop is used to monitor the lowest temperature in the floor surface temperature distribution matrix in real time. The highest dew point temperature calculated from the temperature and humidity field of the indoor vertical space When the temperature difference between the two Less than the set dynamic safety margin When this happens, feedback control is used to trigger the collaborative adjustment mechanism; The feedforward control loop predicts the dew point temperature change curve over a future set time period. If the predicted value of the curve is less than the minimum floor surface temperature and the dynamic safety margin ΔT at some future moment... safeWhen the difference is reached, the coordinated adjustment mechanism is activated in advance; wherein, the coordinated adjustment mechanism is to adjust the central air conditioning system and the radiant floor system in order of priority. Specifically, the feedforward control loop uses a long short-term memory network model based on time series analysis to predict the dew point temperature change curve within a preset time window based on historical temperature and humidity data and external meteorological forecast data. A further technical solution involves dynamically adjusting the safety margin for anti-condensation based on the rate of change of dew point, resulting in a dynamic safety margin. The calculation model is as follows: ; in, The basic safety temperature difference constant is set, representing the minimum safe distance of the system under stable operating conditions. Its value can be set according to engineering requirements. This is the adjustment coefficient used to characterize the system response characteristics; The absolute value of the rate of change of dew point temperature is calculated by analyzing the T values ​​over a recent period. dew,max The sequence is obtained by performing first-order difference or sliding window linear regression, which directly reflects the severity of indoor humidity fluctuations. The system's thermal inertia time constant represents the time required for the system to respond to control commands; Furthermore, the system's thermal inertia time constant The parameters were identified through a building thermodynamic model, using a step response parameter identification method based on a first-order building thermodynamic model. The steps are as follows: Step 2001: Perform a step test and collect steady-state indoor temperature / dew point temperature data. Take the average value in the steady segment before the disturbance and take the average value in the new steady segment after the disturbance to obtain steady-state temperature difference data. Specifically, when the building thermal environment is in a stable state, the integrated actuator can apply a step disturbance command to the radiant floor system or central air conditioning system, such as adjusting the radiant floor water supply temperature by ±3℃ or the air conditioning supply temperature by ±2℃. Indoor temperature / dew point temperature data from the start of the disturbance to the new steady state were recorded at 1-minute intervals, and smoothed after outliers were removed. Step 2002: Construct the observations for the first-order model, and fit the data using the first-order model. The formula is: ; Where, ΔT ss The steady-state temperature difference before and after the disturbance. It is the average temperature before the disturbance; Step 2002: Calculate the slope through linear regression and then back-calculate the system's thermal inertia time constant τ. inertia ; τ is calculated using linear regression. inertiaThe required goodness of fit R 2 If the value is ≥0.9, the weighted average of the three identification results is taken; the final result is limited to 5~20 minutes, and if it exceeds this time, the boundary value is taken.

[0056] This dynamic safety margin The physical meaning of the calculation model is that when indoor humidity is stable or changes slowly, Approaching 0, safety margin Close to the base value This allows the floor temperature to be closer to the dew point for maximum cooling capacity; however, when indoor humidity fluctuates drastically, such as during thunderstorms when windows are opened and a large amount of moisture enters, Significantly increased, The temperature automatically increases from 0.5℃ to 2.5℃, allowing more time for the control system to respond and greatly improving the system's safety.

[0057] In the above embodiments of this example, dynamic power distribution and anti-condensation safety control are integrated, which can realize accurate prediction and efficient distribution of building heating and cooling loads. While effectively avoiding the hidden dangers of condensation in radiant cooling, it significantly improves the quality of indoor thermal and humidity environment and the overall operating efficiency of the system.

[0058] In one specific implementation, the Long Short-Term Memory (LSTM) network model used in the feedforward control loop is pre-trained offline on a high-performance computer using a large amount of historical running data. Its input layer receives a feature vector composed of multivariate time series data, including indoor air temperature time series, indoor relative humidity time series, floor surface minimum temperature time series over a preset past period, and outdoor dry-bulb temperature prediction series and outdoor relative humidity prediction series for a preset future period obtained from external meteorological services. The output is a prediction time series representing the minute-by-minute prediction of the indoor maximum dew point temperature within a preset future time window.

[0059] like Figure 5 As shown, the network structure of a Long Short-Term Memory (LSTM) network model typically includes one or more LSTM hidden layers and a fully connected output layer. The output of the LSM model is a minute-by-minute prediction sequence of the highest indoor dew point temperature within a preset time window. For example, for the next 30 minutes, the output would be... The intelligent control unit 200 runs this long short-term memory network model once in each control cycle. If the prediction results show that at some point in the future... , The temperature will rise rapidly and approach the current floor surface temperature, triggering a preset risk threshold, for example... The system will then activate the collaborative adjustment mechanism in advance to achieve preventative risk avoidance.

[0060] Furthermore, the coordinated adjustment mechanism in the aforementioned feedforward and feedback control loops can execute logic using a priority control principle, first scheduling the dehumidification module of the central air conditioning system to reduce the absolute humidity of the indoor air, and then controlling the dehumidification when the dehumidification capacity of the central air conditioning system reaches saturation or the rate of decrease in the indoor air dew point temperature reaches a certain level. When the water temperature falls below a preset threshold, the multi-channel proportional-integral regulating valve of the radiant floor system is then activated to increase the water supply temperature.

[0061] The intelligent control unit centrally schedules various integrated actuators (variable frequency pumps, proportional-integral valves, compressors, fans, etc.) to achieve coordinated linkage of underlying control signals. Specifically, when the real-time temperature difference is monitored... Less than the dynamically calculated When, or when the aforementioned feedforward prediction model issues an early warning, the system will immediately activate a priority-based collaborative adjustment mechanism, with the following execution logic: Level 1 Response: The system first issues a command to maintain the water supply temperature setpoint of the radiant floor system unchanged, but forcibly increases the dehumidification capacity of the central air conditioning system. Specific measures may include: increasing the airflow of the indoor units of the central air conditioning system to increase the circulating airflow, while simultaneously controlling the outdoor unit compressor and electronic expansion valve 340 to work in tandem to actively reduce the temperature of the evaporator coil, thereby achieving deep dehumidification of the air flowing through it, directly reducing the absolute moisture content of the indoor air, and thus lowering the dew point. decline.

[0062] Secondary Response: If the risk of condensation persists after the primary response has been in effect for a period of time, such as 5-10 minutes, or if the rate of dew point rise is too rapid, rendering the primary response insufficient, the system will activate the secondary response. In this case, the intelligent control unit will send a command to the mixing valve 310 of the radiant floor system to moderately increase the water supply temperature, for example, by increasing it by 0.5°C at a time, directly raising the floor surface temperature. This allows the system to quickly move away from the danger zone. Due to the early intervention and effective buffering of the Level 1 response, in most cases the system only needs to fine-tune the water temperature or even avoid initiating the Level 2 response to resolve the crisis, thus maximizing the continuity and stability of the radiant floor cooling system's cooling capacity.

[0063] This embodiment introduces a feedforward prediction mechanism constructed with a long short-term memory network (LSTM) and a feedback monitoring-based anti-condensation control strategy, as well as a dynamic safety margin model, to improve the system's predictive and responsive capabilities. It also introduces a dynamic safety margin calculation model to adjust the safe operating boundary of the radiant floor cooling system in real time based on the indoor dew point temperature change rate and building thermal inertia, thereby achieving more precise and forward-looking anti-condensation control. In a further technical solution, the intelligent control unit also includes a personalized comfort modeling module based on deep reinforcement learning, which is configured as follows: The system continuously records and stores user manual intervention behavior data for indoor environmental control devices. At the moment of manual intervention, it combines synchronous multi-dimensional environmental state data collected by a distributed multi-mode sensor network 100, and uses an inverse reinforcement learning algorithm to analyze the manual intervention behavior data to infer and quantify the implicit boundary conditions characterizing the user's individual thermal comfort preferences. Based on the inferred implicit boundary conditions, the comfort target value PMV of the comfort evaluation function S(Prad,Pac) in the multi-objective optimization function is dynamically adjusted. target This allows for the generation of personalized collaborative control strategies that align with specific user preferences.

[0064] Optional, manual intervention behavior data may include adjustments to temperature setpoints, fan speed levels, and operating modes. This data may include any adjustments made by the user to indoor environmental control parameters through interactive interfaces such as wall-mounted thermostats, mobile apps, or voice assistants, such as raising or lowering temperature setpoints, increasing or decreasing fan speed levels, or switching operating modes.

[0065] A complete, multi-dimensional snapshot of the environmental conditions collected by the distributed multi-mode sensor network 100, including indoor air temperature, humidity, mean radiant temperature, floor surface temperature, CO2 concentration, etc.

[0066] The reverse reinforcement learning (IRL) algorithm is used to analyze manual intervention behavior data to infer and quantify the implicit boundary conditions, i.e., reward functions, that characterize the user's individual thermal comfort preferences. The steps include the following: Step 31: Align user manual intervention behavior data with synchronous environmental state data in time to construct an experience dataset, discretize it, and then perform clustering to divide it into K state clusters; The empirical dataset is represented as: ; in, For the sample size, For the first The state vector of each sample For users in status The following intervention actions; For the state vector Perform normalization and map to Intervals; the continuous state space is discretized using the K-means clustering algorithm and divided into intervals. There are discrete state clusters, each cluster representing a class of similar environmental scenarios; this facilitates subsequent estimation and stable MDP solution and reward back-calculation. Step 32: Assuming user intervention behavior satisfies the stochastic policy model, initialize the stochastic policy. Initialize reward function It is a constant; Step 33, in the current reward function The following uses a value iteration algorithm to solve for the state-value function of a Markov decision process (MDP). Then calculate the corresponding optimal strategy. : ; in, The discount factor has a value of ; The state transition probability is based on the solved state value function. Update strategy: ; Step 34: Back-update the reward based on KL divergence by minimizing the current policy. User experience strategy KL divergence, update reward function :

[0067] in, The learning rate has a value of [value]. ; For users in status Take action below The probability of; Step 35: Calculate the maximum difference between two consecutive reward functions, and use the difference being less than a set value as the convergence condition. Iterate until the convergence condition is met, and output the final reward function. Specifically, the maximum difference between two consecutive reward functions is calculated. ,like Iterative convergence, outputting the final reward function. The reward function is the implicit boundary condition that represents the user's individual thermal comfort preference; otherwise, return to step 33 to continue iterating.

[0068] Through the above process, the reward function It quantitatively expresses the user's preference for different environmental states, such as in state s. =0.8, while in state s′ = A value of 0.5 indicates that the user prefers the environmental state corresponding to s.

[0069] Optionally, this embodiment may employ algorithms such as maximum entropy IRL to robustly infer the implicit comfort preference boundary from diverse, even noisy, user behaviors.

[0070] The individualized reward function inferred by the IRL algorithm will be used to dynamically adjust the comfort evaluation function in the multi-objective optimization function. Specifically, a fixed, universally applicable comfort target value will no longer be used. Instead, it replaces it with a dynamic, personalized target value. .this It can be a function defined by the learned reward function that changes over time, activity state, etc. Based on the calculated implicit boundary conditions, i.e., the reward function The comfort target value of the comfort evaluation function S(Prad,Pac) in the multi-objective optimization function is dynamically adjusted, and the formula is as follows: ; in, The baseline comfort target value; The value of the user preference reward function after training convergence, with a range of [-1, 1], and the partial derivative of the reward function with respect to the PMV value; Signed functions The mean of the reward function, Take 1 at time. The value is -1, used to reinforce the direction of user preferences; ω1, ω2, and ω3 are calibration coefficients.

[0071] For example, the system might learn that a user prefers a slightly cooler environment when working at rest. During lunch breaks, they prefer a neutral or slightly warm environment. In this way, the collaborative control strategy generated by the intelligent control unit will no longer pursue a standardized optimal environment, but will generate a personalized optimal environment that best suits the preferences of a specific user, thereby achieving customized intelligent environment regulation for different users at the technical level.

[0072] This embodiment uses a personalized comfort modeling module based on inverse reinforcement learning. By analyzing the user's manual adjustment behavior, it can dynamically adjust the PMV target value and form an adaptive and personalized control strategy. Based on the personalized comfort modeling technology of inverse reinforcement learning (IRL), the system can continuously learn the user's adjustment behavior of parameters such as temperature and wind speed, infer their individualized comfort boundary, and dynamically adjust and optimize the PMV target value in the control accordingly, so as to achieve truly intelligent regulation based on user preferences.

[0073] This embodiment effectively solves the problems of strong drafts, noise, and uneven temperature distribution that may exist in traditional single air conditioning systems by precisely coordinating and controlling both radiation and convection heat transfer. The system can control the vertical temperature difference in the room to a very small range, creating a uniform, stable, and quiet thermal environment. Simultaneously, by separately processing latent heat and sensible heat loads, it can achieve independent adjustment of temperature and humidity, providing a higher level of comfort. Furthermore, based on a personalized comfort model using inverse reinforcement learning, the system can learn and adapt to the differences in subjective feelings among different users, providing customized environmental control and ensuring that the control strategy always aligns with the user's actual perceived needs, greatly improving user satisfaction.

[0074] In some embodiments, an integrated actuator is established in communication with the intelligent control unit. The integrated actuator includes a variable frequency circulating pump group and a multi-channel proportional-integral regulating valve belonging to the radiant floor system, as well as a variable frequency compressor, an electronic expansion valve 340 and a variable speed fan belonging to the central air conditioning system. The integrated actuator is used to adjust the fluid flow rate, medium temperature and airflow parameters sent by the central air conditioning system in accordance with the coordinated control command. In a feasible implementation, the integrated actuator 300 encompasses all key adjustable moving parts of the radiant floor system and central air conditioning system, serving as the final executor of control commands. For the radiant floor system, its core is the mixing center. This center is equipped with an electrically operated three-way proportional-integral regulating valve 310 controlled by a 0-10V analog signal, used to precisely adjust the mixing ratio of supply and return water, thereby achieving continuous and smooth adjustment of the radiant coil supply water temperature with an accuracy of ±0.5℃. The circulating water system uses a variable frequency circulating pump set 320, whose speed is also controlled by the control signal output by the intelligent control unit 200 to match real-time load demands.

[0075] For central air conditioning systems, adjustable components include the inverter compressor 330 of the outdoor unit, the electronic expansion valve (EEV) 340 of the indoor unit, and the DC brushless variable speed fan 350 of the indoor unit. All of these components directly receive digital commands from the intelligent control unit 200 via a dedicated RS-485 communication interface and proprietary or publicly available communication protocols. This direct digital communication method, compared to traditional on / off or analog control, enables faster and more precise digital coordinated control of refrigerant flow, evaporation temperature, and air volume.

[0076] In the specific regulation process, the underlying control process of the integrated actuator 300 includes: Optimal power command Decomposed into the water supply temperature setpoint of the radiant floor system and circulating water flow setpoint The required valve opening and pump frequency to achieve the desired power are determined through reverse table lookup or model-based calculation; the optimal power command is then applied. Decomposed into the supply air temperature setpoint of the central air conditioning system and air volume set value The dehumidification requirement is calculated by combining the indoor latent heat load. These set values ​​are then used to generate specific messages or analog signals to control the frequency of the variable frequency compressor, the opening degree of the electronic expansion valve 340, and the speed of the variable speed fan through the specific protocol BACnet.

[0077] To illustrate the operational effectiveness of the system in this embodiment, a typical summer workday scenario is used to describe the complete workflow of the system in actual operation.

[0078] During the pre-cooling start-up phase from 7:00 AM to 9:00 AM, the intelligent control unit 200 retrieves the building usage plan for the day from its internal database, such as knowing that 9:00 AM is the start time for work, and combines this with the day's weather forecast data obtained through the network interface to determine that the day will be hot and humid. To take advantage of off-peak electricity prices at night or in the early morning and the lower outdoor ambient temperature, the system starts the pre-cooling program 1-2 hours before the arrival of key personnel. At this time, the economic weights in the multi-objective optimization function... Set to maximum. The system primarily activates the radiant floor system, slowly pre-cooling the concrete floor structure to a low temperature, such as 19°C, turning it into a giant cold storage unit. During this stage, the central air conditioning system is typically in standby or minimum fresh air operation mode.

[0079] During the normal, stable operation phase from 9:00 AM to 12:00 PM, as office workers gradually enter, the heat load on personnel and equipment inside the room gradually increases. At this time, the optimization algorithm will adjust the comfort weighting. Set to maximum. The system allocates most of the steadily increasing sensible heat load to the radiant floor system, which has already stored sufficient cooling capacity. The central air conditioning system operates at lower power, its main tasks being to handle the fresh air load, compensate for the small amount of sensible heat load that the radiant floor system cannot handle, and, crucially, maintain the indoor relative humidity precisely within the user-set comfort range through its dehumidification function. Under these conditions, because most of the cooling load is provided by the energy-efficient radiant floor system, the overall COP of the entire composite system is at an extremely high level.

[0080] During the high-load and drastically fluctuating period from 12:00 to 14:00, outdoor temperatures reach their daily peak, solar radiation is at its strongest, and this is likely the time when people return to the office after lunch, leading to a sharp increase in indoor heat and humidity load. The sensor network detects a rapid rise in indoor CO2 concentration and humidity, causing the dew point temperature to climb accordingly. At this time, the LSTM prediction model may have already warned of potential condensation risks several tens of minutes in advance, and the feedforward control module will instruct the central air conditioning system to increase dehumidification efforts. Simultaneously, the load observer identifies the total load... Significantly exceeded the maximum safe cooling capacity of the radiant floor system. The dynamic power allocation algorithm reacts immediately, distributing all excess peak load to the central air conditioning system, controlling the inverter compressor to rapidly increase its frequency, and increasing the airflow of the indoor units to quickly respond to and eliminate peak loads. Simultaneously, the water supply temperature of the radiant floor system may be slightly adjusted upwards under dynamic safety margin guidance to ensure that condensation does not occur on the floor surface at any time.

[0081] During the nighttime energy-saving mode from 6:00 PM to the following morning, the indoor load decreases rapidly as people leave. Once the system detects the disappearance of human activity, it automatically switches to energy-saving or standby mode. At this time, the system fully utilizes the cold energy stored in the building structure and floor during the day for passive cooling, gradually reducing and eventually shutting down the central air conditioning system. Throughout the night, it may only maintain a low-speed circulation of the radiant floor system to maintain a basic indoor temperature with extremely low energy consumption, thus achieving maximum quietness and energy saving.

[0082] To quantify and verify the beneficial effects of the embodiments of the present invention, a comparative test experiment was conducted for one month in a modern office building with a floor area of ​​approximately 500 square meters. Two areas with identical physical conditions, internal layout, and usage patterns were selected for the experiment. Area A deployed the integrated collaborative control system described in this embodiment, while Area B adopted a traditional independent control method. Specifically, the radiant floor system operated at a constant water supply temperature of 18°C ​​and was configured with simple dew point shutdown protection; the central air conditioning system used an independent thermostat, set to a constant temperature of 26°C.

[0083] The test results clearly demonstrate the superiority of the embodiments of the present invention, and the overall trend can be referred to... Figure 6As shown. In terms of energy efficiency, after a month of continuous operation, the total power consumption of the HVAC system in Area A was reduced by 26.5% compared to Area B. The energy-saving benefits mainly come from two aspects: first, through dynamic power allocation, the ineffective energy consumption of cooling and heating offsetting was basically eliminated; second, through optimized scheduling, the proportion of high-efficiency radiant floor systems in the total operating time was greatly increased, and peak and off-peak electricity prices were effectively utilized. In terms of indoor environmental quality, the PMV index of Area A was consistently maintained within the international standard level 1 comfort zone of -0.3 to +0.3 throughout the entire working period. The indoor vertical temperature difference was successfully controlled within 1.2℃, and users reported no obvious draft, resulting in a high level of comfort. In contrast, when the load in Area B changed abruptly, such as at the start or end of a meeting, the indoor temperature often fluctuated drastically by more than ±2.5℃, accompanied by obvious complaints of localized overcooling and draft. In terms of system safety, a summer thunderstorm occurred during the test, causing a short-term surge in outdoor humidity. The simple control system in Area B failed to respond in time, resulting in significant condensation on its floor surface, triggering an emergency shutdown and causing a sharp deterioration in indoor comfort. In contrast, the system in Area A, with its feedforward prediction module anticipating the humidity risk, initiated a Level 1 response, successfully maintaining the indoor dew point temperature at a stable level. The floor surface remained dry throughout, and the system operated continuously and stably without any downtime.

[0084] In summary, this embodiment, through innovative deep integration design, utilizes advanced dynamic power allocation algorithms, predictive anti-condensation strategies, and personalized adaptive technology to systematically solve the energy efficiency, comfort, and safety issues faced in the coordinated operation of radiant floor and central air conditioning. It demonstrates high technological advancement and practicality, and has broad engineering application value and market promotion prospects.

[0085] The system in this invention enhances its intelligence, robustness, and maintainability. The constructed control system possesses self-learning and adaptive capabilities, proactively addressing system characteristic drift caused by factors such as building aging, climate pattern changes, and alterations in indoor usage. This eliminates the need for frequent manual readjustment and intervention, ensuring long-term operational stability and efficiency. The introduction of load forecasting and equipment condition monitoring functions enables the system not only to perform energy-saving operations but also to provide early warnings of potential equipment failures by analyzing subtle anomalies in equipment operating parameters. This provides data support for predictive maintenance, effectively extending equipment lifespan and improving overall system reliability.

[0086] The system's overall energy efficiency has been significantly improved, and operating costs have been substantially reduced. Through a dynamic power allocation strategy based on multi-objective optimization, the method of this invention ensures that the composite system always operates at or near its optimal operating point for overall energy efficiency under various internal and external conditions. This strategy rationally allocates the load to the subsystem best suited to bear that load, effectively avoiding heat and cold offsetting between systems and unnecessary high-frequency start-ups and shutdowns of equipment, while fully utilizing the high energy efficiency characteristics of the radiant system. Simulation analysis and measured data verification show that, compared to the traditional independent operation control mode of radiant and air conditioning systems, the system using the technical solution of this invention can achieve an overall energy saving rate of 20% to 30%, bringing considerable economic benefits to building operators.

[0087] Example 2 Based on Embodiment 1, this embodiment provides an integrated collaborative control method for radiant floor and central air conditioning, which can be configured to be implemented in the intelligent control unit 200. The method includes a dynamic power allocation method and a feedforward-feedback hybrid anti-condensation control method. The dynamic power allocation method comprises the following steps: Step 1: Obtain multi-dimensional environmental status data of the area to be monitored; Step 2: Construct a building thermodynamic model and use the Kalman filter algorithm to identify the total building load in real time; Step 3: Construct a multi-objective optimization function with the objectives of minimizing operating costs, maximizing thermal comfort, and improving system response stability. ; Step 4: Use the sequential quadratic programming algorithm to solve the multi-objective optimization function in real time to obtain the optimal power allocation ratio between the radiant floor system and the central air conditioning system; Step 5: Based on the optimal power allocation ratio between the radiant floor system and the central air conditioning system obtained from the solution, power allocation is performed based on the logic of spatial vertical gradient partitioning.

[0088] In this embodiment, a real-time load identification mechanism based on a building thermodynamics model and Kalman filtering algorithm is introduced, enabling dynamic power allocation to be based on accurate load estimation and avoiding the bias problems caused by traditional empirical allocation methods. The multi-objective optimization function J integrates operating costs, thermal comfort, and system response stability into the optimization framework, making the power allocation results of the radiant floor system and central air conditioning system more consistent with the optimal overall performance requirements. The application of a sequential quadratic programming algorithm enhances the solution efficiency and stability of the method under complex constraints, enabling real-time updates of power allocation as environmental conditions change. The power allocation method based on spatial vertical gradient zoning effectively matches the adjustment advantages of radiant and convective energy supply at different heights, reducing uneven heating and cooling and improving the overall energy efficiency and indoor thermal environment quality of the system.

[0089] In step 1, the multi-dimensional environmental status data includes the floor surface temperature distribution matrix, indoor vertical space temperature and humidity field, building envelope internal surface temperature, personnel activity density, carbon dioxide concentration, etc. Step 2 involves constructing a building thermodynamic model and using the Kalman filter algorithm to identify the total building load in real time, including the following steps: Step 21: Abstract the building space into a multi-order resistive-capacitive network, and establish a state-space model as the building thermodynamic model. The specific formula of the model is as follows: Equations of state: ; Observation equation: ; in, The state vector includes indoor air node temperature, building envelope node temperature, and floor node temperature. The observation vector is represented by the indoor air temperature, floor surface temperature, etc. collected by the sensor network. The input vector is represented by outdoor temperature, solar radiation intensity, and heat generated by indoor people and equipment. A is the state matrix, B is the input matrix, C is the observation matrix, and D is the direct transmission matrix. and These are process noise and observation noise, respectively.

[0090] Step 22: For the constructed building thermodynamic model, obtain the current environmental state data, and identify the total sensible heat load acting on the building space at the current moment using the Kalman filter algorithm. With total latent heat load It includes the following steps: Step 2021: Initialize the Kalman filter parameters, including the initial estimates of the state vector. Initial error covariance matrix Process noise covariance matrix and observation noise covariance matrix ; Q and R are determined through offline calibration based on sensor accuracy and system disturbance characteristics; Step 2022: Based on the input vector at the current time. and the state estimate of the previous time step Predicting the prior error covariance matrix using the state equation and the prior estimate of the state vector at the current moment. : ; ; in, for Prior estimate of the state vector at time step, The prior error covariance matrix; Step 2023: Obtain the observation vector at the current time. Based on the predicted prior error covariance matrix, the Kalman gain is calculated. : ; Step 2024, based on Kalman gain The prior estimates are corrected to obtain the posterior estimates of the state vector. and posterior error covariance matrix : ; ; Where I is the identity matrix.

[0091] Step 2025: Calculate the observation residuals By integrating the components representing unmodeled thermal disturbances in the residuals and combining them with the building heat balance equation, the total sensible heat load is obtained. and total latent heat load : ; ; ; Where, η sen The percentage of sensible heat load is calculated in real time based on the trends of indoor temperature and humidity changes and the load characteristics of personnel and equipment.

[0092] The total sensible heat load acting on the building space at the current moment in step 22 With total latent heat load The identification method mainly includes setting the initial state estimate, error covariance matrix, and process / observation noise covariance, and calibrating according to sensor accuracy; using Kalman filtering to combine input and observation values ​​to obtain the optimal state estimate; calculating the difference between the observation value and the model prediction value, which is the unmodeled thermal and humidity disturbance; the total heat load is the thermal change of each state node plus the residual correction amount, the latent heat load is calculated through the indoor humidity change rate, and the sensible heat load is equal to the total heat load minus the latent heat load.

[0093] At the start of each dynamic power allocation control cycle, the system initiates an online heat load observer based on a Kalman filter algorithm. This observer embeds a simplified building thermodynamics model, such as a second- or third-order RC network model describing heat exchange between the building space and the building envelope. The model utilizes indoor air temperature, floor surface temperature, and inner surface temperature of the building envelope from a distributed multi-mode sensor network 100 as state observations, while taking outdoor temperature, solar radiation intensity, and estimated heat dissipation from people and equipment based on CO2 concentration and activity levels as inputs. Through the recursive prediction and correction process of the Kalman filter, the system not only obtains smooth optimal estimates of state variables such as indoor temperature, but more importantly, it attributes the residuals between the observed values ​​and the model predictions to unmodeled disturbances, thereby accurately identifying the total sensible heat load acting on the building space in real time. With total latent heat load The sum of these two load components constitutes the total load demand. .

[0094] In step 3, the constructed multi-objective optimization function aims to minimize operating cost, maximize thermal comfort, and optimize system response stability. as follows: ; in, Let be the value of the comprehensive objective function to be minimized; The power allocated to the radiant floor system, The power allocated to the central air conditioning system; First item Let be the system operating cost function, and be an economic indicator. The calculation formula is as follows: ; in, Real-time electricity pricing, which is the real-time time-of-use pricing obtained from the power grid or energy management system, makes the system tend to reduce energy consumption during peak electricity price periods; and These are the instantaneous energy efficiency ratios of the radiant floor system and the central air conditioning system under the current operating conditions, respectively. and These two values ​​are not fixed constants, but are obtained by looking up tables or calculating based on real-time parameters from the device performance curves or performance models stored in the controller. The calculation formula is as follows: ; in, The water temperature for the radiant floor. The return water temperature, The cooling / heating intensity per unit area of ​​radiant floor. The fitting coefficients were obtained by fitting the equipment using the least squares method from the equipment manufacturer's performance data sheet; a physical correction term was also introduced. It is used to correct energy efficiency degradation when the risk of condensation is approaching. This is a correction factor, with a value ranging from 0.8 to 1.0. This is the dew point temperature. The final energy efficiency value is... .

[0095] ; in, The evaporator evaporation temperature. The condenser condensing temperature. For air conditioning air volume, For fitting coefficients; physical correction terms , This is a correction factor, with a value ranging from 0.7 to 1.0. The air conditioning supply temperature, Set the temperature for the air conditioner. Final energy efficiency rating. .

[0096] Second item The comfort evaluation function represents the user's thermal comfort evaluation index. This embodiment uses Fanger's PMV-PPD model, and the calculation formula is as follows: ; in, The predicted average thermal perception vote is calculated based on current and predicted environmental parameters. The target value for comfort is set to 0 for a neutral feeling, but it can also be dynamically set based on user adjustment data. To take into account air temperature Mean radiant temperature relative humidity of air air velocity Clothing quantity of personnel and metabolic rate The comfort index consists of six basic parameters, and the specific calculation steps are as follows: Calculate skin temperature With the surface temperature of clothing Initial value: ; ; Calculate the heat generated by the human body (H) and the heat dissipated through evaporation (E): ; ; Calculate the radiative heat loss R and the convective heat loss C: ; ; in, This is the area coefficient for clothing. ; Let be the convective heat transfer coefficient, when hour, ,when hour, .

[0097] Iterative computation and until thermal equilibrium is achieved. : ; ; in, It facilitates radiative heat exchange between the skin and clothing.

[0098] Calculate the final PMV value: ; in, To reduce heat loss through diffusion and evaporation. The amount of heat dissipated through perspiration is the amount that is felt. This refers to the amount of heat dissipated through respiration and evaporation.

[0099] Third item The system response stability penalty function is used to suppress the adjustment amplitude and frequency of the actuator. Its purpose is to prevent the control system from excessively pursuing instantaneous optimality, leading to overly frequent and drastic actuator actions, thereby causing equipment wear, increased energy consumption, and system oscillation. In this embodiment, the system response stability penalty function is defined as a function related to the change in power distribution command between two consecutive control cycles, as follows: ; in, Here, t represents the penalty coefficient, and t represents the discrete time step.

[0100] This is a set of weighted coefficients that are dynamically adjusted based on the building's operating mode and environmental conditions. These weights are not fixed but can be dynamically adjusted according to the building's operating mode or user settings. For example, during office hours, the comfort weight w2 should be assigned a higher value; while during off-peak hours at night, the economy weight w1 should dominate.

[0101] The multi-objective optimization function is optimized and solved, with constraints including: 1) Energy balance constraint: the sum of the power borne by the two subsystems, the radiant floor system and the central air conditioning system, must be equal to the currently identified total heat and humidity load demand; 2) Latent heat load constraint, the power P allocated to the central air conditioning system ac It must be at least equal to the total latent heat load Q lat .

[0102] 3) Physical operational boundary constraints, and The value must be within the range of the minimum and maximum power that the respective subsystem equipment can provide. The constraint conditions are expressed in the following formula: ; in, For the building's real-time total load demand, These are the upper and lower limits of the physical operating power for radiant floor systems and central air conditioning systems.

[0103] In step 4, a sequential quadratic programming algorithm is used to solve the multi-objective optimization function in real time to obtain the optimal power allocation ratio between the radiant floor system and the central air conditioning system. The solution method for the multi-objective optimization function includes the following steps: Step 41, Initialization: Initialize the initial iteration point, Lagrange multipliers, maximum number of iterations, and initial values ​​of the constraint function gradient matrix; Set optimization variable vector Initial iteration point Set convergence precision and Maximum number of iterations Lagrange multiplier vectors and initial values ​​of the gradient matrix of the constraint function This step outputs the current iteration point and initial dual / gradient information, providing a baseline for subsequent construction of QP subproblems.

[0104] Step 42: Construct a QP subproblem: Calculate the gradient of the objective function at the current iteration point. With Hessian matrix Calculate the constraint function values ​​for each constraint condition. The gradient is used to construct a QP subproblem by approximating the objective function in the second order and constraining it to the first order linearization, which serves as a solvable local approximate optimization model for this round. At the current iteration point At, for the objective function Taylor expansion to second-order terms, for constraint functions Expanding to first-order terms, construct QP subproblems: ; in: The iteration step size vector, For Hessian matrix, For the objective function in gradient at; For constraint functions in The gradient at that point.

[0105] Step 43, Solution and Update: Solve the QP subproblem using the effective set method to obtain the optimal step size. Then, the step size coefficient is determined through a line search. The iteration points are updated, and the Lagrange multipliers are updated according to the KKT conditions. This will enable the next iteration to more accurately characterize the activity and feasibility of constraints.

[0106] The formula for calculating the new iteration point is: ; in, Set the step size coefficients for the line search; update the Lagrange multipliers. The multiplier values ​​are adjusted based on the KKT conditions to gradually improve the constraint satisfaction. Step 44, Convergence Determination and Output: Calculate the change in the objective function and the constraint violation, and determine whether the convergence condition is met. If convergence is achieved, the optimal solution is obtained; otherwise, proceed to step 42 to execute the next iteration.

[0107] Calculate the change in the objective function and the amount of constraint violation ;like and The iteration converges, and the optimal solution is output. ;like If convergence has not yet occurred, output the current best feasible solution and record a convergence warning message.

[0108] Sequential Quadratic Programming (SQP) is an efficient iterative algorithm for handling nonlinear constrained optimization problems. In each control cycle, the SQP algorithm transforms the nonlinear programming problem into a series of quadratic programming subproblems for solution, ultimately finding the current optimal power allocation solution (Prad) quickly and reliably. Pac In typical operating scenarios, the algorithm's solution naturally reflects the coordinated strategy of base load and peak load: stable, gradually changing sensible heat loads (base load) are primarily allocated to the more energy-efficient radiant floor system, while all latent heat loads and transient, drastic sensible heat loads caused by personnel activity and changes in solar radiation (peak load) are allocated to the faster-responding central air conditioning system. For example, when a large number of people suddenly flood into a quiet office for a meeting, the load monitor will quickly identify a sharp increase in sensible and latent heat loads. At this time, the optimization algorithm will immediately instruct the central air conditioning system to significantly increase its power output to quickly cool and dehumidify, while simultaneously gradually increasing the power command for the radiant floor system, which has high thermal inertia. This avoids thermal discomfort and over-adjustment caused by response lag in traditional control modes.

[0109] In step 5, based on the optimal power allocation ratio between the radiant floor system and the central air conditioning system obtained from the solution, power allocation is performed based on the logic of spatial vertical gradient zoning, including the following process: Step 501: Divide the building's interior vertical space into the ground-affected zone, the core human activity zone, and the upper stagnation zone according to the sensor layout; Step 502: When the building load is in a steady state or slowly changing condition, prioritize scheduling the radiant floor system to bear the basic load required to maintain the thermal comfort of the core human activity area; Step 503: When the building load experiences transient high-frequency fluctuations or there is a large amount of latent heat load, the central air conditioning system is prioritized for scheduling, so that it can handle the hot and humid air in the upper stagnation zone and eliminate indoor peak loads.

[0110] A specific judgment scheme, the quantitative judgment method for building load conditions is as follows: The sensible heat load change rate r is calculated based on real-time indoor dry-bulb temperature, relative humidity, and total system heat supply. Q (t), latent heat load percentage η L (t) Two quantitative indicators are used to determine the sensible heat load change rate threshold; the threshold r is set. Q,th Take 5% / min, latent heat load percentage threshold η L,th Take 30%; a latent heat load with a latent heat load ratio of ≥30% is defined as a large latent heat load. When r Q (t)≤5% / min and η L When (t) < 30%, it is determined to be a steady-state or slowly changing building load condition, and the radiant floor priority scheduling strategy in step 502 is executed. When r Q (t) > 5% / min or η LWhen (t)≥30%, it is determined that the building load is experiencing transient high-frequency fluctuations or a large amount of latent heat load, and the central air conditioning priority scheduling strategy in step 503 is executed.

[0111] Furthermore, the feedforward-feedback hybrid anti-condensation control method includes a feedback control loop and a feedforward control loop; Feedback control loop is used to monitor the lowest temperature in the floor surface temperature distribution matrix in real time. The highest dew point temperature calculated from the temperature and humidity field of the indoor vertical space When the temperature difference between the two Less than the set dynamic safety margin When this happens, feedback control is used to trigger the collaborative adjustment mechanism; The feedforward control loop predicts the dew point temperature change curve within a set future time period. When the predicted value of the curve is less than the difference between the lowest floor surface temperature and the preset risk threshold at a certain future moment, the coordinated adjustment mechanism is activated in advance. Based on the dynamic adjustment of the anti-condensation safety margin according to the dew point change rate, dynamic safety margin The calculation model is as follows: ; in, The basic safety temperature difference constant is set, representing the minimum safe distance of the system under stable operating conditions. Its value can be set according to engineering requirements. This is the adjustment coefficient used to characterize the system response characteristics; The absolute value of the rate of change of dew point temperature is calculated by analyzing the T values ​​over a recent period. dew,max The sequence is obtained by performing first-order difference or sliding window linear regression, which directly reflects the severity of indoor humidity fluctuations. The system's thermal inertia time constant represents the time required for the system to respond to control commands; Furthermore, the system's thermal inertia time constant The parameters were identified through a building thermodynamic model, using a step response parameter identification method based on a first-order building thermodynamic model. The steps are as follows: Step 2001: Perform a step test and collect steady-state indoor temperature / dew point temperature data. Take the average value in the steady segment before the disturbance and take the average value in the new steady segment after the disturbance to obtain steady-state temperature difference data. Specifically, when the building thermal environment is in a stable state, the integrated actuator can apply a step disturbance command to the radiant floor system or central air conditioning system, such as adjusting the radiant floor water supply temperature by ±3℃ or the air conditioning supply temperature by ±2℃. Indoor temperature / dew point temperature data from the start of the disturbance to the new steady state were recorded at 1-minute intervals, and smoothed after outliers were removed. Step 2002: Construct the observations for the first-order model, and fit the data using the first-order model. The formula is: ; Where, ΔT ss The steady-state temperature difference before and after the disturbance. It is the average temperature before the disturbance; Step 2002: Calculate the slope through linear regression and then back-calculate the system's thermal inertia time constant τ. inertia ; τ is calculated using linear regression. inertia The required goodness of fit R 2 If the value is ≥0.9, the weighted average of the three identification results is taken; the final result is limited to 5~20 minutes, and if it exceeds this time, the boundary value is taken.

[0112] This dynamic safety margin The physical meaning of the calculation model is that when indoor humidity is stable or changes slowly, Approaching 0, safety margin Close to the base value This allows the floor temperature to be closer to the dew point for maximum cooling capacity; however, when indoor humidity fluctuates drastically, such as during thunderstorms when windows are opened and a large amount of moisture enters, Significantly increased, The temperature automatically increases, dynamically from 0.5℃ to 2.5℃, thus allowing more time for the control system to respond and greatly improving system safety. In one specific implementation, the feedforward control loop uses a Long Short-Term Memory (LSTM) network model, which is pre-trained offline on a high-performance computer using a large amount of historical operating data. Its input layer receives a feature vector composed of multivariate time series data, including indoor air temperature time series, indoor relative humidity time series, floor surface minimum temperature time series over a preset past period, and outdoor dry-bulb temperature prediction series and outdoor relative humidity prediction series for a preset future period obtained from external meteorological services. The output is a prediction time series representing the minute-by-minute prediction of the indoor maximum dew point temperature within a preset future time window.

[0113] like Figure 5 As shown, the network structure of a Long Short-Term Memory (LSTM) network model typically includes one or more LSTM hidden layers and a fully connected output layer. The output of the LSM model is a minute-by-minute prediction sequence of the highest indoor dew point temperature within a preset time window. For example, for the next 30 minutes, the output would be... The intelligent control unit 200 runs this long short-term memory network model once in each control cycle. If the prediction results show that at some point in the future... , The temperature will rise rapidly and approach the current floor surface temperature, triggering a preset risk threshold, for example... The system will then activate the collaborative adjustment mechanism in advance to achieve preventative risk avoidance.

[0114] Furthermore, the coordinated adjustment mechanism in the aforementioned feedforward and feedback control loops can execute logic using a priority control principle, first scheduling the dehumidification module of the central air conditioning system to reduce the absolute humidity of the indoor air, and then controlling the dehumidification when the dehumidification capacity of the central air conditioning system reaches saturation or the rate of decrease in the indoor air dew point temperature reaches a certain level. When the water temperature falls below a preset threshold, the multi-channel proportional-integral regulating valve of the radiant floor system is then activated to increase the water supply temperature.

[0115] The condensation control method in this embodiment fundamentally eliminates the risk of condensation in radiant cooling and significantly improves the system's cooling capacity. The proposed feedforward-feedback hybrid anti-condensation control strategy, particularly the combination of LSTM trend prediction and a dynamic safety margin model, enables the system to anticipate and proactively respond to humidity changes. This mechanism allows the radiant floor surface temperature to operate safely at a critical state that is closer to but always above the dew point temperature, while ensuring absolute safety. This not only completely avoids condensation accidents but, more importantly, breaks through the limitations of traditional conservative water temperature settings on the cooling capacity of radiant systems. Therefore, it can fully utilize the efficiency of radiant cooling even in hot and humid climates, improving the overall cooling capacity and applicability of the system.

[0116] Furthermore, it also includes a personalized comfort modeling method based on deep reinforcement learning, comprising the following steps: Step S1: Continuously record and store user manual intervention data on indoor environmental control devices; Step S2: For the moment when the manual intervention behavior occurs, based on the synchronously acquired multidimensional environmental state data, the inverse reinforcement learning algorithm is used to analyze the manual intervention behavior data in order to infer and quantify the implicit boundary conditions that characterize the user's individual thermal comfort preference. Step S3: Based on the inferred implicit boundary conditions, dynamically modify the comfort target value of the comfort evaluation function S(Prad,Pac) in the multi-objective optimization function.

[0117] In step S2, the method of analyzing the manual intervention behavior data using the inverse reinforcement learning (IRL) algorithm to infer and quantify the implicit boundary conditions, i.e. the reward function, that characterizes the user's individual thermal comfort preference includes the following: Step S21: Align the user's manual intervention behavior data with the synchronous environmental state data in time, construct an experience dataset, discretize it, and then perform clustering to divide it into K state clusters; Environmental condition data includes radiant floor power P rad Central air conditioning power P ac Including indoor temperature and humidity and other related parameters, the IRL algorithm empirical dataset is represented as follows: ; in, For the sample size, For the first The state vector of each sample For users in status The following intervention actions; For the state vector Perform normalization and map to Intervals; the continuous state space is discretized using the K-means clustering algorithm and divided into intervals. There are discrete state clusters, each cluster representing a class of similar environmental scenarios; this facilitates subsequent estimation and stable MDP solution and reward back-calculation. Step S22: Assuming user intervention behavior satisfies the stochastic policy model, initialize the stochastic policy. Initialize reward function It is a constant; Step S23: In the current reward function The following uses a value iteration algorithm to solve for the state-value function of a Markov decision process (MDP). Then calculate the corresponding optimal strategy. : ; in, The discount factor has a value of ; The state transition probability is based on the solved state value function. Update strategy: ; Step S24: Reward back-engineering update based on KL divergence, by minimizing the current policy. User experience strategy KL divergence, update reward function :

[0118] in, The learning rate has a value of [value]. ; For users in status Take action below The probability of; Step S25: Calculate the maximum difference between two adjacent reward functions, take the difference being less than a set value as the convergence condition, iterate until the convergence condition is met, and output the final reward function. Specifically, the maximum difference between two consecutive reward functions is calculated. ,like Iterative convergence, outputting the final reward function. The reward function is the implicit boundary condition that characterizes the user's individual thermal comfort preference; otherwise, return to step S23 to continue iterating.

[0119] Through the above process, the reward function It quantitatively expresses the user's preference for different environmental states, such as in state s. =0.8, while in state s′ = A value of 0.5 indicates that the user prefers the environmental state corresponding to s.

[0120] In step S3, the reward function is calculated based on the implicit boundary conditions. The comfort target value of the comfort evaluation function S(Prad,Pac) in the multi-objective optimization function is dynamically adjusted, and the formula is as follows: ; in, The baseline comfort target value; The value of the user preference reward function after training convergence, with a range of [-1, 1], and the partial derivative of the reward function with respect to the PMV value; Signed functions The mean of the reward function, Take 1 at time. The value is -1, used to reinforce the direction of user preferences; ω1, ω2, and ω3 are calibration coefficients.

[0121] This embodiment uses a personalized comfort modeling module based on inverse reinforcement learning. By analyzing the user's manual adjustment behavior, it can dynamically adjust the PMV target value and form an adaptive and personalized control strategy. Based on the personalized comfort modeling technology of inverse reinforcement learning (IRL), the system can continuously learn the user's adjustment behavior of parameters such as temperature and wind speed, infer their individualized comfort boundary, and dynamically adjust and optimize the PMV target value in the control accordingly, so as to achieve truly intelligent regulation based on user preferences.

[0122] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

[0123] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for integrated coordinated control of radiant floor and central air conditioning, characterized in that, This includes dynamic power allocation methods and a feedforward-feedback hybrid anti-condensation control method. The dynamic power allocation method has the following steps: Acquire multi-dimensional environmental status data of the area to be monitored; A building thermodynamics model was constructed, and the total building load was identified in real time using the Kalman filter algorithm. To minimize operating costs, maximize thermal comfort, and improve system response stability, a multi-objective optimization function is constructed. ; The sequential quadratic programming algorithm is used to solve the multi-objective optimization function in real time, and the optimal power allocation ratio between the radiant floor system and the central air conditioning system is obtained. Based on the optimal power allocation ratio between the radiant floor system and the central air conditioning system obtained by solving the problem, power allocation is performed based on the logic of spatial vertical gradient partitioning.

2. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, Constructing a building thermodynamic model and using the Kalman filter algorithm to identify the total building load in real time includes the following steps: The building space is abstracted into a multi-order resistance-capacitance network, and a state-space model is established as the building thermodynamic model. For the constructed building thermodynamic model, current environmental state data is obtained, and the total sensible heat load acting on the building space at the current moment is identified using the Kalman filter algorithm. With total latent heat load .

3. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, The constructed multi-objective optimization function aims to minimize operating costs, maximize thermal comfort, and optimize system response stability. as follows: ； in, Let be the value of the comprehensive objective function to be minimized; The power allocated to the radiant floor system, The power allocated to the central air conditioning system, It is a set of weighting coefficients that are dynamically adjusted according to the building's operating mode and environmental conditions; First item The system operating cost function is calculated using the following formula: ； in, For real-time electricity prices; and These are the instantaneous energy efficiency ratios of the radiant floor system and the central air conditioning system under the current operating conditions, respectively. Second item The comfort evaluation function is calculated using the following formula: ； in, The predicted average thermal perception vote is calculated based on current and predicted environmental parameters. The target value for comfort; Third item The system response stability penalty function is given by the following formula: ； in, Here, t represents the penalty coefficient, and t represents the discrete time step.

4. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, The multi-objective optimization function is optimized and solved, with constraints including: Energy balance constraints dictate that the sum of the power supplied by the radiant floor system and the central air conditioning system must equal the currently identified total heat and humidity load demand. Latent heat load constraint, the power P allocated to the central air conditioning system ac It must be at least equal to the total latent heat load Q lat ; Physical operational boundary constraints, and The value must be within the range of minimum and maximum power that the respective subsystem devices can provide.

5. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, A sequential quadratic programming algorithm is used to solve the multi-objective optimization function in real time, obtaining the optimal power allocation ratio between the radiant floor system and the central air conditioning system. The solution method for the multi-objective optimization function includes the following steps: Step 41: Initialize the initial iteration point, Lagrange multipliers, maximum number of iterations, and initial values ​​of the constraint function gradient matrix; Step 42: Calculate the gradient of the objective function and the Hessian matrix at the current iteration point, calculate the constraint function values ​​and gradients in each constraint condition, and construct the second-order approximation of the objective function and the first-order linearization of the constraints into a QP subproblem, which serves as a solvable local approximate optimization model for this round. Step 43: Solve the QP subproblem using the effective set method to obtain the optimal step size. Then, the step size coefficient is determined through a line search. The iteration points are updated, and the Lagrange multipliers are updated according to the KKT conditions. ; Step 44: Calculate the change in the objective function and the amount of constraint violation, and determine whether the convergence condition is met. If convergence is achieved, the optimal solution is obtained; otherwise, proceed to step 42 to execute the next iteration.

6. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, It also includes a personalized comfort modeling method based on deep reinforcement learning, which includes the following steps: Continuously record and store user manual intervention data on indoor environmental control devices; For the moment when manual intervention occurs, based on the synchronously acquired multidimensional environmental state data, the inverse reinforcement learning algorithm is used to analyze the manual intervention data in order to infer and quantify the implicit boundary conditions that characterize the user's individual thermal comfort preference. Based on the inferred implicit boundary conditions, the comfort target value of the comfort evaluation function S(Prad,Pac) in the multi-objective optimization function is dynamically modified.

7. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 1, characterized in that, A feedforward-feedback hybrid anti-condensation control method, comprising a feedback control loop and a feedforward control loop; Feedback control loop is used to monitor the lowest temperature in the floor surface temperature distribution matrix in real time. The highest dew point temperature calculated from the temperature and humidity field of the indoor vertical space When the temperature difference between the two Less than the set dynamic safety margin When this happens, feedback control is used to trigger the collaborative adjustment mechanism; The feedforward control loop predicts the dew point temperature change curve within a set future time period. When the predicted value of the curve is less than the difference between the lowest floor surface temperature and the preset risk threshold at a certain future moment, the coordinated adjustment mechanism is activated in advance.

8. The integrated collaborative control method based on radiant floor and central air conditioning as described in claim 7, characterized in that, Based on the dynamic adjustment of the anti-condensation safety margin according to the dew point change rate, dynamic safety margin The calculation model is as follows: ； in, This is the established basic safety temperature difference constant; This is the adjustment coefficient used to characterize the system's response properties; The absolute value of the rate of change of dew point temperature is calculated by analyzing the T values ​​over a recent period. dew,max The sequence is obtained by performing first-order differencing or sliding window linear regression. is the thermal inertia time constant, which characterizes the time required for the system to respond to control commands.

9. A radiant floor-central air conditioning integrated control system, characterized in that, include: A distributed multi-mode sensor network is deployed at multiple vertical gradient levels in the building space to collect multi-dimensional environmental status data of the building space. The intelligent control unit establishes a communication connection with the distributed multi-mode sensor network and is configured to execute the integrated collaborative control method based on radiant floor-central air conditioning as described in any one of claims 1-7. An integrated actuator is used to execute control commands to regulate the coordinated operation of the radiant floor system and the central air conditioning system.

10. The integrated collaborative control system based on radiant floor and central air conditioning as described in claim 9, characterized in that: The intelligent control unit includes an identification module, a dynamic power distribution module, and an anti-condensation safety control module; The identification module is used to identify the total heat and moisture load of a building based on the constructed building thermodynamic model; The dynamic power allocation module is configured to execute a dynamic power allocation strategy, which constructs and solves a multi-objective optimization function to dynamically decompose the total heat and humidity load into a base load power component borne by the radiant floor system and a peak-shaving power component borne by the central air conditioning system. The anti-condensation safety control module is configured to dynamically adjust the anti-condensation safety margin based on the rate of change of dew point, and to perform anti-condensation regulation by combining a feedforward-feedback hybrid anti-condensation control strategy that combines long short-term memory network dew point trend prediction with real-time monitoring.

Images