Building central air conditioning system comprehensive control method and device

Abstract

The application discloses a kind of building central air conditioning system integrated control method and device, comprising: according to the system state vector of construction, the load and environment prediction of future time domain are carried out, and predicted value is obtained;Using global optimization control strategy, through model predictive control module, global energy consumption optimal basic set value meeting constraint is solved according to predicted value;Based on the control error and its change rate between system desired set value and actual measured output value, the correction amount of global energy consumption optimal basic set value is calculated;Global energy consumption optimal basic set value and correction amount are fused, through physical hard constraint and change rate limit, instruction is fused, limiting and interlocking, and final execution instruction is generated and issued to execution equipment;Execution equipment responds to execution instruction and executes set value, while obtaining actual operation data, and calculates reward value and stores in experience pool for model iteration.The present application can realize intelligent control building central air conditioner.

Description

Technical Field

[0001] This invention relates to the field of central air conditioning system optimization control technology, and in particular to a comprehensive control method and device for a building central air conditioning system. Background Technology

[0002] China is a major energy consumer, with building energy consumption accounting for as much as 25% of total social energy consumption, and this proportion is still rising year by year. Air conditioning systems consistently account for a large portion of building energy consumption. As people's demands for building comfort and energy conservation increase, how to control the terminal equipment of air conditioning systems to achieve maximum energy savings while meeting people's comfort requirements is receiving increasing attention.

[0003] Central air conditioning systems, commonly used in large and medium-sized buildings, are responsible for air conditioning in designated areas, primarily providing cooling, heating, humidification, dehumidification, and air purification. In modern buildings, central air conditioning systems have become a crucial component in ensuring indoor environmental comfort. However, the energy consumption of central air conditioning systems is considerable, especially in large public buildings where more than half of the electricity is used for air conditioning. Therefore, optimizing the control of central air conditioning systems and improving energy efficiency are of great significance for energy conservation and emission reduction.

[0004] Currently, central air conditioning systems are widely used in various buildings, and their energy consumption accounts for a very high proportion of the overall building energy consumption, generally 40%-50%. Therefore, energy conservation and consumption reduction are receiving increasing attention. Furthermore, central air conditioning systems involve numerous devices, each with different efficiency models. The input and output variables for energy-saving optimization control are numerous, exhibiting nonlinear, time-varying, and coupled characteristics. This makes it difficult for each device in a central air conditioning system to operate under relatively optimal conditions. For those skilled in the art, how to enable each device in a central air conditioning system to operate under relatively optimal conditions, thereby reducing the energy consumption of the central air conditioning system, has become an important research topic. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings and defects of the prior art and provide a comprehensive control method and device for building central air conditioning systems based on a hybrid model, aiming to solve the problems of control lag, low global energy efficiency, poor model adaptability and insufficient AI control security in the prior art.

[0006] One aspect of the present invention provides a comprehensive control method for a building central air conditioning system, comprising the steps of: S1. Based on the constructed system state vector, predict the load and environment in the future time domain to obtain the predicted values; S2. A global optimization control strategy is adopted. The model prediction control module solves for the global energy consumption optimal basic setpoint that satisfies the constraints based on the predicted value. S3. Based on the control error and its rate of change between the system's expected setpoint and the actual measured output value, calculate the correction amount for the global energy consumption optimal base setpoint; S4. The global energy consumption optimal base setting value and correction amount are integrated, and the instructions are integrated, limited and interlocked through physical hard constraints and change rate limits to generate the final execution instructions and send them to the execution device; S5. Execute the device response execution command to execute the set value, and at the same time obtain the actual operation data, calculate the reward value and store it in the experience pool for model iteration.

[0007] Preferably, the constructed system state vector includes indoor temperature, indoor relative humidity, real-time cooling load, indoor occupant status, time encoding, and date type.

[0008] Preferably, when making future time-domain load and environmental predictions, a time-series prediction model is used, and predictions are made based on the constructed system state vector and the obtained meteorological data.

[0009] Preferably, the model predictive control module is based on a state-space model, and solves for the optimal control sequence that satisfies the constraints in a finite time domain based on the predicted values, to obtain the global energy consumption optimal basic setpoint that satisfies the constraints, including: Establish a discretized spatial state model: ; This represents state variables, including indoor temperature and relative humidity. This indicates the controlled variables, including chilled water temperature, supply air temperature, and differential pressure. This represents the disturbance variables, including load forecasts and outdoor enthalpy. A, B, E These represent the corresponding state transition matrix, control input matrix, and disturbance input matrix, respectively. Construct the objective function J Minimize the deviation between energy consumption and comfort: ; in, Q, R These are the comfort weight matrix and the energy consumption weight matrix, respectively. Indicates the reference state variable. This indicates the prediction period, and the superscript T indicates transpose. Indicates the total power of the air conditioning system; Set constraints; A quadratic programming (QP) problem is constructed based on the multivariate coupling problem. A solver is used to find the optimal control sequence, and the optimal global energy consumption setpoints are obtained, including: ; in, This represents the baseline setting for optimal global energy consumption. This indicates the optimal baseline setting for the cold water temperature. This indicates the optimal basic setting value for the water pump frequency. This indicates the optimal basic setting value for the cooling tower fan frequency. This indicates the optimal basic setting value for valve opening.

[0010] Preferably, the constraints include state variable constraints, control variable constraints, control variable rate of change constraints, and dehumidification constraints, which ensure that the state variables, control variables, control variable rates of change, and cold water temperature are within a preset threshold range.

[0011] Preferably, in step S3, calculating the correction amount for the global energy consumption optimal baseline setting includes: The model predictive control module utilizes the Adaptive Neural Fuzzy Inference System (ANFIS) to learn the residual between the state-space model and the actual system. Using control error and error change rate as input, it constructs a multi-layer Sugeno-type fuzzy neural network. Through a hybrid learning algorithm, it optimizes the network parameters and outputs the correction amount for the model predictive control quantity, i.e., the correction amount for the global energy consumption optimal baseline setpoint, including: ; in, Indicates the correction amount. Indicates control error. Indicates the rate of change of error. To normalize the rule strength, These are the parameters for the conclusion.

[0012] Preferably, the optimal global energy consumption baseline setting value and the correction amount are combined to obtain a combined value, including: ; in, Indicates the fusion value. This represents the optimal baseline setting for local energy consumption. Indicates the correction amount. The confidence coefficient is initially set to 0.2 and gradually increases to 1.0 as the system becomes more stable. Prior to obtaining the fusion value, the following steps are also included: Perform hard constraint testing, including at least dehumidification constraints, to ensure the dehumidification capacity and dew point temperature limits of the surface cooler, and obtain the final chilled water temperature setting. The rate of change is limited and combined with the historical value from the previous moment to obtain the final set values ​​for the pump frequency and valve opening. Output safety setting value The data is sent down to the underlying controller to collect actual operational data feedback and calculate reward signals. Stored in the experience pool for model iteration; Furthermore, output safety settings. Prior to this, the process also includes a water-air flow matching step: Calculate the total air volume requirement based on the opening distribution of the terminal VAV valves. If the supply air temperature increases, leading to an increase in the total air volume demand, then adjust the total air volume demand accordingly. and pipeline resistance coefficient S Calculate the required pressure difference Check whether the pump head and flow rate meet the hydraulic operating conditions; if the pump capacity is insufficient, limit the rise of the supply air temperature to prevent hydraulic imbalance; otherwise, dynamically adjust the cold water circulation flow rate according to the load prediction sequence to form a pump set operating frequency control command.

[0013] Preferably, when the executing device responds to the execution command to execute the set value, it uses PID control based on adaptive fuzzy PID control technology, including: ; in, This represents the proportional gain at time t. This represents the initial proportional gain. β This indicates the ANFIS corrected weighting coefficient. This indicates the PID gain correction output.

[0014] In another aspect, the present invention provides a building central air conditioning integrated control device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the building central air conditioning system integrated control method.

[0015] The integrated control method for building central air conditioning systems of the present invention solves the problems of control lag, low global energy efficiency, poor model adaptability and insufficient AI control security in the prior art, realizes adaptive intelligent control of building central air conditioning, and meets the intelligent control needs of building central air conditioning. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the integrated control method for a building central air conditioning system according to the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] In this application, the integrated control method for building central air conditioning system adopts the steps of "prediction, optimization, correction and execution" to achieve intelligent and safe operation control of building central air conditioning.

[0019] In an exemplary embodiment of this application, the integrated control method for a building central air conditioning system includes the following steps: S1. Based on the constructed system state vector, predict the load and environment in the future time domain to obtain the predicted values; S2. A global optimization control strategy is adopted (using the model prediction control module to solve for the global energy consumption optimal basic setpoint that satisfies the constraints based on the predicted values). S3. Based on the control error and its rate of change between the system's expected setpoint and the actual measured output value, calculate the correction amount for the global energy consumption optimal base setpoint; S4. The global energy consumption optimal base setting value and correction amount are integrated, and the instructions are integrated, limited and interlocked through physical hard constraints and change rate limits to generate the final execution instructions and send them to the execution device; S5. Execute the device response execution command to execute the set value, and at the same time obtain the actual operation data, calculate the reward value and store it in the experience pool for model iteration.

[0020] Since the raw sensor data collected by sensors (such as temperature, humidity, pressure, flow rate, power, etc. within a building system) often contains noise, drift, or packet loss, this application cleans the raw sensor data using a combination of statistical methods and physical mechanisms to ensure the reliability of the data input to the algorithm. For example, the 3-sigma principle is used for outlier detection, and time series methods are used to adjust for missing and outlier values. Data preprocessing is performed through standardization and normalization methods, and exponential moving average (EMA) filtering is used to obtain the input state vector. The system's state vector is then constructed using the processed data. The constructed system state vector includes indoor temperature, indoor relative humidity, real-time cooling load, indoor occupant status (Occ), time encoding, and date type.

[0021] Specifically, this includes the following processing methods or procedures: 1. Outlier handling based on the 3-sigma principle; Assume the data follows a normal distribution Define the sliding window size as N;

[0022] Decision logic:

[0023] According to the characteristics of normal distribution, 99.73% of the data fall within 3 standard deviations; anything outside this range is considered abnormal. in, This represents the window mean at time k. Let N represent the window variance at time k, and N be the sliding window size. This represents historical data within the window. This represents the data at the current moment; 2. Exponential Moving Average (EMA) Filtering: ; Parameter design, A value of 0.3 is used as the smoothing coefficient, balancing response speed and filtering effect. This represents the original sampled data at the current moment. This is represented as smoothed data.

[0024] In step S1, when predicting future load and environment in the time domain, a time series prediction model is used. The prediction is made based on the constructed system state vector and the obtained meteorological data. LSTM network can be used to predict future load and environmental disturbances, such as meteorological disturbances. For example, by using LSTM to predict future load, actions can be taken in advance to eliminate building thermal inertia lag. LSTM can be replaced by GRU, Transformer or XGBoost regression models. Any model with time series prediction function can be used.

[0025] When using LSTM for prediction, the following steps can be taken: [The steps are described in the original text, which is incomplete and cannot be translated accurately.] Meteorological data is input into the LSTM network, and the hidden state is calculated through the forget gate, input gate, and cell state update. Output the load forecast sequence at the forecast time. and disturbance prediction Specifically, as shown below: 1. Input vector definition: ; 2. LSTM element equations: Forget Gate: ; Input Gate: ; Cell State Update:

[0026] Output Gate:

[0027] Predicted output: ; 3. Loss Function:

[0028] The first term is the mean squared error (MSE), which ensures prediction accuracy; the second term is L2 regularization, which prevents overfitting; and N is the number of samples. This represents the actual cooling load of the k-th sample. Let W represent the predicted cooling load for the k-th sample, and let W represent the model weight matrix. This represents the L2 regularization coefficient.

[0029] In this application, the model predictive control module (MPC) is based on a state-space model. It solves for the optimal control sequence that satisfies the constraints within a finite time domain based on the predicted values, thereby obtaining the optimal global energy consumption setpoint that satisfies the constraints. Specifically, the MPC solves for the minimum global energy consumption of the cold source, distribution, and terminal units while satisfying comfort and physical constraints, including: Establish a discretized spatial state model: ; This represents the current state variables, including the indoor temperature. With relative humidity , This indicates the control variables, including chilled water temperature. air supply temperature With pressure difference , This represents disturbance variables, including load forecasts. Compared with outdoor enthalpy; A, B, E These represent the corresponding state transition matrix, control input matrix, and disturbance input matrix, respectively. Construct the objective function J Minimize the deviation between energy consumption and comfort: ; in, Q, R These are the comfort weight matrix and the energy consumption weight matrix, respectively. Indicates the reference state variable. This indicates the prediction period, and the superscript T indicates transpose. Indicates the total power of the air conditioning system; Set constraints; A quadratic programming (QP) problem is constructed based on the multivariate coupling problem. A solver is used to find the optimal control sequence, and the optimal global energy consumption setpoints are obtained, including: ; in, This represents the baseline setting for optimal global energy consumption. This indicates the optimal baseline setting for the cold water temperature. This indicates the optimal basic setting value for the water pump frequency. This indicates the optimal basic setting value for the cooling tower fan frequency. This indicates the optimal basic setting value for valve opening.

[0030] The solver uses the following formula: ; Solve using OSQP or Gurobi; Where U represents the optimization variable vector, such as the sequence of control actions in the future control time domain (e.g., chilled water valve opening, fan frequency, supply air temperature, etc.); H represents the Hessian matrix, a quadratic coefficient matrix composed of the energy consumption weight matrix R, the control smoothing weight matrix S, etc., which determines the penalty intensity for changes in the control quantity; and f represents the linear coefficient vector, composed of state deviation, reference target, disturbance prediction, etc., reflecting the linear relationship between the control quantity and the target state. Represents the inequality constraint matrix, describing the coefficient matrix of all inequality constraints (such as upper and lower limits of control quantities, rate of change limits, temperature / dew point safety constraints, etc.). This represents the vector on the right-hand side of the inequality constraint, corresponding to the boundary value of the constraint (such as the maximum / minimum valve opening, allowable temperature range, etc.). This represents the secondary cost term. The core optimization objective is to minimize control energy consumption and control motion smoothness. This represents the linear cost term, which is used to assist in the optimization objective: minimizing the deviation between the state and the comfort objective.

[0031] By satisfying all safety constraints, a set of control action sequences U is found that minimizes system energy consumption, ensures smooth control actions, and brings the indoor state closest to the comfort target.

[0032] In this application, the constraints include state variable constraints, control variable constraints, control variable rate of change constraints, and dehumidification constraints. These constraints ensure that the state variables, control variables, control variable rates of change, and cold water temperature remain within preset threshold ranges, such as: ; ; ; That is, the chilled water temperature is less than or equal to the dew point temperature minus 5 degrees.

[0033] It should be noted that in this application, the dehumidification constraint can be replaced with other physical safety constraints, such as chiller minimum flow constraint, anti-surge constraint, etc., or together with the dehumidification constraint, they can constitute the corresponding physical safety constraint conditions.

[0034] In this application, ANFIS is used to learn the residuals between the physical model and the actual system, and adaptive correction is performed to compensate for nonlinear errors and model drift online. Specifically, in step S3, the correction amount of the global energy consumption optimal baseline setting is calculated, including: The model predictive control module utilizes the Adaptive Neural Fuzzy Inference System (ANFIS) to learn the residual between the state-space model and the actual system. Using the control error and error change rate as input, it constructs a multi-layer Sugeno-type fuzzy neural network. Through a hybrid learning algorithm (forward least squares + backward gradient descent), it optimizes the network parameters, updates the parameters online, and outputs the correction amount of the model predictive control quantity, i.e., the correction amount of the global energy consumption optimal baseline setpoint, including: ; in, Indicates the correction amount. Indicates control error. Indicates the rate of change of error. To normalize the rule strength, These are the parameters for the conclusion. Specifically, the control error is calculated. and error change rate The input to the ANFIS network undergoes fuzzification, rule-based excitation, normalization, and conclusion calculation to output the correction value. The processing steps include: 1. Input variables: (Control error) (Rate of change of error).

[0035] 2. Five-layer Sugeno fuzzy neural network structure: Layer 1 (fuzzification): Gaussian membership function.

[0036] ,parameter These are prerequisite parameters.

[0037] Level 2 (Rule-based Incentives): ; Layer 3 (Normalization): ; Level 4 (Conclusion Calculation): First-order linear function.

[0038] parameter These are the parameters for the conclusion.

[0039] Level 5 (Total Output): ; 3. Hybrid Learning Algorithm: Forward propagation: fixed prerequisite parameters The least squares (LSE) method is used to optimally solve for the conclusion parameters. Backpropagation: With the conclusion parameters fixed, the premise parameters are updated using gradient descent. ; Error definition: ; in, Indicates the new / old membership function center. Indicates the width of the new / old membership function. Indicates the learning rate. This represents the actual optimal control quantity. Indicates the basic control quantity of MPC. This indicates the ANFIS correction amount.

[0040] By using the Adaptive Neural Fuzzy Inference System (ANFIS) to compensate for the nonlinear residuals of the Model Predictive Control Module (MPC) in the HVAC system, rather than directly replacing the MPC, the MPC is based on a linearized model. ANFIS compensates for nonlinear errors and unmodeled dynamics, thus balancing optimality and adaptability.

[0041] In this application, the ANFIS model is used as a modified model, and can also be replaced by Gaussian process regression (GPR) or simple fuzzy PID, as long as the model has nonlinear residual compensation function.

[0042] In this embodiment of the application, in step S4, the instructions are fused, limited and interlocked based on fuzzy rules and physical hard constraints (such as dew point limits) to output the final execution instructions, thereby achieving a safety guard, that is, using fuzzy rules and hard constraints (such as dew point limits) to ensure that dehumidification safety and the physical limits of the equipment are not exceeded.

[0043] In this application, the optimal global energy consumption baseline setting value and the correction amount are fused to obtain a fused value, including: ; in, Indicates the fusion value. This represents the optimal baseline setting for local energy consumption. Indicates the correction amount. The confidence coefficient is initially set to 0.2 and gradually increases to 1.0 as the system becomes more stable. The weight of the correction amount is dynamically adjusted to achieve a smooth transition from "shadow mode" to "fully automatic".

[0044] In this application, after obtaining the fusion value, the following steps are also included: Perform hard constraint testing, including at least dehumidification constraints, to ensure the dehumidification capacity and dew point temperature limits of the surface cooler, and obtain the final chilled water temperature setting. The rate of change is limited and combined with the historical value from the previous moment to obtain the final set values ​​for the pump frequency and valve opening. Output safety setting value The data is sent down to the underlying controller to collect actual operational data feedback and calculate reward signals. Stored in the experience pool for model iteration; Furthermore, output safety settings. Prior to this, the process also includes a water-air flow matching step: Calculate the total air volume requirement based on the opening distribution of the terminal VAV valves. If the supply air temperature increases, leading to an increase in the total air volume demand, then adjust the total air volume demand accordingly. and pipeline resistance coefficient S Calculate the required pressure difference Check whether the pump head and flow rate meet the hydraulic operating conditions; if the pump capacity is insufficient, limit the rise of the supply air temperature to prevent hydraulic imbalance; otherwise, dynamically adjust the cold water circulation flow rate according to the load prediction sequence to form a pump set operating frequency control command.

[0045] By forcibly embedding dew point temperature constraints and hydraulic matching verification before the final command output, water-air coordinated safety filtration is achieved, ensuring dehumidification safety and system stability.

[0046] In this application, when the executing device responds to the execution command and executes the set value, it uses PID control based on adaptive fuzzy PID control technology, including: ; in, This represents the proportional gain at time t. This represents the initial proportional gain. β This indicates the ANFIS corrected weighting coefficient. This indicates the PID gain correction output. The PID gain is dynamically adjusted based on the error magnitude; when the error is large, K is increased to speed up the response, and when the error is small, K is decreased to prevent oscillation.

[0047] During the execution of PID control, real-time data is recorded simultaneously to form a quadruple. That is, the four-tuple of state-action-reward-next state, which is used for offline training of subsequent ANFIS or RL models; Among them, rewards ; in, For energy consumption costs, For comfort level violation, This is a comfort weighting coefficient. These represent the current system state, control action, reward value, and the system state at the next moment.

[0048] In another aspect, the present invention provides a building central air conditioning integrated control device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the building central air conditioning system integrated control method.

[0049] In practical applications, the technology of this application can be implemented using cloud-edge-device collaborative control combined with hierarchical control. For example, data acquisition sensors, actuators, and PLC / DDC controllers are deployed on the device layer to handle millisecond-level PID execution; edge computing gateways are deployed on the edge layer to perform load prediction, setpoint solving and correction, and final safety command generation based on the acquired data, handling minute-level optimization; and relevant historical databases, model training platforms, and digital twins are deployed on the cloud layer to handle hourly / day-level model retraining. It should be noted that in this application, all calculations can be performed entirely in the cloud or entirely on a local server, and are not limited to a "cloud-edge-device" architecture.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the claims be included within the invention.

[0051] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A comprehensive control method for a building's central air conditioning system, characterized in that, Including the following steps: S1. Based on the constructed system state vector, predict the load and environment in the future time domain to obtain the predicted values; S2. A global optimization control strategy is adopted. The model prediction control module solves for the global energy consumption optimal basic setpoint that satisfies the constraints based on the predicted value. S3. Based on the control error and its rate of change between the system's expected setpoint and the actual measured output value, calculate the correction amount for the global energy consumption optimal base setpoint; S4. The global energy consumption optimal base setting value and correction amount are integrated, and the instructions are integrated, limited and interlocked through physical hard constraints and change rate limits to generate the final execution instructions and send them to the execution device; S5. Execute the device response execution command to execute the set value, and at the same time obtain the actual operation data, calculate the reward value and store it in the experience pool for model iteration.

2. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S1, the constructed system state vector includes indoor temperature, indoor relative humidity, real-time cooling load, indoor occupant status, time encoding, and date type.

3. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S1, when making future time-domain load and environmental predictions, a time-series prediction model is used to make predictions based on the constructed system state vector and the obtained meteorological data.

4. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S2, the model predictive control module, based on the state-space model, solves for the optimal control sequence that satisfies the constraints within a finite time domain according to the predicted values, to obtain the global energy consumption optimal basic setpoint that satisfies the constraints, including: Establish a discretized spatial state model: ； This represents state variables, including indoor temperature and relative humidity. This indicates the controlled variables, including chilled water temperature, supply air temperature, and differential pressure. This represents the disturbance variables, including load forecasts and outdoor enthalpy. A, B, E These represent the corresponding state transition matrix, control input matrix, and disturbance input matrix, respectively. Construct the objective function J Minimize the deviation between energy consumption and comfort: ； in, Q, R These are the comfort weight matrix and the energy weight matrix, respectively. Indicates the reference state variable. This indicates the prediction period, and the superscript T indicates transpose. Indicates the total power of the air conditioning system; Set constraints; A quadratic programming (QP) problem is constructed based on the multivariate coupling problem. A solver is used to find the optimal control sequence, and the optimal global energy consumption setpoints are obtained, including: ； Among them, This represents the baseline setting for optimal global energy consumption. This indicates the optimal baseline setting for the cold water temperature. This indicates the optimal basic setting value for the water pump frequency. This indicates the optimal basic setting value for the cooling tower fan frequency. This indicates the optimal basic setting value for valve opening.

5. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, The constraints include state variable constraints, control variable constraints, control variable rate of change constraints, and dehumidification constraints. These constraints ensure that the state variables, control variables, control variable rates of change, and cold water temperature are within a preset threshold range.

6. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S3, the correction amount for the global energy consumption optimal baseline setting is calculated, including: The model predictive control module utilizes the Adaptive Neural Fuzzy Inference System (ANFIS) to learn the residual between the state-space model and the actual system. Using control error and error change rate as input, it constructs a multi-layer Sugeno-type fuzzy neural network. Through a hybrid learning algorithm, it optimizes the network parameters and outputs the correction amount for the model predictive control quantity, i.e., the correction amount for the global energy consumption optimal baseline setpoint, including: ； in, Indicates the correction amount. Indicates control error. Indicates the rate of change of error. To normalize the rule strength, These are the parameters for the conclusion.

7. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S4, the global energy consumption optimal baseline setting value and the correction amount are merged to obtain a merged value, including: ； in, Indicates the fusion value. This represents the optimal baseline setting for local energy consumption. Indicates the correction amount. The confidence coefficient is initially set to 0.2 and gradually increases to 1.0 as the system becomes more stable. After obtaining the fusion value, the following steps are also included: Perform hard constraint testing, including at least dehumidification constraints, to ensure the dehumidification capacity and dew point temperature limits of the surface cooler, and obtain the final chilled water temperature setting. The rate of change is limited and combined with the historical value from the previous moment to obtain the final set values ​​for the pump frequency and valve opening. Output safety setting value The data is sent down to the underlying controller to collect actual operational data feedback and calculate reward signals. It is stored in the experience pool for model iteration.

8. The integrated control method for a building central air conditioning system according to claim 7, characterized in that, Output safety setting value Prior to this, the process also includes a water-air flow matching step: Calculate the total air volume requirement based on the opening distribution of the terminal VAV valves. ; If the supply air temperature increases, leading to an increase in the total air volume demand, then adjust according to the total air volume demand. and pipeline resistance coefficient S Calculate the required pressure difference Check whether the pump head and flow rate meet the hydraulic operating conditions; if the pump capacity is insufficient, limit the rise of the supply air temperature to prevent hydraulic imbalance; otherwise, dynamically adjust the cold water circulation flow rate according to the load prediction sequence to form a pump set operating frequency control command.

9. The integrated control method for a building central air conditioning system according to claim 1, characterized in that, In step S5, when the device responds to the execution command and executes the set value, PID control based on adaptive fuzzy PID control technology is used, including: ； in, This represents the proportional gain at time t. This represents the initial proportional gain. β This indicates the ANFIS corrected weighting coefficient. This indicates the PID gain correction output.

10. A building central air conditioning integrated control device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the integrated control method for a building central air conditioning system as described in any one of claims 1-8.

Images