Water-cooling central air conditioner load management and control method based on artificial intelligence

Abstract

The invention discloses a water-cooling central air conditioner load management and control method based on artificial intelligence, and relates to the technical field of load management and control. Load prediction is carried out by combining a long and short-term memory time sequence model and a physical constraint neural network in a water chilling unit load prediction link and utilizing historical data; errors are corrected through an energy balance equation of the water chilling unit, and the prediction accuracy is improved; multi-objective optimization control is adopted, COP maximization of a water chilling unit, energy consumption minimization of a cooling tower fan and cooling water supply temperature stability serve as optimization objectives, environment temperature and humidity prediction is combined, and a cooling water supply temperature set point and a cooling tower fan start-stop strategy are dynamically calculated; based on a cooling water supply temperature set point and a fan start-stop strategy, a water chilling unit, a cooling tower fan and a water pump are dynamically adjusted, and the high-frequency operation phenomenon is reduced through fan start-stop optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of load control, and in particular to an artificial intelligence-based water-cooled central air-conditioning load control method. Background Art

[0002] Water-cooled central air-conditioning systems undertake key temperature control tasks, and their operating efficiency directly affects the overall energy consumption level; with the application of artificial intelligence technology, more and more central air-conditioning systems have begun to adopt intelligent load control methods.

[0003] In scenarios such as data centers and hospitals where temperature control requirements are high, the operating status of chillers and cooling towers are closely related, but the current load control method usually optimizes the two separately, lacking a linkage scheduling strategy, resulting in insufficient energy efficiency optimization of the cooling water system; current artificial intelligence load control methods are mostly focused on the optimization control of chillers, such as using deep learning or time series prediction models to improve load prediction accuracy, and optimizing the operating status of chillers accordingly; however, the accuracy of load prediction is not directly equivalent to maximizing system energy efficiency; in actual operation, the optimal scheduling of chillers is often independent of the operating strategy of cooling towers, while cooling tower fans and water pumps still rely on fixed rules or simple temperature difference thresholds for control, which makes it difficult to adapt to changes in load demand of the chiller in a timely manner, resulting in unstable cooling water temperature regulation, which in turn affects the performance coefficient COP of the chiller.

[0004] This problem is particularly prominent in the hot and humid southern regions or in environments with large changes in external temperature and humidity. The existing hierarchical fan control strategy is only based on the fixed supply and return water temperature difference for start and stop, rather than dynamic optimization, resulting in frequent changes in the operation of the cooling tower fan, further aggravating the burden on the chiller. At the same time, the coordinated operation between the water pump, chiller and cooling tower has not been fully considered, and it is easy to have "frequent start and stop of the chiller, high-frequency operation of the cooling tower fan", resulting in additional energy loss. Although some optimization schemes introduce adaptive adjustment mechanisms, such as adjusting the cooling water temperature set point in combination with the chiller load prediction model, or using fuzzy control to optimize the cooling tower fan operation strategy, they have not been able to completely solve the nonlinear coupling relationship between the chiller and the cooling tower, and the system's dynamic matching ability under different load conditions is still limited.

[0005] It can be seen that although the existing artificial intelligence load control method has partially improved the operating efficiency of water-cooled central air-conditioning, due to the insufficient coordinated optimization of chillers and cooling towers, there are still problems such as system response lag, energy efficiency loss, and frequent equipment start and stop; therefore, on the basis of the current load control method, further improving the coordinated optimization capabilities of cooling towers and chillers, and building an intelligent control strategy based on global energy efficiency optimization will become the key direction for energy-saving optimization of water-cooled central air-conditioning. Summary of the invention

[0006] In view of the above existing problems, the present invention is proposed.

[0007] The present invention provides an artificial intelligence-based water-cooled central air-conditioning load control method to solve the problem that traditional load control methods fail to achieve coordinated optimization of chillers and cooling towers, resulting in lagging cooling water temperature regulation, reduced COP, and serious energy efficiency loss.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: The embodiment of the present invention provides a water-cooled central air-conditioning load control method based on artificial intelligence, which includes: Step S1, obtaining the operation data of the water-cooled central air conditioner, cleaning, denoising and feature extraction of the operation data, and constructing a water-cooled central air conditioner operation data set; Step S2, based on the water-cooled central air-conditioning operation data set of step S1, predicting the load demand of the chiller at a future time; Step S3, calculating the cooling water supply temperature set point and the cooling tower fan start and stop strategy according to the chiller load prediction result of step S2; Step S4, dynamically adjusting the operation of the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the cooling tower fan start and stop strategy; Step S5, based on the water-cooled central air-conditioning operation data monitored in step S1 and combined with the adjustment result of step S4, a reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and adjust according to the actual operation effect.

[0009] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, the operating data includes the load data of the chiller, the supply and return water temperature of cooling water, the ambient temperature and humidity, and the internal load demand of the building.

[0010] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, wherein: in step S2, a long short-term memory LSTM time series model is combined with a physical constraint neural PINN network to construct a chiller load prediction model to predict the chiller load demand at future times.

[0011] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, wherein: the step of using the long short-term memory LSTM time series model combined with the physical constraint neural PINN network to build a chiller load prediction model and predicting the chiller load demand at future moments is as follows: Construct an LSTM time series model. The model formula is: , , in, Represents the time step The LSTM hidden layer state, represents the LSTM hidden layer state at the previous time step, Represents the time step The input feature vector is and are the weight matrices of the hidden layer state and the input layer, and is the bias term, is a nonlinear activation function, is the time step The load forecast output is is the output layer weight matrix; On the basis of LSTM prediction, the physical constraint neural network PINN is used to provide the physical constraint conditions of the chiller load. The physical relationship is based on the energy balance equation of the chiller, which is: , in, Represents the time step The cooling load of the chiller is is the cold water flow rate, is the specific heat capacity of water, and Represents the time step The chiller inlet and outlet water temperatures, Define the loss function combined with the PINN constraints, including: , , , in, is the total loss function, is the LSTM prediction error loss, is the physical constraint error loss, is the weight parameter of physical loss, To predict the time series length, is the time step LSTM predicts load value; The final optimization goal is: , in, Represents the set of trainable parameters of the LSTMPINN model.

[0012] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, wherein: in step S3, a multi-objective optimization control MPC method is adopted to construct a cooling water system energy efficiency optimization model.

[0013] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, wherein: in step S3, the cooling water system energy efficiency optimization model is adopted, with the COP maximization of the chiller, the minimization of the energy consumption of the cooling tower fan and the stability of the cooling water supply temperature as the optimization goals, combined with the ambient temperature and humidity prediction data, the cooling water supply temperature set point and the cooling tower fan start and stop strategy are calculated.

[0014] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, the steps of constructing the cooling water system energy efficiency optimization model by using the multi-objective optimization control MPC method are as follows: The cooling water system optimization control model is established, and the energy efficiency optimization target is defined as : , in, represents the optimization objective function of the cooling water system, represents the coefficient of performance of the chiller, represents the total energy consumption of the cooling tower fan, Indicates the stability index of cooling water supply temperature. is the weight coefficient, The performance coefficient of the chiller is calculated as follows: , in, Indicates the cooling load of the chiller. Indicates the input power of the chiller, The total energy consumption of cooling tower fans is calculated as follows: , in, Represents the time step Cooling tower fan power, represents the time step of the optimization time domain, The stability index calculation formula of cooling water supply temperature is: , in, Represents the time step The cooling water supply temperature, represents the target water supply temperature set point, The final optimization goal is defined as: , The constraints are defined as: , , , in, Represents the dynamic change function of cooling water supply temperature, Represents the time step The ambient temperature, Indicates the maximum cooling load of the chiller. Indicates the maximum power of the cooling tower fan.

[0015] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, the step of calculating the cooling water supply temperature set point and the cooling tower fan start-stop strategy is as follows: Calculate the cooling water supply temperature set point in the MPC calculation framework and the start and stop strategy of the cooling tower fan, Calculate cooling water supply temperature set point , the calculation formula is: , Combined with the ambient temperature and humidity forecast data, the model predictive control MPC method is used to update , the update formula is: , in, represents the learning rate parameter, Define the operating status of cooling tower fans : like ,but , like ,but , in, Represents the time step The start and stop status of the cooling tower fan, 1 means running, 0 means off, Indicates the start and stop power threshold of the fan. Final solution : .

[0016] As a preferred solution of the water-cooled central air-conditioning load control method based on artificial intelligence described in the present invention, the steps of dynamically adjusting the operation of the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the cooling tower fan start-stop strategy are as follows: At the calculated cooling water supply temperature set point and cooling tower fan start and stop strategies Based on the above, the chiller, cooling tower fan and water pump are dynamically adjusted. Adjust the operating load of the chiller based on the optimization target , the adjustment formula is: , in, Represents the time step The cooling load of the chiller, Represents the time step The cooling load, is the load regulation gain parameter, Based on the start-stop strategy Control the fan operating status, like ,but , like ,but , in, Represents the time step The cooling tower fan start and stop status, 1 means running, 0 means shut down, is the start and stop power threshold of the cooling tower fan, Adjust the water pump speed according to the cooling water supply temperature deviation , the adjustment formula is: , in, Represents the time step The operating speed of the cooling water pump, Represents the time step The cooling water pump running speed, Adjust the gain factor for the water pump.

[0017] As a preferred solution of the water-cooled central air conditioner load control method based on artificial intelligence described in the present invention, wherein: the water-cooled central air conditioner operation data monitored in step S1 is combined with the adjustment result of step S4, and a reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and the steps of adjusting according to the actual operation effect are as follows: Define the state variables of the reinforcement learning model as: , in, Represents the time step The system status, Represents the time step The cooling water supply temperature, Represents the time step Cooling tower fan power, Represents the time step The cooling load of the chiller is Represents the time step The ambient temperature, The reinforcement learning decision-making method is: , in, Represents the time step Control actions, including chiller load adjustment, cooling tower fan start and stop, and water pump frequency conversion adjustment, is the time step In Status Take Action The reward function after Define the reward function as : , in, Represents reinforcement learning at time step Take Action The reward value after They represent the loss values ​​of COP performance coefficient, fan energy consumption and temperature stability target respectively. is the corresponding weight parameter, The reinforcement learning algorithm continuously optimizes the control strategy to maximize the long-term reward. The optimization formula is: , in, To reinforce the learning strategy, represents the expected value calculation, is the reward discount factor, To optimize the time step in the time domain.

[0018] The beneficial effects of the present invention are as follows: the present invention combines the long short-term memory time series model and the physical constraint neural network in the load prediction link of the chiller, uses historical data to perform load prediction, corrects errors through the energy balance equation of the chiller, and improves the accuracy of the prediction; adopts multi-objective optimization control, maximizes the COP of the chiller, minimizes the energy consumption of the cooling tower fan, and stabilizes the cooling water supply temperature as optimization goals, and combines the ambient temperature and humidity prediction to dynamically calculate the cooling water supply temperature set point and the cooling tower fan start and stop strategy.

[0019] The present invention dynamically adjusts the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the fan start-stop strategy, and reduces the high-frequency operation phenomenon by optimizing the fan start-stop; and introduces a reinforcement learning algorithm, and continuously adjusts the cooling water temperature control logic and the fan start-stop strategy in combination with the operation data and the optimization target. Through reinforcement learning, long-term experience is accumulated, and the control strategy is continuously optimized according to the dynamic changes of the external environment and load demand, thereby enhancing the adaptability to complex environments, while improving the overall energy efficiency and reducing unnecessary energy consumption.

[0020] In summary, the present invention can perform adaptive adjustments under different load conditions, effectively improve the coordinated control capability of the chiller and the cooling tower, reduce the ineffective operation time of the cooling tower fan, and reduce the additional energy consumption caused by frequent start and stop of the chiller. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0022] Figure 1 The present invention is a flow chart of a water-cooled central air conditioning load control method based on artificial intelligence. DETAILED DESCRIPTION

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0026] Example 1, reference Figure 1 This embodiment provides a water-cooled central air-conditioning load control method based on artificial intelligence, comprising the following steps: Step S1, obtaining the operation data of the water-cooled central air conditioner, cleaning, denoising and feature extraction of the operation data, and constructing a water-cooled central air conditioner operation data set; Operation data includes chiller load data, cooling water supply and return temperatures, ambient temperature and humidity, and internal building load requirements; Step S2, based on the water-cooled central air-conditioning operation data set of step S1, predicting the load demand of the chiller at a future time; In step S2, a long short-term memory (LSTM) time series model is combined with a physical constraint neural network (PINN) to construct a chiller load forecasting model to forecast the chiller load demand at future times. The long short-term memory (LSTM) time series model is combined with the physical constraint neural PINN network to build a chiller load forecasting model. The steps for predicting the chiller load demand at future times are as follows: Construct an LSTM time series model. The model formula is: , , in, Represents the time step The LSTM hidden layer state, represents the LSTM hidden layer state at the previous time step, Represents the time step The input feature vector is and are the weight matrices of the hidden layer state and the input layer, and is the bias term, is a nonlinear activation function, is the time step The load forecast output is is the output layer weight matrix; On the basis of LSTM prediction, the physical constraint neural network PINN is used to provide the physical constraint conditions of the chiller load. The physical relationship is based on the energy balance equation of the chiller, which is: , in, Represents the time step The cooling load of the chiller is is the cold water flow rate, is the specific heat capacity of water, and Represents the time step The chiller inlet and outlet water temperatures, Define the loss function combined with the PINN constraints, including: , , , in, is the total loss function, is the LSTM prediction error loss, is the physical constraint error loss, is the weight parameter of physical loss, To predict the time series length, is the time step LSTM predicts load value; The final optimization goal is: , in, Represents the set of trainable parameters of the LSTMPINN model.

[0027] Specifically, LSTM is used to process time series data to predict the chiller load, and long-term dependencies are captured through hidden state propagation. PINN corrects the LSTM prediction error through the physical loss term, and then optimizes the LSTMPINN joint model by minimizing the loss function, thereby predicting the chiller load at future moments. Step S3, calculating the cooling water supply temperature set point and the cooling tower fan start and stop strategy according to the chiller load prediction result of step S2;

[0028] In step S3, a multi-objective optimization control MPC method is used to construct an energy efficiency optimization model for the cooling water system;

[0029] In step S3, the cooling water system energy efficiency optimization model is used to maximize the COP of the chiller, minimize the energy consumption of the cooling tower fan, and stabilize the cooling water supply temperature. The cooling water supply temperature set point and the cooling tower fan start-stop strategy are calculated in combination with the ambient temperature and humidity prediction data.

[0030] Using the multi-objective optimization control MPC method, the steps of constructing the cooling water system energy efficiency optimization model are as follows: The cooling water system optimization control model is established, and the energy efficiency optimization target is defined as : , in, represents the optimization objective function of the cooling water system, represents the coefficient of performance of the chiller, represents the total energy consumption of the cooling tower fan, Indicates the stability index of cooling water supply temperature. is the weight coefficient, The performance coefficient of the chiller is calculated as follows: , in, Indicates the cooling load of the chiller. Indicates the input power of the chiller, The total energy consumption of cooling tower fans is calculated as follows: , in, Represents the time step Cooling tower fan power, represents the time step of the optimization time domain, The stability index calculation formula of cooling water supply temperature is: , in, Represents the time step The cooling water supply temperature, represents the target water supply temperature set point, The final optimization goal is defined as: , The constraints are defined as: , , , in, Represents the dynamic change function of cooling water supply temperature, Represents the time step The ambient temperature, Indicates the maximum cooling load of the chiller. Indicates the maximum power of the cooling tower fan;

[0031] The steps to calculate the cooling water supply temperature set point and the cooling tower fan start and stop strategy are: Calculate the cooling water supply temperature set point in the MPC calculation framework and the start and stop strategy of the cooling tower fan, Calculate cooling water supply temperature set point , the calculation formula is: , Combined with the ambient temperature and humidity forecast data, the model predictive control MPC method is used to update , the update formula is: , in, represents the learning rate parameter, Define the operating status of cooling tower fans : like ,but , like ,but , in, Represents the time step The start and stop status of the cooling tower fan, 1 means running, 0 means off, Indicates the start and stop power threshold of the fan. Final solution : ; Specifically, based on the multi-objective optimization control MPC method, an energy efficiency optimization model of the cooling water system is constructed. Specifically: The optimization direction is to maximize the COP of the chiller, minimize the energy consumption of the cooling tower fan and stabilize the cooling water supply temperature. The influence of different objectives is balanced through weight parameters. Combined with the energy efficiency characteristics of the chiller, the COP is calculated using the energy balance equation and the fan energy consumption loss function is constructed. The cooling water temperature stability index is established, the optimization problem is solved to calculate the cooling water supply temperature set point, and it is adjusted in real time based on the ambient temperature and humidity data. The start and stop strategies of cooling tower fans are determined based on the optimization objectives. The start and stop states are calculated based on the fan power consumption threshold. The system operation is dynamically adjusted through the optimal control strategy. While ensuring the system stability, the overall energy efficiency is improved and unnecessary fan energy consumption is effectively reduced. Step S4, dynamically adjusting the operation of the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the cooling tower fan start and stop strategy;

[0032] Based on the cooling water supply temperature set point and the cooling tower fan start and stop strategy, the steps for dynamically adjusting the operation of the chiller, cooling tower fan and water pump are as follows: At the calculated cooling water supply temperature set point and cooling tower fan start and stop strategies Based on the above, the chiller, cooling tower fan and water pump are dynamically adjusted. Adjust the operating load of the chiller based on the optimization target , the adjustment formula is: , in, Represents the time step The cooling load of the chiller, Represents the time step The cooling load, is the load regulation gain parameter, Based on the start-stop strategy Control the fan operating status, like ,but , like ,but , in, Represents the time step The cooling tower fan start and stop status, 1 means running, 0 means shut down, is the start and stop power threshold of the cooling tower fan, Adjust the water pump speed according to the cooling water supply temperature deviation , the adjustment formula is: , in, Represents the time step The operating speed of the cooling water pump, Represents the time step The cooling water pump running speed, Adjust the gain factor for the water pump; Specifically, based on the cooling water supply temperature set point and the cooling tower fan start-stop strategy, the operation of the chiller, cooling tower fan and water pump is dynamically adjusted to achieve better energy-saving control; Adjust the load of the chiller to respond to the water supply temperature deviation and fan power change, and calculate the load adjustment of the chiller; dynamically control the start and stop status of the fan based on the fan power consumption threshold and optimal strategy to reduce the ineffective operation time of the fan; finally, for the variable frequency control strategy of the water pump, adopt the incremental adjustment method based on temperature deviation to make the water pump operating speed adaptively adjusted with the change of cooling water temperature; improve energy utilization and reduce equipment operation costs; Step S5, based on the water-cooled central air-conditioning operation data monitored in step S1 and combined with the adjustment result of step S4, a reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and adjust according to the actual operation effect; Based on the water-cooled central air-conditioning operation data monitored in step S1 and combined with the adjustment result of step S4, the reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and the steps of adjusting according to the actual operation effect are as follows: Define the state variables of the reinforcement learning model as: , in, Represents the time step The system status, Represents the time step The cooling water supply temperature, Represents the time step Cooling tower fan power, Represents the time step The cooling load of the chiller is Represents the time step The ambient temperature, The reinforcement learning decision-making method is: , in, Represents the time step Control actions, including chiller load adjustment, cooling tower fan start and stop, and water pump frequency conversion adjustment, is the time step In Status Take Action The reward function after Define the reward function as : , in, Represents reinforcement learning at time step Take Action The reward value after They represent the loss values ​​of COP performance coefficient, fan energy consumption and temperature stability target respectively. is the corresponding weight parameter, The reinforcement learning algorithm continuously optimizes the control strategy to maximize the long-term reward. The optimization formula is: , in, To reinforce the learning strategy, represents the expected value calculation, is the reward discount factor, To optimize the time step in the time domain;

[0033] Specifically, a reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy. Based on the reinforcement learning strategy, key decision variables such as the cooling water supply temperature set point, fan start-stop and water pump regulation are dynamically optimized, and the energy efficiency performance of different decisions is quantified through the reward function. The reinforcement learning algorithm accumulates experience through long-term operation and continuously adjusts the control strategy, making the system more adaptable in complex environments, improving operating efficiency, and reducing unnecessary energy consumption.

[0034] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A water-cooled central air-conditioning load control method based on artificial intelligence, characterized in that: include, Step S1, obtaining the operation data of the water-cooled central air conditioner, cleaning, denoising and feature extraction of the operation data, and constructing a water-cooled central air conditioner operation data set; Step S2, based on the water-cooled central air-conditioning operation data set of step S1, predicting the load demand of the chiller at a future time; Step S3, calculating the cooling water supply temperature set point and the cooling tower fan start and stop strategy according to the chiller load prediction result of step S2; Step S4, dynamically adjusting the operation of the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the cooling tower fan start and stop strategy; Step S5, based on the water-cooled central air-conditioning operation data monitored in step S1 and combined with the adjustment result of step S4, a reinforcement learning algorithm is introduced to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and adjust according to the actual operation effect.

2. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence according to claim 1, characterized in that: The operating data includes chiller load data, cooling water supply and return water temperature, ambient temperature and humidity, and internal building load demand.

3. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence as claimed in claim 2, characterized in that: In step S2, a long short-term memory (LSTM) time series model is combined with a physical constraint neural network (PINN) to construct a chiller load forecasting model to predict the chiller load demand at future times.

4. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence as claimed in claim 3, characterized in that: The steps of constructing a chiller load prediction model by using a long short-term memory (LSTM) time series model combined with a physical constraint neural network (PINN) to predict the chiller load demand at a future time are as follows: Construct an LSTM time series model. The model formula is: ， ， in, Represents the time step The LSTM hidden layer state, represents the LSTM hidden layer state at the previous time step, Represents the time step The input feature vector is and are the weight matrices of the hidden layer state and the input layer, and is the bias term, is a nonlinear activation function, is the time step The load forecast output is is the output layer weight matrix; On the basis of LSTM prediction, the physical constraint neural network PINN is used to provide the physical constraint conditions of the chiller load. The physical relationship is based on the energy balance equation of the chiller, which is: ， in, Represents the time step The cooling load of the chiller is is the cold water flow rate, is the specific heat capacity of water, and Represents the time step The chiller inlet and outlet water temperatures, Define the loss function combined with the PINN constraints, including: ， ， ， in, is the total loss function, is the LSTM prediction error loss, is the physical constraint error loss, is the weight parameter of physical loss, To predict the time series length, is the time step LSTM predicts load value; The final optimization goal is: ， in, Represents the set of trainable parameters of the LSTMPINN model.

5. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence as claimed in claim 4, characterized in that: In step S3, a multi-objective optimization control (MPC) method is used to construct an energy efficiency optimization model for the cooling water system.

6. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence as claimed in claim 5, characterized in that: In step S3, the cooling water system energy efficiency optimization model is adopted, with the COP maximization of the chiller, the minimization of the energy consumption of the cooling tower fan and the stability of the cooling water supply temperature as the optimization goals, and the cooling water supply temperature set point and the cooling tower fan start and stop strategy are calculated in combination with the ambient temperature and humidity prediction data.

7. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence according to claim 6, characterized in that: The steps of constructing the cooling water system energy efficiency optimization model by using the multi-objective optimization control MPC method are as follows: The cooling water system optimization control model is established, and the energy efficiency optimization target is defined as : ， in, represents the optimization objective function of the cooling water system, represents the coefficient of performance of the chiller, represents the total energy consumption of the cooling tower fan, Indicates the stability index of cooling water supply temperature. is the weight coefficient, The performance coefficient of the chiller is calculated as follows: ， in, Indicates the cooling load of the chiller. Indicates the input power of the chiller, The total energy consumption of cooling tower fans is calculated as follows: ， in, Represents the time step Cooling tower fan power, represents the time step of the optimization time domain, The stability index calculation formula of cooling water supply temperature is: ， in, Represents the time step The cooling water supply temperature, represents the target water supply temperature set point, The final optimization goal is defined as: ， The constraints are defined as: ， ， ， in, Represents the dynamic change function of cooling water supply temperature, Represents the time step The ambient temperature, Indicates the maximum cooling load of the chiller. Indicates the maximum power of the cooling tower fan.

8. The method for load control of water-cooled central air conditioner based on artificial intelligence as claimed in claim 7, characterized in that: The steps of calculating the cooling water supply temperature set point and the cooling tower fan start and stop strategy are: Calculate the cooling water supply temperature set point in the MPC calculation framework and the start and stop strategy of the cooling tower fan, Calculate cooling water supply temperature set point , the calculation formula is: ， Combined with the ambient temperature and humidity forecast data, the model predictive control MPC method is used to update , the update formula is: ， in, represents the learning rate parameter, Define the operating status of cooling tower fans : like ,but , like ,but , in, Represents the time step The start and stop status of the cooling tower fan, 1 means running, 0 means off, Indicates the start and stop power threshold of the fan. Final solution : 。 9. The method for controlling the load of a water-cooled central air conditioner based on artificial intelligence according to claim 8, characterized in that: The steps of dynamically adjusting the operation of the chiller, cooling tower fan and water pump based on the cooling water supply temperature set point and the cooling tower fan start-stop strategy are: At the calculated cooling water supply temperature set point and cooling tower fan start and stop strategies Based on the above, the chiller, cooling tower fan and water pump are dynamically adjusted. Adjust the operating load of the chiller based on the optimization target , the adjustment formula is: ， in, Represents the time step The cooling load of the chiller, Represents the time step The cooling load, is the load regulation gain parameter, Based on the start-stop strategy Control the fan operating status, like ,but , like ,but , in, Represents the time step The cooling tower fan start and stop status, 1 means running, 0 means shut down, is the start and stop power threshold of the cooling tower fan, Adjust the water pump speed according to the cooling water supply temperature deviation , the adjustment formula is: ， in, Represents the time step The operating speed of the cooling water pump, Represents the time step The cooling water pump running speed, Adjust the gain factor for the water pump.

10. The water-cooled central air conditioner load control method based on artificial intelligence as claimed in claim 9, characterized in that: The steps of introducing a reinforcement learning algorithm based on the water-cooled central air-conditioning operation data monitored in step S1 and combining the adjustment result of step S4 to adaptively optimize the cooling water temperature control logic and the cooling tower fan start-stop strategy, and adjusting according to the actual operation effect are as follows: Define the state variables of the reinforcement learning model as: ， in, Represents the time step The system status, Represents the time step The cooling water supply temperature, Represents the time step Cooling tower fan power, Represents the time step The cooling load of the chiller is Represents the time step The ambient temperature, The reinforcement learning decision-making method is: ， in, Represents the time step Control actions, including chiller load adjustment, cooling tower fan start and stop, and water pump frequency conversion adjustment, is the time step In Status Take Action The reward function after Define the reward function as : ， in, Represents reinforcement learning at time step Take Action The reward value after They represent the loss values ​​of COP performance coefficient, fan energy consumption and temperature stability target respectively. is the corresponding weight parameter The reinforcement learning algorithm continuously optimizes the control strategy to maximize the long-term reward. The optimization formula is: ， in, To reinforce the learning strategy, represents the expected value calculation, is the reward discount factor, To optimize the time step in the time domain.

Images