Central air conditioning water system control methods, devices, electronic equipment and storage media

Abstract

This application discloses a control method, device, electronic device, and storage medium for a central air conditioning water system. The method includes: acquiring operating status data of the central air conditioning water system; generating an action candidate set based on the operating status data using an optimization algorithm; inputting the operating status data and the action candidate set into a multi-agent reinforcement learning decision module to obtain target control parameters; wherein the multi-agent reinforcement learning decision module includes multiple agents, and is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters corresponding to the subsystems from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters; and adjusting the central air conditioning water system according to the target control parameters.

Description

Technical Field

[0001] This application relates to the field of central air conditioning technology, and in particular to a central air conditioning water system control method, device, electronic equipment and storage medium. Background Technology

[0002] As a core component of the HVAC system, the central air conditioning water system accounts for 40%-50% of the building's total energy consumption. The central air conditioning water system mainly consists of chillers, chilled water pumps, cooling water pumps, and cooling towers. Due to the strong coupling between the various devices, the dynamic changes in operating load over time, and the nonlinearity of thermodynamic characteristics, traditional control methods (such as rule-based or PID control methods) cannot dynamically adjust according to the real-time changes in the system state. Under partial load conditions, the system deviates from its efficient operating range for extended periods, resulting in high energy consumption. Summary of the Invention

[0003] This application aims to provide a control method, device, electronic equipment, and storage medium for a central air conditioning water system, which can reduce the energy consumption of the central air conditioning water system.

[0004] In a first aspect, embodiments of this application provide a central air conditioning water system control method, including: Obtain the operating status data of the central air conditioning water system; Based on the operational status data, an action candidate set is generated by searching through an optimization algorithm; wherein, the action candidate set contains multiple candidate actions, each of the candidate actions contains multiple sets of control parameters, and the multiple sets of control parameters correspond to multiple subsystems in the central air conditioning water system; The operating status data and the action candidate set are input into the multi-agent reinforcement learning decision module to obtain the target control parameters. The multi-agent reinforcement learning decision module includes multiple agents. The multi-agent reinforcement learning decision module is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters of the corresponding subsystem from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters. The central air conditioning water system is adjusted according to the target control parameters.

[0005] According to some embodiments of this application, the step of generating an action candidate set by searching through an optimization algorithm based on the running state data includes: Multiple sets of control parameters characterizing the candidate actions are encoded into chromosomes to generate multiple individuals, resulting in an initial population comprising multiple individuals. Using the initial population as the target population, multiple iterative operations are performed on the target population, and the population obtained in the last iteration is determined as the action candidate set. The iterative operations include: Based on the operational status data, the total energy consumption of the central air conditioning water system corresponding to each individual in the target population is calculated using a preset system energy consumption model of the central air conditioning water system, and the fitness of each individual is calculated based on the total system energy consumption. Based on the fitness, selection, crossover, and mutation operations are performed on the target population to generate a new generation population; The new generation population will be used as the target population for the next iteration.

[0006] According to some embodiments of this application, the system energy consumption model includes an objective function, a chiller unit energy consumption model, a chilled water primary pump energy consumption model, a chilled water secondary pump energy consumption model, a cooling water pump energy consumption model, a cooling tower fan energy consumption model, and a frequency correlation model between the cooling water pump and the cooling tower fan. The expression for the objective function is: ; in, For the energy consumption of the refrigeration unit, Energy consumption of the primary chilled water pump. Energy consumption of the chilled water secondary pump. For cooling water pump energy consumption, For cooling tower fan energy consumption, The expression for the chiller unit energy consumption model is: ; in, This refers to the nominal cooling capacity of a single refrigeration unit. This represents the actual cooling capacity. The energy efficiency ratio is the ratio under full load. The chilled water inlet temperature of the refrigeration unit. This refers to the number of refrigeration units in operation. The coefficient is constant. When the chilled water primary pump is set to deliver a constant flow rate, the expression for the energy consumption model of the chilled water primary pump is: ; in, This refers to the nominal energy consumption of a single chilled water primary pump. This represents the actual flow rate of the chilled water primary pump. This refers to the nominal flow rate of the chilled water primary pump. This refers to the number of operating primary chilled water pumps. The coefficient is constant. When the chilled water secondary pump is configured for variable speed delivery, the energy consumption model of the chilled water secondary pump is expressed as follows: ; in This refers to the water flow rate of the chilled water secondary pump. For the head of the chilled water secondary pump, For the efficiency of the chilled water secondary pump, It is a constant. This refers to the number of operating secondary pumps for chilled water; The energy consumption model of the cooling water pump is expressed as follows: ; in The water flow rate of the cooling water pump. For the head of the cooling water pump, For the efficiency of the cooling water pump, This refers to the number of operating cooling water pumps. The energy consumption model of the cooling tower fan is expressed as follows: ; in The airflow of the cooling tower fan. For the total pressure of the cooling tower fan, To improve the efficiency of cooling tower fans, This refers to the number of operating cooling tower fans. It is a constant; The expression for the frequency correlation model between the cooling water pump and the cooling tower fan is as follows: ; in, The frequency ratio of the cooling water pump to the cooling tower fan. is the regression coefficient.

[0007] According to some embodiments of this application, the step of selecting and outputting a set of control parameters of the corresponding subsystem from the action candidate set by multiple intelligent agents based on the operating state data includes: By using multiple agents based on a deep Q-network, the control parameters of each group of the subsystems in the action candidate set are evaluated to obtain the corresponding Q value. The control parameters with the highest Q value are selected as the output.

[0008] According to some embodiments of this application, the deep Q-network is obtained through the following steps: Obtain historical operating data of the central air conditioning water system; Based on the optimization algorithm, a training set including multiple training samples is generated using the historical operating data, and the training samples include historical control parameters. Using the training set, an iterative training operation is performed on a preset initial deep Q-network until a preset convergence condition is met, thereby obtaining the deep Q-network. The iterative training operation includes: One training sample is selected from the training set to obtain the target training sample; The target training samples are input into the system energy consumption model to determine the total system energy consumption of the central air conditioning water system. The reward value is calculated based on the total energy consumption of the system and the preset constraints. Based on the reward value, update the network parameters of the initial depth Q network.

[0009] According to some embodiments of this application, the constraints include: ; in, For the specific heat of water, This refers to the outlet water temperature of the chiller unit; ; in, Let i be the chilled water flow rate through the i-th terminal heat exchanger. ; ; ; in, The minimum chilled water temperature parameter set for refrigeration unit equipment. The maximum value of the chilled water temperature parameter set for the refrigeration unit equipment; ; ; in, The minimum value of the chilled water flow rate parameter set for the terminal heat exchange equipment. The maximum value of the chilled water flow rate parameter set for the terminal heat exchange equipment. This refers to the frequency of the cooling tower fan.

[0010] According to some embodiments of this application, the reward value is calculated using the following formula: in, The reward value is... Total system energy consumption Penalties for violating constraints These are the weighting coefficients.

[0011] Secondly, embodiments of this application provide a central air conditioning water system control device, comprising: Data acquisition module, the data acquisition module is used to acquire the operating status data of the central air conditioning water system; An action generation module is used to generate an action candidate set by searching through an optimization algorithm based on the operating status data; wherein, the action candidate set contains multiple candidate actions, each of the candidate actions contains multiple sets of control parameters, and the multiple sets of control parameters correspond to multiple subsystems in the central air conditioning water system; A parameter determination module is used to input the operating status data and the action candidate set into a multi-agent reinforcement learning decision module to obtain target control parameters. The multi-agent reinforcement learning decision module includes multiple agents, and is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters corresponding to the subsystem from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters. The system adjustment module is used to adjust the central air conditioning water system according to the target control parameters.

[0012] Thirdly, embodiments of this application provide an electronic device, including: At least one processor; At least one memory for storing at least one program; The central air conditioning water system control method described above is implemented when at least one of the programs is executed by at least one of the processors.

[0013] Fourthly, embodiments of this application provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the central air conditioning water system control method described above.

[0014] In this embodiment, based on the current operating status data of the central air conditioning water system, a high-quality action candidate set is generated in real time using an optimization algorithm, which effectively constrains the search space. Then, through the intelligent agents of each subsystem of the central air conditioning water system, with the goal of minimizing the total energy consumption of the central air conditioning water system, the candidate actions are quickly evaluated and collaboratively selected, and the globally optimal target control parameters are output. This improves the overall energy efficiency and adaptive control capability of the central air conditioning water system under complex dynamic conditions and significantly reduces the total energy consumption of the system.

[0015] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0016] The present application will be further described below with reference to the accompanying drawings and embodiments, wherein: Figure 1 A flowchart illustrating an embodiment of the central air conditioning water system control method provided in this application; Figure 2 Simulation data diagram showing the impact of optimizing the number of operating devices on system energy consumption in the embodiments of the central air conditioning water system control method provided in this application; Figure 3 Simulation data diagram showing the impact of chiller inlet water temperature optimization on system energy consumption in the embodiments of the central air conditioning water system control method provided in this application; Figure 4 Simulation data diagram showing the impact of water pump and cooling tower fan frequency optimization on system energy consumption in the embodiments of the central air conditioning water system control method provided in this application; Figure 5 A schematic diagram of an embodiment of the electronic device provided in this application.

[0017] Figure label: Electronic device 100, processor 110, memory 120. Detailed Implementation

[0018] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0019] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0020] In the description of this application, "multiple" refers to two or more. The use of "first" and "second" is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of technical features indicated, or the order in which the technical features are indicated.

[0021] In the description of this application, unless otherwise expressly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0022] To better understand this application, the central air conditioning water system used in this application is described below.

[0023] The central air conditioning water system includes a refrigeration unit module, which is used to install refrigeration unit equipment and generate chilled water through refrigerant circulation; Chilled water circulation module is used to deliver chilled water to end devices to achieve heat exchange; The cooling water circulation module is used to remove the heat generated during the operation of the refrigeration equipment; The pressure stabilization module is used to maintain the stability of the chilled water output and the system pressure; Terminal heat exchange modules are used to install heat exchange equipment to achieve heat exchange between chilled water and air; The water supply and drainage module is used to replace the circulating water and replenish the water lost during the cooling process; The system monitoring module is used to realize automated monitoring and control of the air conditioning system, and includes a sensing unit, a control unit and an interaction unit. Energy-saving module, used to improve the energy efficiency of air conditioning system, includes frequency converter drive unit and heat recovery unit.

[0024] The refrigeration unit module includes a compressor, condenser, evaporator, and throttling components. The throttling components include an expansion valve and a capillary tube, used to regulate the refrigerant flow rate. The chilled water circulation module includes a chilled water pump, piping system, valve assembly, and monitoring instruments. The chilled water pump is divided into a primary pump and a secondary pump. The piping system includes a supply pipe, a return pipe, a distributor, and a collector. The monitoring instruments include a pressure gauge, a thermometer, and a flow meter. The sensing unit of the system monitoring module includes temperature sensors (chilled water supply / return temperature, cooling water supply / return temperature, outdoor wet-bulb temperature), flow sensors (chilled water flow rate, cooling water flow rate), load sensors (system cooling load), and power sensors (energy consumption of each device). The control unit is used to execute the central air conditioning water system control method of this application. The interaction unit consists of a control panel and central monitoring software.

[0025] The cooling water circulation module includes a cooling water pump, a cooling tower, and a water treatment device, which includes a descaling device, a dosing device, and a bypass filter. The pressure stabilization module includes a pressure stabilization device consisting of an expansion tank, a pressure tank, and a makeup water pump, as well as a water softening device, a filter, and a sterilization and disinfection device. The terminal heat exchange module includes an air heat exchange device consisting of a fan coil unit, an air handling unit, and a fresh air unit. The water supply and drainage module includes a makeup water tank, a pressure makeup water valve, and drain pipes and floor drains. The variable frequency drive unit of the energy-saving module is used to control the variable frequency water pump and the variable frequency fan of the cooling tower, and the heat recovery unit is used to recover condensation heat.

[0026] The following is based on Figures 1 to 5 This application describes a central air conditioning water system control method, apparatus, electronic device, and storage medium provided in its embodiments.

[0027] This application provides a method for controlling a central air conditioning water system, such as... Figure 1 As shown, it includes: Step S100: Obtain the operating status data of the central air conditioning water system; Step S200: Based on the operating status data, an optimization algorithm is used to search and generate an action candidate set; wherein, the action candidate set contains multiple candidate actions, each candidate action contains multiple sets of control parameters, and the multiple sets of control parameters correspond to multiple subsystems in the central air conditioning water system; Step S300: Input the operating status data and action candidate set into the multi-agent reinforcement learning decision module to obtain the target control parameters; wherein, the multi-agent reinforcement learning decision module includes multiple agents, and the multi-agent reinforcement learning decision module is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters corresponding to the subsystems from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters; Step S400: Adjust the central air conditioning water system according to the target control parameters.

[0028] In this embodiment, based on the current operating status data of the central air conditioning water system, a high-quality action candidate set is generated in real time using an optimization algorithm, which effectively constrains the search space. Then, through the intelligent agents of each subsystem of the central air conditioning water system, with the goal of minimizing the total energy consumption of the central air conditioning water system, rapid evaluation and collaborative selection are performed, and the globally optimal target control parameters are output. This improves the overall energy efficiency and adaptive control capability of the central air conditioning water system under complex dynamic conditions and significantly reduces the total energy consumption of the system.

[0029] In step S100 above, the current operating status data of the central air conditioning water system is collected in real time by the sensing units in the system monitoring module. The sensing units include: temperature sensors (for measuring chilled water supply / return water temperature, cooling water supply / return water temperature, and outdoor wet-bulb temperature), flow sensors (for measuring the flow rate of the chilled water main and branch pipes, and the cooling water flow rate), load sensors (for calculating or measuring the real-time cooling load of the system), and power sensors (for monitoring the real-time energy consumption of each device). The collected data is integrated into a state vector representing the current instantaneous operating condition of the system to obtain operating status data, which serves as input for subsequent optimization and decision-making.

[0030] In step S200 above, an optimization algorithm is used to generate an action candidate set. The type of optimization algorithm is not limited; for example, genetic algorithms, particle swarm optimization algorithms, etc., can be used. The action candidate set includes a first candidate subset related to the control parameters of the chilled water subsystem and a second candidate subset related to the control parameters of the cooling water subsystem. Control parameters include, but are not limited to, the number of operating chiller units, chilled water inlet temperature, chilled water pump frequency, cooling water pump frequency, and the equivalent fan frequency of the cooling tower.

[0031] In step S300 above, the number of agents in the multi-agent reinforcement learning decision-making module is not limited and can be set according to the number of subsystems. For example, the multi-agent reinforcement learning decision-making module includes two agents: a chilled water agent and a cooling water agent. Each agent contains a pre-trained deep Q-network. The chilled water agent receives operating state data and a first candidate subset, evaluates each set of control parameters on the chilled water side, outputs its Q value, and selects the set of parameters with the highest Q value as its output. Similarly, the cooling water agent evaluates, selects, and outputs each set of control parameters on the cooling water side in the second candidate subset. The control parameters output by the chilled water agent and the control parameters output by the cooling water agent are combined to form a complete set of target control parameters for the current operating state, in order to minimize the total system energy consumption after execution.

[0032] In step S400 above, the obtained target control parameters are sent to the execution unit of the central air conditioning water system. The execution unit includes the frequency converters and start / stop controllers of each device. The system adjusts according to the specific instructions in the target control parameters: for example, it starts and stops the corresponding number of chiller units according to the target control parameters, adjusts the setpoint of the chilled water supply temperature, adjusts the frequency of the chilled water pump and cooling water pump, and adjusts the operating status of the cooling tower fan. The system monitoring module continuously monitors the adjusted operating status and repeats the above steps in the next control cycle (e.g., every 5-10 minutes), thereby realizing the continuous, online energy-saving optimized operation of the central air conditioning water system under all-weather dynamic load.

[0033] In some embodiments of this application, the step S200, "based on the running status data, searching through an optimization algorithm to generate a candidate set of actions," is further explained. Step S200 includes: Step S210: Encode multiple sets of control parameters representing candidate actions into chromosomes to generate multiple individuals, resulting in an initial population containing multiple individuals; Step S220: Using the initial population as the target population, perform multiple iterations on the target population, and determine the population obtained in the last iteration as the action candidate set. The iteration operations include: Step S221: Based on the operating status data, calculate the total energy consumption of the central air conditioning water system for each individual in the target population using the preset system energy consumption model of the central air conditioning water system, and calculate the fitness of each individual based on the total system energy consumption. Step S222: Based on fitness, perform selection, crossover, and mutation operations on the target population to generate a new generation population; Step S223: Use the new generation population as the target population for the next iteration.

[0034] In this embodiment, by introducing a genetic algorithm as the mechanism for generating the action candidate set, a crucial pre-search capability is provided for the real-time optimization control of the central air conditioning water system. This transforms the complex, multi-variable, and strongly coupled control parameter optimization problem into a structured parallel search process. Through encoding operations, continuous and discrete control variables are uniformly represented, enabling the synchronous optimization of heterogeneous parameters such as the number of chiller units, temperature setpoints, and frequencies. Employing a population iterative evolution strategy, each generation rapidly evaluates and filters a large number of candidate solutions based on a high-fidelity system energy consumption model. Selection, crossover, and mutation operators are used to effectively explore the solution space, combining global search capability with guidance. This process can converge and extract a high-quality set of near-optimal solutions—the action candidate set—from a vast solution space within a limited number of iterations. The action candidate set significantly narrows the search range for subsequent agent decisions, overcoming the efficiency bottleneck of blindly exploring in a high-dimensional continuous action space.

[0035] In step S210 above, the combination of control parameters to be optimized is encoded into a fixed-length chromosome. For example, the control parameters include the number of operating chiller units, the chilled water inlet temperature of the chiller units, the chilled water pump frequency, the cooling water pump frequency, and the equivalent fan frequency of the cooling tower. Each control parameter is mapped to one or more gene loci on the chromosome according to its physical meaning and value range. Next, population initialization is performed. An initial population containing K individuals is randomly generated, with no limitation on the value of K, for example, K can be 40. Each individual is a chromosome, representing a complete, randomly generated control parameter scheme.

[0036] In step S220 above, the iterative optimization process is initiated. Using the current system operating status data (including real-time data such as load, temperature, and flow rate) as fixed input, the population undergoes M generations of evolution. The value of M is not limited; for example, M can be 6. Each generation of evolution includes the following operations: For each individual in the population, its chromosomes are decoded and restored to a specific combination of control parameters. This set of parameters, along with the current system operating status, is input into a preset system energy consumption model. This system energy consumption model integrates the physical energy consumption characteristics of chillers, chilled water pumps, cooling water pumps, and cooling tower fans. The system energy consumption model calculates the predicted total energy consumption of the entire central air conditioning water system when executing this set of control parameters under the current operating state. The fitness of an individual is directly defined as the negative value of the predicted total energy consumption. The lower the predicted total energy consumption, the higher the fitness value, and the more superior the individual. Based on the calculated fitness value, a selection mechanism such as roulette wheel is used to select dominant individuals from the current population to enter the mating pool. Individuals with higher fitness have a greater probability of being selected. Single-point or multi-point crossover operations are performed on individuals in the mating pool, exchanging some chromosome segments to generate new individuals that incorporate parental characteristics. Mutation operations are performed on the newly generated individuals with probability, randomly changing the values ​​of certain gene loci to introduce new genetic characteristics and maintain population diversity. After selection, crossover, and mutation operations, a new generation of the population is obtained. After completing the preset M generations of evolution, the iteration is terminated. From the final generation population, the K individuals with the highest fitness (e.g., K = 40) are selected. Decoding these individuals yields K sets of optimized control parameter combinations corresponding to the current running state, thus obtaining the action candidate set.

[0037] In some embodiments of this application, the system energy consumption model includes an objective function, a chiller unit energy consumption model, a chilled water primary pump energy consumption model, a chilled water secondary pump energy consumption model, a cooling water pump energy consumption model, a cooling tower fan energy consumption model, and a frequency correlation model between the cooling water pump and the cooling tower fan. The expression for the objective function is: ; in, For the energy consumption of the refrigeration unit, Energy consumption of the primary chilled water pump. Energy consumption of the chilled water secondary pump. For cooling water pump energy consumption, For cooling tower fan energy consumption, The expression for the chiller unit energy consumption model is: ; in, This refers to the nominal cooling capacity of a single refrigeration unit. This represents the actual cooling capacity. The energy efficiency ratio is the ratio under full load. The chilled water inlet temperature of the refrigeration unit. This refers to the number of refrigeration units in operation. These are constant coefficients, determined through regression fitting using historical data. For example, a0 is -0.346587, a1 is 0.076446, a2 is -0.003575, b0 is -0.283348, b1 is 23.7169, and b2 is -13.548119.

[0038] Assuming the chilled water primary pump operates at a constant flow rate, the energy consumption model for the chilled water primary pump is expressed as follows: ; in, This refers to the nominal energy consumption of a single chilled water primary pump. This represents the actual flow rate of the chilled water primary pump. This refers to the nominal flow rate of the chilled water primary pump. This refers to the number of operating primary chilled water pumps. The constant coefficient, The constant coefficients are determined by fitting regression to historical data. For example, d0 is 0.008165930606975225, d1 is 0.05427126740879877, and d2 is 19.371966638508297. When the chilled water secondary pump is configured for variable speed delivery, the energy consumption model of the chilled water secondary pump is expressed as follows: ; in This refers to the water flow rate of the chilled water secondary pump. For the head of the chilled water secondary pump, For the efficiency of the chilled water secondary pump, For example, constants (e.g.) (9.8) This refers to the number of operating secondary pumps for chilled water; The energy consumption model of the cooling water pump is expressed as follows: ; in The water flow rate of the cooling water pump. For the head of the cooling water pump, For the efficiency of the cooling water pump, This refers to the number of operating cooling water pumps. The energy consumption model of the cooling tower fan is expressed as follows: ; in The airflow of the cooling tower fan. For the total pressure of the cooling tower fan, To improve the efficiency of cooling tower fans, This refers to the number of operating cooling tower fans. It is a constant; The expression for the frequency correlation model between the cooling water pump and the cooling tower fan is as follows: ; in, The frequency ratio of the cooling water pump to the cooling tower fan. is the regression coefficient.

[0039] In this implementation, an objective function aimed at minimizing the total system energy consumption was constructed, establishing a clear direction for overall optimization. Furthermore, dedicated energy consumption sub-models were established for core energy-consuming equipment such as chillers, primary and secondary chilled water pumps, cooling water pumps, and cooling tower fans, each conforming to their operational characteristics. These dedicated energy consumption sub-models fully consider the nonlinear operating characteristics of the equipment, such as the complex relationship between chiller energy consumption and cooling load and inlet water temperature, and the correlation between pump energy consumption and flow rate, head, and efficiency. A frequency-based energy consumption correlation model for cooling water pumps and cooling tower fans was also introduced, accurately reflecting the energy consumption characteristics under variable frequency control, a core energy-saving method. The system energy consumption model enables the optimization algorithm and intelligent agent to rely on high-fidelity energy consumption predictions rather than empirical estimations during search, evaluation, and decision-making. This ensures that the final output control parameters are physically feasible and accurately guide the system to the optimal state of global energy efficiency, fundamentally guaranteeing the effectiveness and engineering practicality of the optimized control.

[0040] In some embodiments of this application, further control is performed in step S300 by "selecting and outputting a set of control parameters corresponding to the subsystem from the action candidate set based on the running state data by multiple intelligent agents". Step S300 includes: Step S310: Using multiple agents based on a deep Q-network, evaluate each set of control parameters of the corresponding subsystem in the action candidate set to obtain the corresponding Q value; Step S320: Select the set of control parameters with the highest Q value as the output.

[0041] In this implementation, each agent uses the Q-value output by its deep Q-network to quantitatively evaluate the long-term energy efficiency gains of different combinations of control parameters while satisfying the current system operating state, and selects the optimal solution accordingly. The complex global joint optimization problem is decomposed into multiple parallel, dimensionally simplified distributed decision-making processes, significantly improving decision-making efficiency and reliability in high-dimensional space while maintaining global consistency. The locally optimal parameters output by each agent are ultimately integrated into a unified target control parameter, thereby achieving automatic coordination and global energy efficiency optimization across multiple subsystems.

[0042] For example, the multi-agent reinforcement learning decision-making module specifically includes two agents: a freezing water agent and a cooling water agent. Each agent internally encapsulates a pre-trained deep Q-network with fixed parameters. The deep Q-network adopts a fully connected structure, contains two hidden layers, each with 128 neurons, and uses the ReLU function as the activation function.

[0043] After the decision-making process begins, the operational status data and the action candidate set generated by the optimization algorithm are synchronously input into the multi-agent reinforcement learning decision-making module. The action candidate set includes a first candidate subset related to the chilled water circulation and a second candidate subset related to the cooling water circulation. The chilled water agent receives the operational status data and the first candidate subset. Each set of candidate parameters in the first candidate subset contains control parameters for the chilled water subsystem, such as a combination of the number of operating chiller units, the chilled water inlet temperature setpoint, and the chilled water pump frequency. For each set of candidate parameters in the first candidate subset, the chilled water agent inputs it along with the operational status data into its internal deep Q-network. The deep Q-network calculates and outputs a corresponding Q-value using forward propagation. This Q-value represents an estimate of the long-term cumulative discounted reward that can be obtained by executing this set of control parameters under the current operational status, i.e., a quantitative assessment of long-term energy efficiency benefits. The agent iterates through and evaluates all candidate sets in the first candidate subset, and then selects the set of control parameters with the highest output Q-value as its decision result. The cooling water agent executes the same decision-making process in parallel. After the two agents complete their independent evaluation and selection, they combine the optimal control parameters selected by each agent to obtain the target control parameters.

[0044] In some embodiments of this application, the deep Q-network in step S310 is obtained through the following steps: Step S311: Obtain historical operating data of the central air conditioning water system; Step S312: Based on the optimization algorithm, use historical running data to generate a training set including multiple training samples, the training samples including historical control parameters; Step S313: Using the training set, iteratively train the preset initial deep Q-network until the preset convergence condition is met to obtain the deep Q-network. The iterative training operation includes: Step S314: Select a training sample from the training set to obtain the target training sample; Step S315: Input the target training samples into the system energy consumption model to determine the total system energy consumption of the central air conditioning water system; Step S315: Calculate the reward value based on the total system energy consumption and preset constraints; Step S316: Update the network parameters of the initial depth Q network based on the reward value.

[0045] In this embodiment, historical operating data accumulated from the central air conditioning water system is used as the source of training samples. A training set covering various operating conditions is constructed by reusing historical control parameters and guiding the search of optimization algorithms. During iterative training, each training sample is calculated using a pre-established system energy consumption model. This model accurately simulates the energy consumption output and state changes of the central air conditioning water system after executing actions with specific control parameters, and generates corresponding reward values ​​strictly according to preset constraints. The initial deep Q-network continuously learns these calculated "state-action-reward" mapping relationships, gradually optimizing its network parameters to efficiently learn the globally energy-optimal control strategy.

[0046] For example, a deep Q-network has two hidden layers (128 neurons / layer), the activation function is ReLU, and the objective function is expressed as: ; Among them, the target value For the target network parameters, Discount factor; First, acquire historical operating data of the central air conditioning water system, including but not limited to timestamps, system cooling load, outdoor dry and wet bulb temperatures, chilled water and cooling water temperatures and flow rates, and energy consumption records of major equipment. Based on this historical operating data, determine the constant coefficients of each sub-model in the system energy consumption model through regression fitting, thereby establishing a complete system energy consumption model that can be used for simulation. Simultaneously, extract or preset system operating constraints (such as upper and lower frequency limits, temperature range, minimum COP value, etc.) from the historical operating data.

[0047] Based on historical operational data, optimization algorithms (such as genetic algorithms) are used to generate a training set for training the initial deep Q-network. Specifically, the system state (e.g., load, temperature) at each moment in the historical records, along with the actual control parameters used at that time (e.g., number of units, frequency), are used as a base sample. The genetic algorithm perturbs, recombines, and optimizes these historical control parameters within their feasible domain, generating a series of new control parameter combinations associated with each historical state but with different parameters. These pairs of historical states and control parameter combinations constitute the training samples, and the set of all training samples forms the training set.

[0048] Initialize an initial deep Q-network (its structure may contain multiple hidden layers, such as two layers with 128 neurons each) and set training hyperparameters (such as learning rate, discount factor γ, etc.). Then, perform multiple rounds of iterative training using the training set until the network performance converges. Each iteration of training includes the following steps: Based on a greedy strategy, a training sample is selected from the training set. The training sample contains a specific historical system state and a corresponding set of control parameters. The control parameters from the selected training sample are input into the system energy consumption model. Based on the input control parameters and the corresponding historical system state, the system energy consumption model simulates and calculates the total system energy consumption under this condition, and determines whether there are constraint violations according to preset constraints. Subsequently, based on the calculated total energy consumption and constraint violation status, the reward value corresponding to that sample is calculated using a preset reward function.

[0049] The historical system states, control parameters, and calculated reward values ​​from the training samples are used as a set of training data to update the network parameters of the initial deep Q-network. The goal of the update is to make the long-term value prediction of the control parameter for that state more closely resemble the target value calculated based on that reward value. A gradient descent-based backpropagation algorithm can be used for the update. Training is considered converged when the training loss stabilizes or reaches a preset number of iterations (e.g., 10 training epochs). All connection weights and bias parameters of the initial deep Q-network at this point are saved, resulting in the trained deep Q-network model.

[0050] In some embodiments of this application, the constraints include: Based on the constraints of the refrigeration capacity and chilled water flow rate of the chiller unit, an energy balance expression is established: ; in, For the specific heat of water, This refers to the outlet water temperature of the chiller unit; Based on the chilled water circulation process, construct the mass balance expression: ; in, Let i be the chilled water flow rate through the i-th terminal heat exchanger. Considering the matching relationship between chilled water pumps and cooling water pumps, the number of chilled water pumps and cooling water pumps is set to be equal during operation: ; Based on the constraints regarding the number of chiller units and water pumps, it is necessary to ensure that at least the chiller units, chilled water primary pumps, and cooling water pumps are in operation and do not exceed the total number of units. Therefore: ; The set temperature of the chilled water entering the chiller unit should meet the following requirements: ; in, The minimum chilled water temperature parameter set for refrigeration unit equipment. The maximum value of the chilled water temperature parameter set for the refrigeration unit equipment; The frequency ratio of the cooling water pump to the cooling tower fan should meet the following requirements: ; The chilled water flow rate of the chiller unit should be set to meet the following requirements: ; in, The minimum value of the chilled water flow rate parameter set for the terminal heat exchange equipment, and The maximum value of the chilled water flow rate parameter set for the terminal heat exchange equipment.

[0051] According to some embodiments of this application, the reward value is calculated using the following formula: ; in, As a reward value, Total system energy consumption Penalties for violating constraints These are the weighting coefficients.

[0052] In some embodiments of this application, the central air conditioning water system includes three parallel chillers (rated power 314kW, cooling capacity 1878kW) and three chilled water pumps (rated power 29.39kW, head 24m, flow rate 346m³ / h). 3 / h), 3 cooling water pumps (rated power 38.72kW, head 24m, flow rate 450m³ / h), 3 / h), 3 cooling towers (rated power 11.6kW, flow rate 130100m³ / h), 3 / h); The system monitoring module consists of a sensor controller, a sensor unit, an execution unit, and an interaction unit. The sensor unit includes temperature, pressure, flow, and differential pressure sensors. The execution unit includes a frequency converter and an electric valve for regulating the speed of the water pump and fan. The interaction unit consists of a control panel and central monitoring software. The energy-saving module consists of a frequency converter drive unit and a heat recovery unit. The frequency converter drive system is used to control the frequency converter water pump and the frequency converter fan of the cooling tower. The heat recovery unit recovers condensation heat to save energy consumption.

[0053] The overall control process of the air conditioning water system includes: 1) Refrigeration cycle: After the system detects the cooling demand of the terminal heat exchange equipment, it starts the refrigeration unit. The compressor first compresses the low-temperature, low-pressure gaseous refrigerant into a high-temperature, high-pressure gas and sends it into the condenser. The cooling water from the cooling tower absorbs the heat of the refrigerant and condenses the refrigerant into a high-pressure liquid. The liquid refrigerant enters the evaporator after being depressurized and cooled by the throttling device. The refrigerant absorbs heat through evaporation to further cool the chilled water. 2) Cooling capacity delivery and distribution: The system controls the primary chilled water pump to operate at a constant flow rate, directly pumping chilled water from the evaporator to the distributor. Then, the secondary pump adjusts its speed according to the load changes of the terminal equipment through the frequency converter to achieve variable flow delivery. At this time, the distributor distributes the low-temperature chilled water to each branch, and then enters the terminal equipment and exchanges heat with the air through the surface cooler. After the chilled water absorbs heat and rises in temperature, it is collected by the water collector and returned to the evaporator. 3) Cooling water circulation and heat dissipation: Start the cooling water pump to pump cooling water from the condenser outlet to the cooling tower. In the tower, the cooling water is evenly sprayed onto the packing material through the water distributor. After full contact with the air, it evaporates and dissipates heat. With the help of the fan ventilation, the heat is dissipated more quickly. The cooled cooling water is returned to the condenser to achieve the cooling water circulation effect. The side filter and chemical dosing device can continuously remove impurities and prevent scaling and microbial growth. 4) Terminal equipment temperature regulation: The fan coil unit drives the air to flow through the coil, transferring the cooling capacity of the chilled water to the indoor environment. The air handling unit mixes the fresh air and return air, filters and cools it, and then sends it to various areas of the room through the air duct. At the same time, the condensate generated by the terminal equipment is discharged to the floor drain through the drain pipe to avoid water accumulation and bacterial growth. 5) Dynamic adjustment of the control system: Based on the real-time monitoring of parameters such as the temperature, pressure and flow of chilled water supply and return water temperature and pipeline pressure difference by temperature, pressure and flow sensors, the multi-agent reinforcement learning decision module adjusts the compressor frequency, water pump speed and cooling tower fan frequency according to load changes, and controls the flow of each branch through electric regulating valves to balance the hydraulic system. 6) Energy-saving and optimized operation: Based on frequency conversion technology, under partial load, the target control parameters output by the multi-agent reinforcement learning decision module reduce the speed of chilled water pump, cooling water pump and fan to reduce energy consumption. Furthermore, by recovering heat from the condenser, the recovered heat is used for domestic hot water or heating, further reducing energy consumption.

[0054] In some embodiments of this application, the impact of optimizing the number of operating devices, the inlet water temperature of the chiller unit, and the frequency of the water pump and cooling tower fan on the energy consumption optimization of the system is studied, including: 1) The impact of optimizing the number of operating devices on system energy consumption Traditional control methods adjust the number of chiller units based on the actual cooling load of the air conditioning system, and simultaneously adjust the number of cooling water pumps and chilled water primary pumps, with decisions relying on empirical rules. This application utilizes a multi-agent reinforcement learning decision-making module, which generates a joint action candidate set using a genetic algorithm. Two agents then coordinate to determine the number of operating units and pumps based on the minimum energy consumption of the objective function, achieving a globally optimal match. Combined with... Figure 2Simulation data shows that when the load is at an intermediate value (12.6~19.8kW), this application saves an average of about 10% energy compared to the traditional method. At high load (19.8~21.4kW) and low load (10.9~12.6kW), this application still maintains an energy saving advantage of 5%~8%, and the switching of the number of operating units is smoother, avoiding energy waste caused by frequent start-stop.

[0055] 2) The impact of optimizing the chiller inlet water temperature on system energy consumption Traditional methods fix the inlet temperature of the chiller unit and meet load demands by changing the chilled water flow rate, which cannot match the optimal energy efficiency under dynamic loads. This application, however, dynamically adjusts the chilled water agent based on the overall system energy-saving target, using a set of action candidates generated by a genetic algorithm, allowing it to adaptively optimize within a 9-12℃ range as the load changes. Figure 3 As shown, compared with the traditional fixed temperature setting, the method of this application saves an average of about 7% of energy consumption. Moreover, during periods of large load fluctuations (such as 10:00~16:00), the inlet water temperature is adjusted synchronously with the load, the COP value of the chiller unit is significantly improved, and the energy-saving effect is more prominent.

[0056] 3) The impact of optimizing the frequency of water pumps and cooling tower fans on system energy consumption Traditional methods operate at fixed pump / fan frequencies, adjusting flow solely through valve throttling, resulting in significant hydraulic losses and high energy consumption under partial load. This application, however, optimizes the chilled water pump frequency using a chilled water intelligent agent and collaboratively optimizes the cooling water pump and cooling tower fan frequencies using a cooling water intelligent agent, enabling dynamic matching of equipment frequencies with load. Figure 4 As shown, under full load conditions, the pump and fan operate at the rated frequency, and the total energy consumption of this application is close to that of the traditional method. Under partial load conditions, the optimal frequency (30~40Hz) is output by the multi-agent reinforcement learning decision module to reduce the operating speed of the equipment. At this time, the energy consumption of this application is reduced by an average of about 8% compared with the traditional method, and the frequency adjustment is smooth, avoiding the life loss caused by drastic fluctuations in the equipment.

[0057] In addition, this application provides a central air conditioning water system control device, including: The data acquisition module is used to acquire the operating status data of the central air conditioning water system. The action generation module is used to search and generate an action candidate set based on the running status data and through optimization algorithms. The action candidate set contains multiple candidate actions, and each candidate action contains multiple sets of control parameters, which correspond to multiple subsystems in the central air conditioning water system. The parameter determination module is used to input the operating status data and action candidate set into the multi-agent reinforcement learning decision module to obtain the target control parameters. The multi-agent reinforcement learning decision module includes multiple agents. The multi-agent reinforcement learning decision module is used to minimize the total energy consumption of the central air conditioning water system. Based on the operating status data, the multiple agents select and output a set of control parameters corresponding to the subsystems from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters. The system adjustment module is used to adjust the central air conditioning water system according to the target control parameters.

[0058] The central air conditioning water system control device provided in this application embodiment can realize the various processes implemented in the above method embodiment and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0059] In addition, one embodiment of this application also discloses an electronic device 100, such as... Figure 5 As shown, it includes: At least one processor 110; At least one memory 120 is used to store at least one program; The central air conditioning water system control method described above is implemented when at least one program is executed by at least one processor 110.

[0060] The electronic device 100 provided in this application embodiment can implement the various processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0061] In addition, embodiments of this application provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the central air conditioning water system control method described above.

[0062] The computer-readable storage medium provided in this application embodiment can implement the various processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0063] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0064] The embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application.

Claims

1. A method for controlling a central air conditioning water system, characterized in that, include: Obtain the operating status data of the central air conditioning water system; Based on the operational status data, an action candidate set is generated by searching through an optimization algorithm; wherein, the action candidate set contains multiple candidate actions, each of the candidate actions contains multiple sets of control parameters, and the multiple sets of control parameters correspond to multiple subsystems in the central air conditioning water system; The operating status data and the action candidate set are input into the multi-agent reinforcement learning decision module to obtain the target control parameters. The multi-agent reinforcement learning decision module includes multiple agents. The multi-agent reinforcement learning decision module is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters of the corresponding subsystem from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters. The central air conditioning water system is adjusted according to the target control parameters.

2. The central air conditioning water system control method according to claim 1, characterized in that, Based on the operational status data, an action candidate set is generated through an optimization algorithm, including: Multiple sets of control parameters characterizing the candidate actions are encoded into chromosomes to generate multiple individuals, resulting in an initial population comprising multiple individuals. Using the initial population as the target population, multiple iterative operations are performed on the target population, and the population obtained in the last iteration is determined as the action candidate set. The iterative operations include: Based on the operational status data, the total energy consumption of the central air conditioning water system corresponding to each individual in the target population is calculated using a preset system energy consumption model of the central air conditioning water system, and the fitness of each individual is calculated based on the total system energy consumption. Based on the fitness, selection, crossover, and mutation operations are performed on the target population to generate a new generation population; The new generation population will be used as the target population for the next iteration.

3. The central air conditioning water system control method according to claim 2, characterized in that, The system energy consumption model includes an objective function, a chiller unit energy consumption model, a chilled water primary pump energy consumption model, a chilled water secondary pump energy consumption model, a cooling water pump energy consumption model, a cooling tower fan energy consumption model, and a frequency correlation model between the cooling water pump and the cooling tower fan. The expression for the objective function is: ； in, For the energy consumption of the refrigeration unit, Energy consumption of the primary chilled water pump. Energy consumption of the chilled water secondary pump. For cooling water pump energy consumption, For cooling tower fan energy consumption, The expression for the chiller unit energy consumption model is: ； in, This refers to the nominal cooling capacity of a single refrigeration unit. This represents the actual cooling capacity. The energy efficiency ratio is the ratio under full load. The chilled water inlet temperature of the refrigeration unit. This refers to the number of refrigeration units in operation. The coefficient is constant. When the chilled water primary pump is set to deliver a constant flow rate, the expression for the energy consumption model of the chilled water primary pump is: ； in, This refers to the nominal energy consumption of a single chilled water primary pump. This represents the actual flow rate of the chilled water primary pump. This refers to the nominal flow rate of the chilled water primary pump. This refers to the number of operating primary chilled water pumps. The coefficient is constant. When the chilled water secondary pump is configured for variable speed delivery, the energy consumption model of the chilled water secondary pump is expressed as follows: ； in This refers to the water flow rate of the chilled water secondary pump. For the head of the chilled water secondary pump, For the efficiency of the chilled water secondary pump, It is a constant. This refers to the number of operating secondary pumps for chilled water; The energy consumption model of the cooling water pump is expressed as follows: ； in The water flow rate of the cooling water pump. For the head of the cooling water pump, For the efficiency of the cooling water pump, This refers to the number of operating cooling water pumps. The energy consumption model of the cooling tower fan is expressed as follows: ； in The airflow of the cooling tower fan. For the total pressure of the cooling tower fan, To improve the efficiency of cooling tower fans, This refers to the number of operating cooling tower fans. It is a constant; The expression for the frequency correlation model between the cooling water pump and the cooling tower fan is as follows: ； in, The frequency ratio of the cooling water pump to the cooling tower fan. is the regression coefficient.

4. The central air conditioning water system control method according to claim 1, characterized in that, The step of selecting and outputting a set of control parameters for the corresponding subsystem from the action candidate set by multiple intelligent agents based on the operational status data includes: By using multiple agents based on a deep Q-network, the control parameters of each group of the subsystems in the action candidate set are evaluated to obtain the corresponding Q value. The control parameters with the highest Q value are selected as the output.

5. The central air conditioning water system control method according to claim 4, characterized in that, The deep Q-network is obtained through the following steps: Obtain historical operating data of the central air conditioning water system; Based on the optimization algorithm, a training set including multiple training samples is generated using the historical operating data, and the training samples include historical control parameters. Using the training set, an iterative training operation is performed on a preset initial deep Q-network until a preset convergence condition is met, thereby obtaining the deep Q-network. The iterative training operation includes: One training sample is selected from the training set to obtain the target training sample; The target training samples are input into the system energy consumption model to determine the total system energy consumption of the central air conditioning water system. The reward value is calculated based on the total energy consumption of the system and the preset constraints. Based on the reward value, update the network parameters of the initial depth Q network.

6. The central air conditioning water system control method according to claim 5, characterized in that, The constraints include: ； in, For the specific heat of water, This refers to the outlet water temperature of the chiller unit; ； in, Let i be the chilled water flow rate through the i-th terminal heat exchanger. ； ； ； in, The minimum chilled water temperature parameter set for refrigeration unit equipment. The maximum value of the chilled water temperature parameter set for the refrigeration unit equipment; ； ； in, The minimum value of the chilled water flow rate parameter set for the terminal heat exchange equipment. The maximum value of the chilled water flow rate parameter set for the terminal heat exchange equipment. This refers to the frequency of the cooling tower fan.

7. The central air conditioning water system control method according to claim 5, characterized in that, The reward value is calculated using the following formula: ； in, The reward value is... Total system energy consumption Penalties for violating constraints These are the weighting coefficients.

8. A central air conditioning water system control device, characterized in that, include: Data acquisition module, the data acquisition module is used to acquire the operating status data of the central air conditioning water system; An action generation module is used to generate an action candidate set by searching through an optimization algorithm based on the operating status data; wherein, the action candidate set contains multiple candidate actions, each of the candidate actions contains multiple sets of control parameters, and the multiple sets of control parameters correspond to multiple subsystems in the central air conditioning water system; A parameter determination module is used to input the operating status data and the action candidate set into a multi-agent reinforcement learning decision module to obtain target control parameters. The multi-agent reinforcement learning decision module includes multiple agents, and is used to minimize the total energy consumption of the central air conditioning water system as the optimization objective. Based on the operating status data, the multiple agents select and output a set of control parameters corresponding to the subsystem from the action candidate set, and determine the control parameters output by the multiple agents as the target control parameters. The system adjustment module is used to adjust the central air conditioning water system according to the target control parameters.

9. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; The central air conditioning water system control method as described in any one of claims 1 to 7 is implemented when at least one of the programs is executed by at least one of the processors.

10. A computer-readable storage medium, characterized in that, It stores a processor-executable program, which, when executed by the processor, is used to implement the central air conditioning water system control method as described in any one of claims 1 to 7.

Images