Temperature control system for water-cooled air conditioners in computer rooms

Abstract

The present invention discloses a temperature control system for water-cooled air conditioners in computer rooms, and relates to the field of temperature control technology. The system includes a coupling analysis module, a collaborative control module, a dynamic control module, and a correction module. By collecting the temperature distribution and power consumption data of the cabinet in real time, a heat flow propagation model is constructed using a graph neural network, and the refrigerant flow rate and the terminal outlet temperature setting value are output. Combined with the efficiency curves of water pumps, fans, and compressors, a non-dominated sorting genetic algorithm is used to optimize the total energy consumption of the system, the equipment life, and the refrigeration stability, and to generate control instructions. According to the layout of the computer room and the control instructions, the thermodynamic constraints are verified and the refrigerant allocation priority matrix is ​​generated to ensure coordinated response of the equipment. The actual and simulated data are compared through a digital twin platform, and the edge weights of the heat flow propagation model are reversely corrected to optimize the system performance. The system can realize precise control of the computer room environment, improve energy efficiency and equipment life, and ensure refrigeration stability.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature control, and in particular to a temperature control system applied to a water-cooled air conditioner in a computer room. Background Art

[0002] With the rapid development of information technology and cloud computing services, data centers and computer rooms have increasing requirements for temperature control systems. Electronic equipment such as servers, storage devices, and network devices in the computer room are constantly running and generate a lot of heat, which requires an efficient air conditioning system to perform temperature control to ensure the stable operation of the equipment. Traditional air conditioning systems mostly rely on simple temperature sensors for global temperature control. However, with the increasing sophistication of equipment and the expansion of computer room scale, a single temperature control method can no longer meet the needs of high efficiency and stability. In addition, modern computer rooms often have a more complex spatial layout, and traditional water-cooled air-conditioning systems face the challenge of optimizing refrigerant distribution, improving energy efficiency, and extending equipment life. Therefore, how to improve the temperature control efficiency and equipment management of the water-cooled air-conditioning system in the computer room through intelligent technology has become a problem that needs to be solved urgently.

[0003] The existing water-cooled air-conditioning temperature control system in computer rooms generally has defects such as dynamic response lag, single energy efficiency optimization dimension, and poor cross-scenario adaptability. Its static control model cannot accurately match the real-time fluctuations of server load and complex environmental interference, resulting in uneven cooling distribution, local overheating or redundant cooling problems. In addition, the operation and maintenance mode that relies on manual experience to adjust parameters is difficult to ensure the stability of high-density computer rooms under extreme working conditions. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a temperature control system for water-cooled air conditioners in computer rooms. Through multimodal data fusion analysis and transfer learning adaptive control, combined with environmental coupling modeling, multi-objective collaborative optimization and digital twin feedback correction, it breaks through the limitations of traditional single threshold control and single device tuning, realizes dynamic planning of refrigerant distribution paths, global balance between equipment life and energy consumption, and second-level response to sudden load scenarios, significantly reduces the overall energy consumption of computer rooms and improves temperature control accuracy, and provides efficient, reliable and adaptive temperature control solutions for high-density computing scenarios. The above-mentioned background technology problems are solved.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a temperature control system for water-cooled air conditioners in computer rooms, comprising the following modules: a coupling analysis module, a collaborative control module, a dynamic control module, and a correction module; the coupling analysis module is used to collect temperature distribution data of server cabinets, and obtain cabinet power consumption data in combination with current sensors, and construct a computer room heat flow propagation model through a graph neural network, and output a refrigerant branch flow rate setting value and a terminal outlet air temperature setting value; the collaborative control module is used to receive the refrigerant branch flow rate setting value and the terminal outlet air temperature setting value, and combine the water pump efficiency curve, the cooling tower heat dissipation coefficient and the compressor health index to obtain the total energy consumption of the system, the equipment Taking life balance and refrigeration stability as optimization goals, the non-dominated sorting genetic algorithm is used to generate water pump inverter frequency control instructions, cooling tower fan speed instructions and chilled water valve opening instructions; the dynamic control module is used to verify whether the instructions meet the constraints of the thermodynamic equation according to the spatial layout characteristics of the computer room and the control instructions, generate the refrigerant allocation priority matrix and equipment response timing instructions, and send them to the water pump inverter, cooling tower fan and chilled water valve controller through the industrial communication protocol; the correction module is used to compare the actual temperature distribution with the simulation data through the digital twin platform, calculate the cooling capacity distribution deviation rate and energy consumption overflow, and reversely correct the edge weight parameters of the heat flow propagation model.

[0006] Furthermore, the specific process of constructing a heat flow propagation model for a computer room through a graph neural network is as follows: the server cabinets are abstracted as nodes in a graph structure, and the physical connection channels and airflow paths between cabinets are defined as edges; the temperature distribution data and cabinet power consumption data are encoded as temperature feature vectors and power consumption feature vectors of the nodes, respectively; based on the node feature aggregation mechanism of the graph neural network, the edge weight parameters are updated through multi-layer message passing to characterize the intensity of heat flow propagation; the external environment temperature and humidity data are integrated, the attenuation coefficient of the heat flow path is dynamically corrected, and a three-dimensional heat flow propagation topology map of the computer room is generated.

[0007] Furthermore, the specific process of outputting the refrigerant branch flow rate threshold and the terminal outlet temperature setting value is as follows: according to the edge weight parameters of each branch in the heat flow propagation topology diagram, the priority level of the refrigerant branches is divided; according to the inverse proportional relationship between the priority level and the heat flow propagation intensity, the adjustment range of the refrigerant flow rate of each branch is dynamically calculated; combined with the real-time monitoring data of the terminal outlet temperature sensor, the outlet temperature setting value is iteratively corrected through the thermodynamic equilibrium equation to ensure the dynamic matching of the refrigerant supply and the heat load.

[0008] Furthermore, combining the water pump efficiency curve, cooling tower heat dissipation coefficient and compressor health index, the specific process of taking the total system energy consumption, equipment life balance and refrigeration stability as the optimization goals is as follows: extract the optimal working range in the water pump efficiency curve, and establish a nonlinear mapping relationship between energy consumption and refrigerant flow; integrate the dynamic change characteristics of the cooling tower heat dissipation coefficient and the degradation trend of the compressor health index to construct an equipment life loss balance evaluation function; set multi-objective optimization constraints, including the refrigerant flow rate threshold, the terminal outlet temperature deviation limit and the refrigeration stability error range.

[0009] Furthermore, the specific process of generating water pump inverter frequency control instructions, cooling tower fan speed instructions and chilled water valve opening instructions through non-dominated sorting genetic algorithm is as follows: the water pump frequency, fan speed and valve opening parameters are encoded into chromosome vectors of the multi-objective solution space; based on the Pareto dominance relationship and reference point screening strategy, the non-dominated solution set with the lowest energy consumption, the most balanced life loss and the highest refrigeration stability is iteratively calculated; the chromosome vector in the Pareto optimal solution set is decoded to generate water pump inverter frequency values, cooling tower fan speed values ​​and chilled water valve opening gradient values ​​that are compatible with the industrial controller interface.

[0010] Furthermore, according to the spatial layout characteristics of the computer room and the control instructions, the specific process of verifying whether the instructions meet the constraints of the thermodynamic equations is as follows: load the structural characteristic parameters of the three-dimensional layout of the computer room, including the cabinet spacing, the distribution of air supply channels and the topological connection relationship of the refrigerant branches; predict the temperature field distribution and the refrigerant flow velocity field distribution of the computer room under the action of the control instructions through the forward propagation calculation of the physical information neural network; input the prediction results into the embedded thermodynamic equation solver to verify whether the simplified form of the Navier-Stokes equations and the energy conservation equation constraints are met, and screen the candidate instruction set that conforms to the laws of physics.

[0011] Furthermore, the specific process of generating the refrigerant distribution priority matrix and the equipment response timing instructions is as follows: according to the refrigerant branch flow rate adjustment amplitude of the candidate instruction set and the edge weight parameters of the heat flow propagation topology diagram, the refrigerant distribution priority matrix under the burst load scenario is constructed; based on the control delay characteristics of water pumps, fans and valves and the equipment start-stop inertia parameters, a device instruction sequence for staggered response in the time dimension is generated; the priority matrix and instruction sequence are encoded into a sequential logic instruction set executable by the industrial controller to ensure the spatiotemporal coordination of cooling capacity distribution and equipment action.

[0012] Furthermore, the specific process of comparing the actual temperature distribution with the simulation data through the digital twin platform and calculating the cooling capacity allocation deviation rate and energy consumption overflow is as follows: align the actual temperature distribution data with the simulated temperature field of the digital twin in time and space, and extract the temperature gradient distribution differences in key areas; identify abnormal temperature aggregation areas based on the adaptive clustering algorithm, and calculate the matching degree between the cooling capacity allocation deviation rate and the expected refrigerant flow rate; synchronize the real-time power data of the equipment energy consumption monitoring system, compare the simulated expected energy consumption curve, and quantify the energy consumption overflow of the cooling tower, water pump and valve.

[0013] Furthermore, the specific process of reversely correcting the edge weight parameters of the heat flow propagation model is as follows: constructing a corrected gradient function of the edge weights of the heat flow propagation model according to the cold distribution deviation rate and the energy consumption overflow; updating the edge weight parameters of the graph neural network through a dynamic learning rate adjustment mechanism combined with the convergence trend of historical correction data; inputting the corrected heat flow propagation model into the digital twin platform for verification and iteration until the temperature gradient distribution difference is lower than the preset convergence threshold.

[0014] The present invention has the following beneficial effects:

[0015] (1) Applied to the temperature control system of water-cooled air conditioners in computer rooms. The computer room heat flow propagation model constructed by graph neural network is combined with multi-modal fusion of cabinet-level temperature and power consumption data to realize dynamic modeling of heat flow paths, solve the problems of large computational complexity and lagging cooling capacity allocation of traditional grid simulation models, and significantly reduce the cooling capacity allocation error rate; multi-objective optimization based on non-dominated sorting genetic algorithm, synchronously optimizes the total energy consumption of the system, equipment life balance and refrigeration stability, breaks through the conflict between energy consumption and life caused by single-objective tuning, and significantly improves the overall energy efficiency ratio. The compressor health index and the dynamic heat dissipation coefficient of the cooling tower are introduced to construct an equipment life loss balance evaluation function, actively balance the loss rate of multiple devices, and avoid the risk of system downtime caused by overload of a single device; combined with the nonlinear mapping relationship of the water pump efficiency curve, dynamically match the refrigerant flow and load demand, reduce redundant energy consumption under inefficient conditions, and significantly extend the service life of key equipment.

[0016] (2) Applied to the temperature control system of the water-cooled air conditioner in the computer room, the thermodynamic equation constraints are embedded in the neural network to verify the physical feasibility of the control command, solve the command oscillation problem caused by the traditional pure data-driven model ignoring the natural laws, and significantly shorten the response delay under the sudden load scenario; the spatiotemporal collaborative control logic ensures the precise matching of cooling capacity distribution and equipment action, avoids the instantaneous overload caused by the simultaneous start and stop of multiple devices, and significantly reduces the peak load of the equipment. The digital twin platform compares the actual and simulated data, locates the abnormal area of ​​cooling capacity distribution through the adaptive clustering algorithm, and reversely corrects the parameters of the heat flow propagation model in combination with the dynamic learning rate mechanism to achieve continuous iterative optimization of the model prediction accuracy.

[0017] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a flow chart of a temperature control system for a water-cooled air conditioner in a computer room. DETAILED DESCRIPTION

[0019] The embodiment of the present application solves the problems of uneven temperature distribution in the computer room, low energy efficiency, short equipment life, and inability to effectively cope with sudden loads and extreme working conditions in the traditional air-conditioning system through the temperature control system applied to the water-cooled air-conditioning in the computer room. By introducing coupling analysis, collaborative control, dynamic control and correction modules, the system can accurately set the refrigerant flow rate and temperature, optimize the equipment operating parameters, and adjust the control strategy according to real-time feedback, thereby achieving more efficient energy consumption management, improving equipment stability and long-term equipment protection, and ensuring the intelligent, refined and efficient operation of the computer room air-conditioning system.

[0020] The overall idea of ​​the solution in the embodiments of this application is as follows:

[0021] The temperature distribution data of the server cabinet is collected, and the cabinet power consumption data is obtained by combining the current sensor. The heat flow propagation model of the computer room is constructed through the graph neural network, and the refrigerant branch flow rate setting value and the terminal outlet air temperature setting value are output.

[0022] Receive the refrigerant branch flow rate set value and the terminal outlet air temperature set value, combine the water pump efficiency curve, cooling tower heat dissipation coefficient and compressor health index, take the system total energy consumption, equipment life balance and refrigeration stability as optimization goals, and generate the water pump inverter frequency control command, cooling tower fan speed command and chilled water valve opening command through the non-dominated sorting genetic algorithm.

[0023] According to the spatial layout characteristics of the computer room and the control instructions, verify whether the instructions meet the constraints of the thermodynamic equations, generate the refrigerant allocation priority matrix and equipment response timing instructions, and send them to the water pump inverter, cooling tower fan and chilled water valve controller through the industrial communication protocol.

[0024] The actual temperature distribution is compared with the simulation data through the digital twin platform, the cooling capacity distribution deviation rate and energy consumption overflow are calculated, and the edge weight parameters of the heat flow propagation model are reversely corrected.

[0025] See also Figure 1The embodiment of the present invention provides a technical solution: a temperature control system for a water-cooled air conditioner in a computer room, comprising the following modules: a coupling analysis module, a collaborative control module, a dynamic control module, and a correction module; the coupling analysis module is used to collect temperature distribution data of a server cabinet, obtain cabinet power consumption data in combination with a current sensor, construct a computer room heat flow propagation model through a graph neural network, and output a refrigerant branch flow velocity setting value and a terminal outlet air temperature setting value; the collaborative control module is used to receive the refrigerant branch flow velocity setting value and the terminal outlet air temperature setting value, and combine the water pump efficiency curve, the cooling tower heat dissipation coefficient, and the compressor health index to balance the total energy consumption of the system and the equipment life The degree and refrigeration stability are taken as the optimization goals, and the frequency control instructions of the water pump inverter, the speed instructions of the cooling tower fan and the opening instructions of the chilled water valve are generated through the non-dominated sorting genetic algorithm; the dynamic control module is used to verify whether the instructions meet the constraints of the thermodynamic equations according to the spatial layout characteristics of the computer room and the control instructions, generate the refrigerant allocation priority matrix and the equipment response timing instructions, and send them to the water pump inverter, the cooling tower fan and the chilled water valve controller through the industrial communication protocol; the correction module is used to compare the actual temperature distribution with the simulation data through the digital twin platform, calculate the cooling capacity distribution deviation rate and the energy consumption overflow, and reversely correct the edge weight parameters of the heat flow propagation model.

[0026] In this implementation scheme, the coupled analysis module: Temperature distribution data: The temperature values ​​of the key points on the surface or inside of each cabinet in the computer room obtained by the distributed temperature sensor reflect the spatial distribution of the heat source. Cabinet power consumption data: The real-time power consumption of the server cabinet measured by the current sensor represents the size of the heat load. Graph neural network (GNN): A deep learning model based on a graph structure, which is used to process non-Euclidean data (such as the topology of the computer room), and model complex relationships through node feature aggregation and edge weight update. Heat flow propagation model: A mathematical model that describes the heat transfer path from the heat source (server) to the cold source (air conditioning terminal) in the computer room, which is used to predict the cooling demand. Refrigerant branch flow rate setting value: The target value of the cooling water flow rate in each refrigerant pipeline, which is used to adjust the cooling capacity distribution. Terminal outlet temperature setting value: The temperature target value of the air outlet of the air conditioner terminal, which is used to control the supply air temperature. Collaborative control module: Water pump efficiency curve: Describes the energy consumption characteristic curve of the water pump under different flow rates and heads, which is used to optimize the operation efficiency of the water pump. Cooling tower heat dissipation coefficient: A parameter that characterizes the heat dissipation capacity of the cooling tower, reflecting the performance of the cooling tower under different working conditions. Compressor health index: a parameter that evaluates the remaining life of the compressor based on its operating status (such as vibration, temperature, and current), which is used to optimize the equipment usage strategy. Non-dominated sorting genetic algorithm: a multi-objective optimization algorithm that uses the Pareto front screening strategy to find the optimal solution set for energy consumption, life, stability and other objectives. Water pump inverter frequency control instruction: a control signal that adjusts the speed of the water pump motor to change the refrigerant flow. Cooling tower fan speed instruction: a control signal that adjusts the speed of the cooling tower fan to adjust the heat dissipation efficiency. Chilled water valve opening instruction: a control signal that adjusts the opening of the chilled water valve to control the cooling capacity distribution ratio. Dynamic control module: room space layout characteristics: including structural parameters such as cabinet spacing, air supply channel distribution, and refrigerant branch topological connection relationship. Thermodynamic equation constraints: a simplified form based on the Navier-Stokes equations and the law of conservation of energy, used to verify the physical feasibility of control instructions. Refrigerant allocation priority matrix: a matrix that describes the priority of cooling capacity allocation for each refrigerant branch under sudden load scenarios, which is used to dynamically adjust the cooling capacity supply. Equipment response timing instructions: Plan the instruction sequence of the time sequence of the action of equipment such as water pumps, fans, and valves to avoid instantaneous overload. Industrial communication protocol: Used to send control instructions to industrial controllers (such as PLCs). Correction module: Digital twin platform: A virtual simulation platform based on physical models and real-time data, used to simulate the temperature field and energy consumption status of the computer room. Cooling capacity distribution deviation rate: The ratio of the difference between the actual cooling capacity distribution and the simulation expectation, reflecting the control accuracy. Energy consumption overflow: The difference between the actual energy consumption of the equipment and the simulated expected energy consumption, used to locate the source of energy consumption anomalies. Edge weight parameter: A parameter that describes the strength of the heat flow propagation path in the graph neural network, used to correct the cooling capacity distribution path prediction.

[0027] Specifically, the specific process of constructing a heat flow propagation model for a computer room through a graph neural network is as follows: the server cabinets are abstracted as nodes in a graph structure, and the physical connection channels and airflow paths between cabinets are defined as edges; the temperature distribution data and the cabinet power consumption data are encoded as the temperature feature vector and the power consumption feature vector of the node respectively; based on the node feature aggregation mechanism of the graph neural network, the edge weight parameters are updated through multi-layer message passing to characterize the intensity of heat flow propagation; the external environment temperature and humidity data are integrated to dynamically correct the attenuation coefficient of the heat flow path to generate a three-dimensional heat flow propagation topology map of the computer room.

[0028] For example, the implementation scenario of a data center computer room is as follows:

[0029] 1. Equipment room parameters:

[0030]

[0031] 2. Data collection example:

[0032] 2.1 Temperature distribution data (unit: °C):

[0033]

[0034] 2.2 Power consumption data (unit: kW):

[0035]

[0036] 3. Graph Neural Network Implementation Parameters:

[0037]

[0038] In this implementation scheme, the definition of nodes and edges is as follows: Node: Each server cabinet is regarded as a node in the graph structure. Each node has specific attributes, usually including the cabinet temperature data (temperature feature vector) and cabinet power consumption data (power consumption feature vector). Edge: The physical connection channels and airflow paths between cabinets are defined as edges in the graph. The edges represent the paths of heat flow propagation, and the weight values ​​of the edges represent the intensity of the heat flow. Encoding of node features The features of each node include: Temperature feature vector ,in Represents the node index, which represents the temperature data of the cabinet. Power consumption feature vector ,in Represents the node index, which represents the power consumption data of the cabinet. These features can be represented by multi-dimensional vectors, combined with the real-time temperature and power consumption information of the cabinet. The core idea of ​​the graph neural network is to update the node status by aggregating node features. The specific process is as follows: For each node , its updated features Update by aggregating the features of adjacent nodes. The formula is: ;in: Is a node In the The updated features of the layer. Is a node In the The characteristics of the layer, Is a node Neighbors. Representation Node The set of neighbor nodes. and are the learned weight matrix and bias term respectively. is the activation function, which uses the ReLU nonlinear function. Through the message passing mechanism, the edge weight (i.e. the intensity of heat flow propagation) will be updated in each layer of the network. The update formula of the edge weight is as follows: ;in: is the edge weight between node u and node v at the t+1th layer (i.e., the intensity of heat flow propagation). and are the degrees (number of neighboring nodes) of nodes u and v respectively. and are the features of nodes u and v at layer t respectively. is the attenuation coefficient, which represents the correction coefficient of the external environment temperature and humidity on the heat flow propagation path. In the actual environment, the propagation of heat flow is not only affected by the physical connection between the institutions, but also by the changes in the external environment, such as temperature and humidity. Therefore, the external environment data needs to be added to the model to dynamically correct the attenuation coefficient of the positive flow path. The generation of the three-dimensional heat flow propagation topology map, based on the iterative calculation of the graph neural network and the dynamically corrected heat flow propagation model, will generate a three-dimensional heat flow propagation topology map of the computer room. This map shows the temperature distribution, heat flow propagation path and its intensity in different areas of the computer room, and can be updated in real time according to the actual operation data, providing data support for the cooling management of the computer room. In summary, the graph neural network can accurately predict and optimize the heat flow propagation process in the computer room through iterative calculation and information transmission, and dynamically adjust the heat flow propagation path according to real-time environmental data, providing strong support for the heat dissipation and energy efficiency management of the computer room.

[0039] Specifically, the specific process of outputting the refrigerant branch flow rate threshold and the terminal outlet temperature setting value is as follows: according to the edge weight parameters of each branch in the heat flow propagation topology diagram, the priority level of the refrigerant branch is divided; according to the inverse proportional relationship between the priority level and the heat flow propagation intensity, the adjustment range of the refrigerant flow rate of each branch is dynamically calculated; combined with the real-time monitoring data of the terminal air outlet temperature sensor, the outlet temperature setting value is iteratively corrected through the thermodynamic equilibrium equation to ensure the dynamic matching of the refrigerant supply and the heat load.

[0040] In this implementation plan, the relationship between the priority division of the refrigerant branch and the heat flow transmission intensity is as follows: First, the priority level of the refrigerant branch is divided according to the edge weight parameters of each branch in the heat flow transmission topology diagram of the computer room. The edge weight parameter reflects the heat flow transmission intensity of the refrigerant branch, that is, the heat flow conduction capacity. The priority of each branch is inversely proportional to the edge weight, that is, the higher the branch edge weight, the greater the heat flow transmission intensity, and the lower its priority; vice versa. Priority division: Assume is the edge weight of the i-th refrigerant branch, is the priority level of the branch. Then the inverse proportional relationship between priority and heat flux intensity can be defined as: ;in: is the priority of the i-th refrigerant branch. is the heat flow propagation intensity corresponding to branch i (i.e., edge weight parameter). Calculation of refrigerant flow rate adjustment range: When dynamically calculating the adjustment range of the refrigerant branch flow rate, the relationship between the priority level and the heat flow propagation intensity needs to be considered. According to the priority level, the flow rate adjustment range of each branch is dynamically calculated. Usually, branches with higher priorities (stronger heat flow propagation) will be assigned lower refrigerant flow rates, while branches with lower priorities will be allowed higher flow rates. Calculation of flow rate adjustment range: Assume the refrigerant flow rate adjustment range and For the i-th branch, the flow rate adjustment range and the priority of the branch Inversely proportional, it can be expressed as: ;in: is the minimum refrigerant flow rate of the ith branch. is the maximum refrigerant flow rate of the ith branch. The basic flow rate (constant) determines the proportion of the flow rate range. Correction of the terminal outlet temperature set value: Dynamically adjust the terminal outlet temperature set value by real-time monitoring of the data collected by the temperature sensor. This process is based on the thermodynamic equilibrium equation and takes into account the dynamic impact of changes in flow rate on the heat load. Through the thermodynamic equilibrium equation, the temperature set value should be iteratively corrected according to the relationship between the refrigerant flow rate and the heat load. Thermodynamic equilibrium equation: Assume that the relationship between the refrigerant flow rate V and the outlet temperature is described by the thermodynamic equilibrium equation. Set the relationship between the refrigerant flow rate and temperature change as: ;in: is the change of the terminal outlet air temperature at time t. is the cooling load (affected by external temperature, heat flow, etc.). is the temperature change of the refrigerant. is the specific heat capacity of the refrigerant. is the refrigerant flow rate, which is adjusted dynamically over time. According to this equation, the terminal outlet temperature setting value is adjusted dynamically : ;in: is the terminal outlet air temperature at time t. is the initial target temperature.

[0041] Specifically, combining the water pump efficiency curve, cooling tower heat dissipation coefficient and compressor health index, the specific process of taking system total energy consumption, equipment life balance and refrigeration stability as optimization targets is as follows: extract the optimal working range in the water pump efficiency curve, and establish a nonlinear mapping relationship between energy consumption and refrigerant flow; integrate the dynamic change characteristics of the cooling tower heat dissipation coefficient and the degradation trend of the compressor health index to construct an equipment life loss balance evaluation function; set multi-objective optimization constraints, including refrigerant flow rate threshold, terminal outlet temperature deviation limit and refrigeration stability error range, to form a three-dimensional optimization space of energy consumption-life-stability.

[0042] In this implementation, the optimal working range of the water pump efficiency curve is extracted and nonlinear energy consumption mapping is performed: the relationship between the efficiency and flow rate of the water pump presents nonlinear characteristics. In the optimization, it is first necessary to extract the optimal working range in the water pump efficiency curve, and establish the relationship between the energy consumption of the water pump and the refrigerant flow rate as a nonlinear energy consumption mapping. The nonlinear mapping relationship between the energy consumption of the water pump and the refrigerant flow rate: Assume that the energy consumption of the water pump is ,in For the refrigerant flow rate, the nonlinear relationship can be expressed by the following formula: ;in: For the pump at flow The energy consumption below. is the refrigerant flow rate. is the fitting coefficient, describing the nonlinear relationship between pump performance and flow rate. Integration of cooling tower heat dissipation coefficient and compressor health degradation trend: The cooling tower heat dissipation performance and compressor health degradation trend will affect the long-term operation efficiency of the equipment. Therefore, the equipment life loss balance evaluation function is constructed by combining the cooling tower heat dissipation coefficient and the compressor health index. Equipment life loss balance evaluation function: Assume the compressor health is , the cooling tower heat dissipation coefficient is , the life loss evaluation function is: ;in: It is the estimated value of equipment life loss. For time Always check the compressor health. is the cooling tower heat dissipation coefficient. For the time interval. is the adjustment coefficient, which characterizes the influence of various factors on life loss. Setting of multi-objective optimization constraints: During the optimization process, multiple constraints need to be considered, such as refrigerant flow rate, terminal outlet temperature deviation, and refrigeration stability, to ensure that the system can meet the requirements. Refrigerant flow rate constraint: Set the upper and lower limits of the refrigerant flow rate to and , it is required to keep the refrigerant flow rate within this range: ; Terminal outlet air temperature deviation limit: Set the terminal outlet air temperature target value to , then the terminal outlet temperature deviation Must meet: ;in The maximum tolerance value of temperature deviation. Refrigeration stability error range: Set the refrigeration stability error , requiring the system to remain within the error range: ; Construct three-dimensional optimized space based on energy consumption, equipment life balance and refrigeration stability.

[0043] Specifically, the specific process of generating water pump inverter frequency control instructions, cooling tower fan speed instructions and chilled water valve opening instructions through non-dominated sorting genetic algorithm is as follows: the water pump frequency, fan speed and valve opening parameters are encoded into chromosome vectors in the multi-objective solution space; based on the Pareto dominance relationship and reference point screening strategy, the non-dominated solution set with the lowest energy consumption, the most balanced life loss and the highest refrigeration stability is iteratively calculated; the chromosome vector in the Pareto optimal solution set is decoded to generate the water pump inverter frequency value, cooling tower fan speed value and chilled water valve opening gradient value that are compatible with the industrial controller interface.

[0044] For example, 4. Example output of the heat flow propagation model:

[0045] 4.1 Refrigerant branch flow rate setting value (unit: m³ / h):

[0046]

[0047] 4.2 Terminal outlet air temperature setting value (unit: ℃):

[0048]

[0049] 5. Collaborative control optimization example:

[0050] 5.1 Pump efficiency curve fitting equation:

[0051]

[0052] 5.2 NSGA-II algorithm parameters:

[0053]

[0054] 5.3 Pareto optimal solution example:

[0055]

[0056] In this implementation, control parameters such as water pump frequency, cooling tower fan speed, and chilled water valve opening are encoded as chromosome vectors. Each chromosome vector contains multiple genes, representing water pump frequency, fan speed, and valve opening. is a chromosome vector, where: Represents the frequency of the water pump inverter. Represents the speed of the cooling tower fan. Represents the opening of the chilled water valve. Multiple objective functions: minimizing energy consumption, balancing equipment life loss, and maximizing refrigeration stability. The objective functions are expressed as follows: Energy consumption target: Assume is the energy consumption objective function. Life loss balance objective: Assume is the life loss balance function. Refrigeration stability target: Assume is the cooling stability objective function. Use non-dominated sorting to sort the chromosomes and obtain the Pareto optimal solution set. The optimal solution is selected through the Pareto dominance relationship. Assume and There are two solutions, if Not inferior to , and is better than , then it is called Dominate , recorded as The reference point screening strategy helps to reduce the redundancy of the solution set and ensure the diversity of the solution set in the target space. is the kth reference point, the objective function The corresponding solution will be screened according to its distance from the reference point. Decoding and controller interface generation: After obtaining the Pareto optimal solution set, the solution vector is decoded into the actual control parameters. For each Pareto optimal solution vector , perform decoding operation: Represents the frequency command of the water pump inverter. Represents the speed command of the cooling tower fan. Represents the opening gradient value of the chilled water valve. Energy consumption objective function: in: is the power of the kth water pump. is the operating time of the kth water pump. Life loss balance objective function: in: is the life loss of the kth water pump. is the life loss of the kth fan. is the life loss of the kth valve. Refrigeration stability objective function: in: is the temperature difference of the kth cooling point.

[0057] Specifically, according to the spatial layout characteristics of the computer room and the control instructions, the specific process of verifying whether the instructions meet the constraints of the thermodynamic equations is as follows: load the structural characteristic parameters of the three-dimensional layout of the computer room, including cabinet spacing, air supply channel distribution, and refrigerant branch topological connection relationship; predict the temperature field distribution and refrigerant flow velocity field distribution of the computer room under the action of the control instructions through the forward propagation calculation of the physical information neural network; input the prediction results into the embedded thermodynamic equation solver to verify whether the simplified form of the Navier-Stokes equations and the energy conservation equation constraints are met, and screen the candidate instruction set that conforms to the laws of physics.

[0058] In this implementation scheme, it is necessary to obtain the spatial layout characteristics of the computer room. These characteristic parameters include: Cabinet spacing: describes the spatial distance between cabinets, which has an important impact on airflow distribution and heat conduction. Air supply channel distribution: indicates how cold air passes through various areas of the computer room. This affects the speed and direction of the airflow, which in turn affects the temperature distribution. Refrigerant branch topological connection relationship: describes the connection relationship of the refrigerant flow path. The refrigerant flow rate and temperature are crucial to the heat dissipation of the computer room, so accurate modeling of the refrigerant flow path is very important. These parameters will be used as inputs to the physical information neural network to help the model better predict the temperature and refrigerant flow rate distribution in the computer room. Use the physical information neural network for forward propagation calculations to predict the impact of control instructions on the temperature field and refrigerant flow rate field of the computer room. The advantage of the physical information neural network is that it can combine physical constraints (such as fluid dynamics equations and heat conduction equations) to optimize predictions. Computer room temperature field distribution: By simulating and calculating air flow and heat conduction, predict the temperature changes at different locations in the computer room under given control instructions (such as water pump frequency, fan speed, etc.). Refrigerant velocity field distribution: The refrigerant velocity distribution is predicted based on the refrigerant flow rate and the refrigerant branch topology. This directly affects the cooling efficiency and system stability of the computer room. The temperature field and refrigerant velocity field results predicted by the physical information neural network model are input into the embedded thermodynamic equation solver for further analysis. The solver is based on the classical thermodynamic principles and fluid mechanics equations for verification to ensure that the prediction results conform to the actual physical laws. Simplified form of the Navier-Stokes equation: This equation is used to describe the movement of incompressible fluids. The simplified equation is usually used to represent the flow of air and refrigerant in this application. The core of the equation is to describe the relationship between the velocity field, pressure field and temperature field of the fluid, and consider the effects of viscosity, inertia and external forces (such as fan and water pump forces) on the flow of the fluid. Energy conservation equation: This equation describes the energy conversion and transfer in the system. In the computer room environment, the main focus is on the distribution of temperature and the transfer of thermal energy. The energy conservation equation ensures that the heat in the computer room will not be generated or disappeared out of thin air. It maintains temperature balance through heat exchange between the refrigerant and the air. After verification by the embedded solver, control instruction sets that conform to the laws of physics are screened out. These instructions can ensure that the temperature and refrigerant flow rate in the computer room meet the desired thermodynamic behavior. Candidate instruction sets: These instruction sets include control parameters such as water pump frequency, cooling tower fan speed, and chilled water valve opening. They need to meet the constraints of the above physical equations to ensure efficient heat dissipation and cooling stability in the computer room. The control instructions can be dynamically adjusted in a complex computer room environment to optimize the operating status of the equipment and ensure that it meets the thermodynamic constraints, thereby improving the energy efficiency and stability of the system.

[0059] Specifically, the specific process of generating the refrigerant distribution priority matrix and the equipment response timing instructions is as follows: according to the refrigerant branch flow rate adjustment amplitude of the candidate instruction set and the edge weight parameters of the heat flow propagation topology diagram, the refrigerant distribution priority matrix under the sudden load scenario is constructed; based on the control delay characteristics of water pumps, fans and valves and the equipment start-stop inertia parameters, a device instruction sequence for staggered response in the time dimension is generated; the priority matrix and instruction sequence are encoded into a sequential logic instruction set executable by the industrial controller to ensure the spatiotemporal coordination of cooling capacity distribution and equipment action.

[0060] For example, 6. Dynamic control verification instance:

[0061] 6.1 Thermodynamic constraint verification equation:

[0062]

[0063] 6.2 Device response timing instruction example:

[0064]

[0065] In this implementation scheme, a refrigerant allocation priority matrix is ​​constructed: In this step, the refrigerant allocation priority matrix needs to be constructed based on the adjustment range of the refrigerant branch flow rate and the edge weight parameters in the heat flow propagation topology. Refrigerant branch flow rate adjustment range: This parameter indicates the adjustable range of each refrigerant branch flow rate in the system. In a burst load scenario, it may be necessary to adjust the refrigerant flow rate in real time to cope with temperature fluctuations based on changes in load distribution. Edge weight parameters of the heat flow propagation topology: In the heat flow propagation topology, the edge weight represents the intensity or efficiency of heat flow transfer from one node (such as a cabinet) to another node. Paths with larger edge weight values ​​represent stronger heat flow transfer, and refrigerant needs to be allocated preferentially for adjustment. Generate device response timing instructions: The generation process of the device response timing instructions is based on the control delay characteristics and start-stop inertia parameters of the device to ensure that the timing of the device response conforms to the dynamic changes of the system. Control delay characteristics: After receiving the control instructions, devices such as water pumps, fans and valves will have a certain response delay. The delay characteristics need to be taken into account to ensure the match between the instructions and the actual device responses. Start-stop inertia parameters: The start-stop of the equipment has inertia, especially high-power equipment (water pumps, fans), which will have a certain lag time when starting and stopping. It is necessary to add a certain degree of foresight to the timing instructions to balance the working state of the equipment. The generation of the equipment response timing instructions can be expressed as follows through the optimization problem: ; : Equipment control instructions at time g (water pump frequency, fan speed) : The target value of the control command is calculated based on real-time load, temperature changes and refrigerant allocation priority : Equipment control delay or start-stop inertia delay : Encoding the current state (working state) of the equipment at the gth moment. Timing logic instruction set: Encode the logic generated by the prior matrix and the equipment response timing instruction into a timing instruction set executable by the industrial controller. Time-space coordination: Ensure the synchronization of refrigerant distribution and equipment action in space and time. That is, the distribution of cooling capacity between different refrigerant branches and cabinets should be coordinated with the control instructions of the equipment (such as water pumps, fans, etc.) to avoid overload of equipment or uneven distribution of cooling capacity. The control logic coding can be expressed as follows: ; : The control instruction set at time f includes the refrigerant distribution priority matrix and equipment control instructions. : Refrigerant distribution priority between the bth refrigerant branch and the cth cabinet. : The device responds to the instruction at time f (such as the frequency of the water pump, etc.). Finally, the timing instruction set It will be executed through industrial controllers to ensure that refrigerant distribution and equipment operations are highly coordinated in time and space.

[0066] Specifically, the specific process of comparing the actual temperature distribution with the simulation data through the digital twin platform and calculating the cooling distribution deviation rate and energy consumption overflow is as follows: align the actual temperature distribution data with the simulated temperature field of the digital twin in time and space, and extract the temperature gradient distribution differences in key areas; identify abnormal temperature aggregation areas based on the adaptive clustering algorithm, and calculate the matching degree between the cooling distribution deviation rate and the expected refrigerant flow rate; synchronize the real-time power data of the equipment energy consumption monitoring system, compare the simulated expected energy consumption curve, and quantify the energy consumption overflow of the cooling tower, water pump and valve.

[0067] For example, 7. Digital twin correction example

[0068] 7.1 Analysis of cooling capacity distribution deviation:

[0069]

[0070] 7.2 Edge weight correction gradient calculation:

[0071]

[0072] 7.3 Comparison of energy consumption overflow:

[0073]

[0074] In this implementation scheme, the actual temperature distribution and simulation data are aligned in time and space: In this step, the temperature distribution data of the actual computer room needs to be compared with the temperature field calculated by the digital twin simulation, and then aligned in time and space. Actual temperature distribution data: real-time temperature data from various sensors in the computer room. Digital twin simulation temperature field: predicted temperature distribution data obtained by simulation calculation through the digital twin platform. The key to time and space alignment is to ensure that the two temperature field data can correspond to the same spatial position and time point to ensure the accuracy of data comparison. Time and space alignment can be achieved by interpolation or data pairing. Extract the temperature gradient distribution difference in key areas: By comparing the actual temperature distribution and the simulated temperature distribution, extract the areas with poor cooling effect and calculate their temperature gradient distribution differences. Temperature gradient: represents the rate of temperature change, usually calculated based on spatial coordinates. A large temperature gradient usually means uneven distribution of cooling or poor cooling effect in local areas. Identify abnormal temperature clustering areas based on adaptive clustering algorithms: Adaptive clustering algorithms can be used to identify and group abnormal areas in temperature distribution, such as places with excessively high temperatures, which usually require more refrigerant flow to compensate. Adaptive clustering algorithm: Classify areas based on temperature gradient differences and identify areas with large temperature differences (usually temperature anomaly areas), which may be hot spots of uneven cooling distribution. The goal of this process is to discover temperature anomaly clusters and calculate their cooling distribution deviation rates. Calculate the matching degree of cooling distribution deviation rate with expected refrigerant flow rate Based on the identified abnormal areas, calculate the cooling distribution deviation rate and compare the actual refrigerant flow rate with the expected refrigerant flow rate. Cooling distribution deviation rate: used to measure the accuracy of refrigerant distribution. Evaluate the effectiveness of cooling distribution by calculating the difference between actual cooling and expected cooling. The cooling distribution deviation rate formula can be expressed as: ; : Cooling capacity distribution deviation rate. : The actual cooling capacity allocated to zone e. :Assume the expected cooling capacity allocated to area e. Synchronize equipment energy consumption monitoring and quantify the energy overflow of cooling towers, water pumps and valves. Finally, the system needs to obtain the power consumption of equipment (such as cooling towers, water pumps, valves, etc.) in real time through the data of the synchronized energy consumption monitoring system, and compare it with the energy consumption curve expected by simulation to calculate the energy overflow. Energy overflow: refers to the part where the actual energy consumption exceeds the expected energy consumption.

[0075] Specifically, the specific process of reversely correcting the edge weight parameters of the heat flow propagation model is as follows: construct a corrected gradient function of the edge weights of the heat flow propagation model according to the cold distribution deviation rate and the energy consumption overflow; update the edge weight parameters of the graph neural network through a dynamic learning rate adjustment mechanism combined with the convergence trend of historical correction data; input the corrected heat flow propagation model into the digital twin platform for verification and iteration until the temperature gradient distribution difference is lower than the preset convergence threshold.

[0076] In this implementation scheme, in this process, based on the cold capacity allocation deviation rate and energy consumption overflow, the edge weights in the heat flow propagation model are gradually corrected, thereby improving the accuracy of the model and matching the temperature gradient distribution with the actual computer room data. The specific steps are as follows: Calculation of cold capacity allocation deviation rate and energy consumption overflow: First, by comparing the actual cold capacity allocation and simulation data, the deviation rate of cold capacity allocation is calculated, and combined with the real-time energy consumption monitoring data of the equipment, the energy consumption overflow of the cooling tower, water pump and valve is quantified. Construct a corrected gradient function: Based on the cold capacity deviation rate and energy consumption overflow calculated above, a corrected gradient function is established to measure the amplitude of the weight adjustment on each edge (the air flow path connecting the cabinets) in the heat flow propagation model. The calculation of the corrected gradient not only considers the current temperature difference, but also considers the overall energy efficiency and stability requirements of the system. Dynamic learning rate adjustment mechanism: In order to ensure that the weight adjustment is both efficient and stable in each iteration, a dynamic learning rate strategy is adopted, combined with the convergence trend of the historical correction data, to adjust the step size of each correction. Specifically, if the difference in system correction becomes smaller, the learning rate is gradually reduced to avoid over-adjustment; if the correction difference is large, the learning rate is appropriately increased to accelerate convergence. Update the edge weight parameters of the graph neural network: Use the calculated correction gradient to update the weight parameters of each edge in the graph neural network. These edge weights determine the intensity of heat flux propagation, thereby affecting the distribution of the temperature field and refrigerant flow velocity field in the computer room. Model verification and iteration: The corrected heat flux propagation model will be input into the digital twin platform for verification. By comparing the output of the model with the actual computer room temperature gradient distribution data, check whether the Navier-Stokes equations and energy conservation constraints are met. If the temperature gradient distribution difference exceeds the preset convergence threshold, continue the iterative correction process until the accuracy requirements are met. Assume that the edge weight of the heat flux propagation model is , where p and q represent two nodes in the graph, Represents the heat flow transmission intensity between nodes p and q. ΔC is used to represent the correction gradient. The correction gradient function constructed based on the cooling deviation rate and energy consumption overflow can be expressed as: ;in: : represents the edge weight between node p and node q, that is, the intensity of heat flow propagation; : Dynamically adjusted learning rate; : The gradient of the current edge weight to the loss function C; : Correction item of cooling capacity distribution deviation rate; : Correction term for energy consumption overflow; t: represents the current number of iterations; Through this formula, the corrected edge weight Gradual adjustments will be made based on the errors in cooling capacity and energy consumption until the difference between the system's temperature gradient distribution and the actual data is lower than the preset convergence threshold.

[0077] In summary, this application has at least the following effects:

[0078] The temperature control system used for water-cooled air conditioners in computer rooms includes four modules: coupling analysis, collaborative control, dynamic control, and correction. The system collects temperature distribution and power consumption data in the computer room, builds a heat flow propagation model, and generates optimized control instructions through a non-dominated sorting genetic algorithm to achieve coordinated control of water pumps, fans, and valves. At the same time, the digital twin platform compares actual and simulated temperature data, calculates the cooling capacity distribution deviation rate, and performs model correction to ensure that the system maintains a stable cooling effect while optimizing energy efficiency and extending equipment life. The system improves the accuracy and efficiency of air conditioning control and meets the needs of intelligent management of computer room environments.

[0079] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0080] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0081] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0082] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0083] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0084] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. The temperature control system used for water-cooled air conditioners in computer rooms is characterized by: It includes the following modules: coupling analysis module, collaborative control module, dynamic regulation module, and correction module; The coupling analysis module is used to collect the temperature distribution data of the server cabinet, obtain the cabinet power consumption data in combination with the current sensor, build a heat flow propagation model of the computer room through the graph neural network, and output the refrigerant branch flow rate setting value and the terminal outlet temperature setting value; The collaborative control module is used to receive the refrigerant branch flow rate setting value and the terminal outlet air temperature setting value, and combines the water pump efficiency curve, the cooling tower heat dissipation coefficient and the compressor health index, with the total system energy consumption, equipment life balance and refrigeration stability as the optimization goals, and generates the water pump inverter frequency control instruction, the cooling tower fan speed instruction and the chilled water valve opening instruction through the non-dominated sorting genetic algorithm; The dynamic control module is used to verify whether the instructions meet the constraints of the thermodynamic equation according to the spatial layout characteristics of the computer room and the control instructions, generate the refrigerant allocation priority matrix and the equipment response timing instructions, and send them to the water pump inverter, cooling tower fan and chilled water valve controller through the industrial communication protocol; The correction module is used to compare the actual temperature distribution with the simulation data through the digital twin platform, calculate the cooling capacity distribution deviation rate and energy consumption overflow, and reversely correct the edge weight parameters of the heat flow propagation model; The specific process of generating the water pump inverter frequency control command, cooling tower fan speed command and chilled water valve opening command through the non-dominated sorting genetic algorithm is as follows: Encode the pump frequency, fan speed and valve opening parameters into chromosome vectors in the multi-objective solution space; Based on the Pareto dominance relationship and reference point screening strategy, the non-dominated solution set with the lowest energy consumption, the most balanced life loss and the highest refrigeration stability is iteratively calculated; Decode the chromosome vector in the Pareto optimal solution set to generate the water pump inverter frequency value, cooling tower fan speed value and chilled water valve opening gradient value compatible with the industrial controller interface; According to the spatial layout characteristics of the computer room and the control instructions, the specific process of verifying whether the instructions meet the constraints of the thermodynamic equations is as follows: Load the structural characteristic parameters of the three-dimensional layout of the computer room, including the cabinet spacing, air supply channel distribution, and refrigerant branch topology connection relationship; Through the forward propagation calculation of the physical information neural network, the temperature field distribution and refrigerant flow rate field distribution of the computer room under the control instructions are predicted; The prediction results are input into the embedded thermodynamic equation solver to verify whether the simplified form of the Navier-Stokes equation and the energy conservation equation constraints are met, and the candidate instruction sets that conform to the laws of physics are screened; The specific process of generating the refrigerant allocation priority matrix and the equipment response timing instructions is as follows: According to the refrigerant branch flow rate adjustment range of the candidate instruction set and the edge weight parameter of the heat flow propagation topology graph, a refrigerant allocation priority matrix under the burst load scenario is constructed; Based on the control delay characteristics of water pumps, fans and valves and the start-stop inertia parameters of equipment, a sequence of equipment command for peak-shifting response in the time dimension is generated; The priority matrix and instruction sequence are encoded into a sequential logic instruction set executable by the industrial controller to ensure the spatiotemporal coordination of cooling capacity distribution and equipment action.

2. The temperature control system for water-cooled air conditioners in computer rooms according to claim 1 is characterized in that: The specific process of constructing the heat flow propagation model of the computer room through the graph neural network is as follows: The server cabinets are abstracted as nodes in the graph structure, and the physical connection channels and airflow paths between cabinets are defined as edges; The temperature distribution data and the cabinet power consumption data are encoded into the temperature feature vector and the power consumption feature vector of the node respectively; Based on the node feature aggregation mechanism of graph neural network, edge weight parameters are updated through multi-layer message passing to characterize the intensity of heat flow propagation; Integrate external environment temperature and humidity data, dynamically correct the attenuation coefficient of the heat flow path, and generate a three-dimensional heat flow propagation topology map of the computer room.

3. The temperature control system for water-cooled air conditioners in computer rooms according to claim 2 is characterized in that: The specific process of outputting the refrigerant branch flow rate threshold and the terminal outlet air temperature setting value is as follows: According to the edge weight parameters of each branch in the three-dimensional heat flow propagation topology diagram of the computer room, the priority level of the refrigerant branch is divided; According to the inverse proportional relationship between the priority level and the heat flow transmission intensity, the adjustment range of the refrigerant flow rate of each branch is dynamically calculated; Combined with the real-time monitoring data of the terminal air outlet temperature sensor, the air outlet temperature setting value is iteratively corrected through the thermodynamic equilibrium equation to ensure dynamic matching of refrigerant supply and heat load.

4. The temperature control system for water-cooled air conditioners in computer rooms according to claim 3 is characterized in that: Combining the water pump efficiency curve, cooling tower heat dissipation coefficient and compressor health index, the specific process of optimizing the total system energy consumption, equipment life balance and refrigeration stability is as follows: Extract the optimal working range in the water pump efficiency curve and establish a nonlinear mapping relationship between energy consumption and refrigerant flow; The dynamic change characteristics of the cooling tower heat dissipation coefficient and the degradation trend of the compressor health index are integrated to construct the equipment life loss balance evaluation function; Set multi-objective optimization constraints, including refrigerant flow rate threshold, terminal air outlet temperature deviation limit and refrigeration stability error range.

5. The temperature control system for water-cooled air conditioners in computer rooms according to claim 4 is characterized in that: The specific process of comparing the actual temperature distribution with the simulation data through the digital twin platform and calculating the cooling capacity allocation deviation rate and energy consumption overflow is as follows: Align the actual temperature distribution data with the simulated temperature field of the digital twin in time and space to extract the temperature gradient distribution differences in key areas; Based on the adaptive clustering algorithm, the abnormal temperature clustering area is identified and the matching degree between the cooling capacity distribution deviation rate and the expected refrigerant flow rate is calculated; Synchronize the real-time power data of the equipment energy consumption monitoring system, compare the simulated expected energy consumption curve, and quantify the energy consumption overflow of cooling towers, water pumps and valves.

6. The temperature control system for water-cooled air conditioners in computer rooms according to claim 5 is characterized in that: The specific process of reversely correcting the edge weight parameters of the heat flow propagation model is as follows: According to the cooling capacity allocation deviation rate and energy consumption overflow, a modified gradient function of the edge weight of the heat flow propagation model is constructed; Through the dynamic learning rate adjustment mechanism, combined with the convergence trend of historical correction data, the edge weight parameters of the graph neural network are updated; The modified heat flow propagation model is input into the digital twin platform for verification and iteration until the temperature gradient distribution difference is lower than the preset convergence threshold.

Images