Method for energy consumption control of a data center cooling system

Abstract

The application provides an energy consumption control method of a data center cooling system, adopts an LSTM network with an attention mechanism to accurately predict future load and climate, uses a recursive least square method to update second-order RC thermal dynamic model parameters online, so that the control model always fits the real dynamic characteristics of the system; with the minimum total energy consumption of the cooling system as the target, an improved particle swarm algorithm with tabu search and Gaussian mutation is used to solve the optimal control sequence under the temperature safety constraint, effectively avoiding local optimum and realizing smooth switching of the control quantity; the penalty coefficient is dynamically adjusted according to the actual temperature deviation, ensuring the dynamic balance of thermal safety and energy efficiency. Through the above mode, the application can flexibly cope with IT load fluctuation and outdoor climate dynamic change, significantly reduce the comprehensive operation energy consumption of the primary side water pump, the secondary side water pump and the cooling tower fan, and at the same time, strictly guarantee that the server inlet temperature does not exceed the limit, and has strong robustness and adaptability.

Description

Technical Field

[0001] This invention relates to the field of data center air conditioning technology, and in particular to an energy consumption control method for data center cooling systems. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence technologies, the scale and computing density of global data centers continue to climb, making data centers a core infrastructure supporting the development of the digital economy in modern society. The increasing power consumption of IT equipment within data centers leads to a sharp increase in heat generation, placing higher demands on cooling systems. Liquid cooling technology, due to its efficient heat transfer capabilities, is gradually becoming the mainstream cooling solution for data centers. In a typical liquid-cooled data center cooling system, the main energy-consuming equipment includes chillers, primary-side cooling water pumps, secondary-side chilled water pumps, and cooling tower fans. The energy consumption of pumps and fans typically accounts for a large proportion of the total energy consumption of the cooling system. How to accurately and efficiently control the energy consumption of these power devices is a key technical challenge for reducing the overall operating costs of data centers and improving energy efficiency.

[0003] Various energy consumption control methods for data center cooling systems have been proposed in existing technologies. For example, some solutions employ rule-based PID control or constant temperature difference control strategies, adjusting pump frequency by maintaining a constant temperature difference between chilled water supply and return water or cooling water. However, these methods lag in response to load fluctuations and outdoor climate changes, and cannot globally optimize the coupling relationships between multiple devices. Other solutions attempt to establish steady-state energy consumption models based on physical mechanisms and use genetic algorithms or particle swarm optimization algorithms for optimization. However, these methods typically rely on fixed model parameters identified offline, and cannot adapt to the drift of data center thermal dynamics caused by seasonal changes, equipment aging, or changes in business deployment. Furthermore, some existing technologies use black-box models such as neural networks for load or energy consumption prediction, but they lack physical constraints. When the system operating state exceeds the range of training data, the prediction accuracy drops significantly, and it is difficult to interpret and guarantee temperature safety.

[0004] Therefore, existing control methods generally suffer from poor adaptability, insufficient optimization, and difficulty in dynamically balancing thermal safety and energy consumption control. There is an urgent need for an intelligent energy consumption control method that can identify the dynamic characteristics of the system online, proactively handle load changes and thermal constraints, and achieve collaborative optimization among multiple devices. Summary of the Invention

[0005] The purpose of this invention is to provide an energy consumption control method for a data center cooling system to solve the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] This invention provides an energy consumption control method for a data center cooling system, the data center cooling system including a cooling tower, a primary side water pump, a secondary side water pump, and a server, the method comprising:

[0008] The system acquires the current operating status data and attribute data of the data center cooling system. The operating status data includes at least one of the following: current outdoor temperature, current outdoor humidity, and server load power. The attribute data includes at least one of the following: specific heat capacity of the primary fluid, specific heat capacity of the secondary fluid, server safe temperature, minimum outlet water temperature of the cooling tower, reference value of water pump energy consumption power, reference value of water pump flow rate at reference power, reference value of fan energy consumption power, and reference value of fan flow rate at reference power.

[0009] The operating status data and attribute data are input into the energy consumption optimization model based on model predictive control for processing to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump, and the third target flow rate of the cooling tower fan in the cooling tower. The energy consumption optimization model based on model predictive control includes a heat load prediction unit, a system identification unit, a rolling optimization unit, and a performance evaluation unit. The rolling optimization unit aims to minimize the total energy consumption of the cooling system, solves for the optimal control sequence in the future control time domain under the condition of satisfying temperature safety constraints, and extracts the control quantity of the first control step as the target adjustment parameter.

[0010] Adjust the current flow rate of the primary water pump to the first target flow rate, adjust the current flow rate of the secondary water pump to the second target flow rate, and adjust the current flow rate of the cooling tower fan to the third target flow rate.

[0011] Preferably, the step of inputting the operating status data and the attribute data into an energy consumption optimization model based on model predictive control for processing to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump, and the third target flow rate of the cooling tower fan in the cooling tower includes:

[0012] The operating status data is input into the heat load prediction unit, and a long short-term memory network with an attention mechanism is used to predict the server load power and outdoor dry-bulb temperature in the future prediction time domain.

[0013] The operating status data is input to the system identification unit, and the parameters of the data center thermal dynamic model are identified online using the recursive least squares method; the thermal dynamic model is a second-order RC lumped parameter model, and its state variables include the indoor average temperature and the equivalent temperature of the building envelope;

[0014] The rolling optimization unit constructs an optimization problem based on the predicted server load power, outdoor dry-bulb temperature, and online identified thermal dynamic model parameters. The problem aims to minimize the total energy consumption of the cooling system and is constrained by ensuring that the indoor temperature does not exceed the safety limit. An improved particle swarm optimization algorithm with tabu search and Gaussian mutation is used to solve the problem and obtain the optimal control sequence in the future control time domain.

[0015] The performance evaluation unit dynamically adjusts the temperature penalty coefficient in the objective function of the rolling optimization unit based on the deviation between the actual measured temperature and the predicted temperature, and triggers the system identification unit to re-identify parameters when the deviation continues to exceed the threshold.

[0016] Preferably, the objective function of the optimization problem constructed by the rolling optimization unit is:

[0017] ;

[0018] in, To predict the length of the time domain, To control the length of the time domain, For the predicted total power of the cooling system, This is the temperature penalty coefficient. For the predicted average indoor temperature, To the upper limit of temperature safety, To control the smoothing coefficient, The control vector includes chilled water set temperature, cooling water set temperature, water pump frequency, and fan speed.

[0019] Preferably, the second-order RC lumped-parameter thermal dynamic model used by the system identification unit is:

[0020] ;

[0021] ;

[0022] in, The average indoor temperature. The equivalent temperature of the building envelope. , The heat capacity of indoor air and building envelope. , Thermal resistance from the interior to the building envelope and from the building envelope to the exterior. The heat transfer coefficient is the cooling coefficient. for Total power of equipment Outdoor dry-bulb temperature, For cooling water flow rate, The supply air temperature; the system identification unit uses the recursive least squares method to estimate the parameter vector online.θ =[ , , , , ] T And set the forgetting factor λ=0.98.

[0023] Preferably, the position update formula of the improved particle swarm algorithm is:

[0024] ;

[0025] ;

[0026] in, For inertial weights, The speed decreased linearly from 0.9 to 0.4, which is the speed of the previous generation. , The learning factor has a value of 1.5. , A random number within the range [0,1] For the first The individual best position in the history of each particle. For the first The current position vector of each particle. This is the globally optimal position for the entire population. The updated particle position; the improvement also includes: setting a tabu list, whereby if a particle has not updated its optimal position for 5 consecutive generations, the particle position is subjected to Gaussian mutation with a 20% probability:

[0027] ;

[0028] in, , These are the upper and lower bounds for the decision variables.

[0029] Preferably, the performance evaluation unit dynamically adjusts the temperature penalty coefficient based on the actual temperature deviation. The method is as follows:

[0030] ;

[0031] in, The deviation between the actual temperature and the predicted temperature. To adjust the step size coefficient, a value of 0.1 is used. Set the temperature value; when |

[0032] When the temperature exceeds 0.5℃ and the duration exceeds two control cycles, the system identification unit is triggered to re-identify the parameters.

[0033] Preferably, in the optimal control sequence obtained by the rolling optimization unit, except for the control quantity of the first control step used for execution in the current cycle, the control quantities of the remaining control steps are cached as the initial solution for rolling optimization in the next control cycle.

[0034] The present invention also provides an energy consumption control system for a data center cooling system, the system comprising:

[0035] The data acquisition unit is used to acquire the current operating status data and attribute data of the data center cooling system; the operating status data includes at least one of the current outdoor temperature, current outdoor humidity, and server load power; the attribute data includes at least one of the following: specific heat capacity of the primary fluid, specific heat capacity of the secondary fluid, server safe temperature, minimum outlet water temperature of the cooling tower, reference value of water pump energy consumption power, reference value of water pump flow rate at reference power, reference value of fan energy consumption power, and reference value of fan flow rate at reference power.

[0036] A heat load prediction unit, connected to the data acquisition unit, is used to predict the server load power and outdoor dry-bulb temperature in the future time domain based on the operating status data.

[0037] The system identification unit is connected to the data acquisition unit and is used to identify the thermal dynamic model parameters of the data center online.

[0038] The rolling optimization unit, connected to the heat load prediction unit and the system identification unit, is used to solve the optimal control sequence in the future control time domain with the goal of minimizing the total energy consumption of the cooling system and satisfying the temperature safety constraints, and extract the control quantity of the first control step as the target adjustment parameter to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump and the third target flow rate of the cooling tower fan in the cooling tower.

[0039] A control execution unit, connected to the rolling optimization unit, is used to adjust the current flow rate of the primary side water pump to the first target flow rate, adjust the current flow rate of the secondary side water pump to the second target flow rate, and adjust the current flow rate of the cooling tower fan to the third target flow rate;

[0040] The performance evaluation unit, connected to the data acquisition unit, the rolling optimization unit, and the system identification unit, is used to dynamically adjust the temperature penalty coefficient in the objective function of the rolling optimization unit according to the deviation between the actual measured temperature and the predicted temperature, and to trigger the system identification unit to perform parameter re-identification when the deviation continues to exceed a threshold.

[0041] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of an energy consumption control method for a data center cooling system.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an energy consumption control method for a data center cooling system.

[0043] The present invention achieves the following beneficial technical effects compared to the prior art:

[0044] This invention provides an energy consumption control method for a data center cooling system. By constructing a model predictive control framework comprising a heat load prediction unit, a system identification unit, a rolling optimization unit, and a performance evaluation unit, it achieves closed-loop dynamic optimization control of the data center cooling system's energy consumption. Specifically, the heat load prediction unit uses an attention-based LSTM network to accurately predict future load and climate conditions; the system identification unit uses recursive least squares to update the parameters of the second-order RC thermal dynamic model online, ensuring the control model always closely matches the system's actual dynamic characteristics; the rolling optimization unit aims to minimize the total energy consumption of the cooling system and, under temperature safety constraints, uses an improved particle swarm optimization algorithm with tabu search and Gaussian mutation to solve for the optimal control sequence, effectively avoiding local optima and achieving smooth switching of control variables; the performance evaluation unit dynamically adjusts the penalty coefficient based on actual temperature deviations, ensuring a dynamic balance between thermal safety and energy efficiency. Through the above methods, the present invention can flexibly respond to IT load fluctuations and dynamic changes in outdoor climate, significantly reduce the overall operating energy consumption of primary side water pumps, secondary side water pumps and cooling tower fans, and strictly ensure that the server intake air temperature does not exceed the limit. It has strong robustness and adaptability, and provides a reliable technical solution for solving the problems of high energy consumption and difficult control of data center cooling systems. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart of the energy consumption control method for a data center cooling system provided by the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The purpose of this invention is to provide an energy consumption control method for a data center cooling system to solve the problems existing in the prior art.

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Example 1:

[0051] like Figure 1 As shown, this embodiment provides an energy consumption control method for a data center cooling system. This method can be applied to systems such as… Figure 1 The computer equipment shown is connected to a data center cooling system, which includes a cooling tower, primary-side water pumps, secondary-side water pumps, plate heat exchangers, and servers. The computer equipment is used to regulate the flow rates of the primary-side water pumps, secondary-side water pumps, and cooling tower fans to minimize the total energy consumption of the cooling system.

[0052] The method in this embodiment first performs a system initialization step. The control period is set to 5 minutes, the prediction time domain length is 12 (corresponding to the next hour), and the control time domain length is 6 (corresponding to the next 30 minutes). The upper limit of temperature safety is set to 27 degrees Celsius, the initial value of the temperature penalty coefficient is set to 100 based on experience, and the control smoothing coefficient is set to 0.5. The pre-trained attention mechanism long short-term memory network model weights are loaded, and the recursive least squares identification parameters are initialized, where the forgetting factor is set to 0.98, and the initial covariance matrix is ​​set to 10 to the power of 5 multiplied by the identity matrix.

[0053] The data acquisition process then begins. Computer equipment acquires the current outdoor dry-bulb temperature and relative humidity using temperature and humidity sensors deployed outside the data center. It also monitors server load power in real time using a pre-defined status query function and obtains the current flow rates of the primary and secondary water pumps using flow sensors. Simultaneously, the computer equipment retrieves attribute data from equipment parameter manuals or online databases, including the specific heat capacity of the primary and secondary fluids, server safe temperature, minimum cooling tower outlet temperature, reference values ​​for water pump energy consumption power, reference values ​​for water pump flow rates at the reference power, reference values ​​for fan energy consumption power, and reference values ​​for fan flow rates at the reference power.

[0054] After acquiring operational status and attribute data, the process proceeds to the heat load prediction step. The computer equipment inputs the server load power and outdoor dry-bulb temperature sequence from the past hour into an attention-based Long Short-Term Memory (LSTM) network, which outputs the future... The network generates predictions in steps, namely a sequence of server load power predictions and a sequence of outdoor dry-bulb temperature predictions, with each prediction step corresponding to a control cycle. By weighting historical time-series information through an attention mechanism, the network can accurately capture the long-term and short-term dependencies of the load.

[0055] Simultaneously, the system identification unit updates the data center thermal dynamic model parameters online. This embodiment uses a second-order RC lumped-parameter thermal model, whose differential equation form is:

[0056] ;

[0057] ;

[0058] in, The average indoor temperature. The equivalent temperature of the building envelope. , The heat capacity of indoor air and building envelope. , Thermal resistance from the interior to the building envelope and from the building envelope to the exterior. The heat transfer coefficient is the cooling coefficient. for Total power of equipment Outdoor dry-bulb temperature, For cooling water flow rate, Let be the supply air temperature. Discretize the model with a sampling period to obtain the state equation:

[0059] ;

[0060] The system identification unit uses real-time data from the past hour to estimate the parameter vector online using the recursive least squares method. θ =[ , , , , ] T The forgetting factor is set to 0.98 to ensure the ability to track time-varying parameters.

[0061] After prediction and identification are completed, a rolling optimization unit is constructed and the optimization problem is solved. The calculation of the total power of the cooling system in the prediction time domain integrates the models of each component. The chiller unit power model is as follows: ,in The cooling load is approximately equal to the server load power plus the building envelope load. and These are the cooling water outlet temperature and the cooling water inlet temperature, respectively. The water pump power model is as follows: ,and It is proportional to the frequency. The total power model of a precision air conditioner fan is as follows: The cooling tower fan power model is .

[0062] The objective function of the rolling optimization unit is set as follows:

[0063] ;

[0064] in, To predict the length of the time domain, To control the length of the time domain, For the predicted total power of the cooling system, This is the temperature penalty coefficient. For the predicted average indoor temperature, To the upper limit of temperature safety, To control the smoothing coefficient, The control vector includes the chilled water setpoint temperature, cooling water setpoint temperature, pump frequency, and fan speed. The objective function's first term is the sum of the total cooling system power at each step within the prediction time domain; the second term is a temperature over-limit penalty term, applied only when the predicted indoor temperature exceeds the safe upper limit; and the third term is a control action smoothing term, used to suppress drastic fluctuations in control variables between adjacent control cycles. Penalty coefficient. Used to balance the weights between energy consumption optimization and thermal safety, controlling the smoothing coefficient. This results in a smoother change in the control sequence, which is beneficial to actuator lifespan and equipment stability. The decision variable is the future. The control vector for each step includes chilled water set temperature, cooling water set temperature, water pump frequency, and fan speed, and each needs to meet its own upper and lower limit constraints.

[0065] To solve the above optimization problem, this embodiment employs an improved particle swarm optimization algorithm. First, 30 particles are initialized, with each particle's position vector representing a set of candidate solutions for the control sequence, and its velocity vector randomly initialized. The maximum number of iterations is set to 50. The inertia weight is linearly decreased from 0.9 to 0.4, and both the individual learning factor and the social learning factor are set to 1.5. The velocity update formula is:

[0066] ;

[0067] The position update formula is:

[0068] ;

[0069] in, For inertial weights, The speed decreased linearly from 0.9 to 0.4, which is the speed of the previous generation. , The learning factor has a value of 1.5. , A random number within the range [0,1] For the first The individual best position in the history of each particle. For the first The current position vector of each particle. This is the globally optimal position for the entire population. This represents the updated particle position. To enhance global search capabilities and avoid getting trapped in local optima, this embodiment introduces a tabu search mechanism and a Gaussian mutation strategy. Specifically, a tabu list is set up to record recently visited better solutions. When a particle fails to update its individual optimal position for 5 consecutive generations, a Gaussian mutation is applied to the particle's position with a 20% probability. The mutation formula is:

[0070] ;

[0071] in, , The upper and lower bounds of the decision variables are defined by a Gaussian distribution with a mean of 0 and a variance of . This mutation helps particles escape local optima, increasing the diversity of solutions.

[0072] By iteratively solving the improved particle swarm optimization algorithm, the optimal control sequence in the future control time domain is obtained. The optimal control vector to be executed in the current cycle includes the first target flow rate of the primary pump, the second target flow rate of the secondary pump, and the third target flow rate of the cooling tower fan. The remaining control sequences are cached as the initial solution for hot-start optimization in the next control cycle, thus significantly reducing online computation time.

[0073] Next, the control execution step begins. The computer equipment sends the first target flow rate to the inverter of the primary water pump via industrial Ethernet or fieldbus protocol, adjusting the pump frequency to the target value; it then sends the second target flow rate to the inverter of the secondary water pump; and the third target flow rate to the inverter of the cooling tower fan. Each actuator adjusts its operating status according to the received setpoints, thus realizing the issuance of energy consumption control commands for the cooling system.

[0074] At the end of each control cycle, the performance evaluation unit collects actual measurement data, including actual total cooling power and actual room temperature, and calculates the prediction error. If the absolute value of the temperature prediction error is | If the temperature deviation is greater than 0.5 degrees Celsius and lasts for more than two control cycles, a re-identification trigger signal is sent to the system identification unit. The system identification unit will then perform a complete recursive least squares re-estimation of all parameters θ, enabling the model to quickly adapt to the current thermal dynamics. Simultaneously, the performance evaluation unit dynamically adjusts the penalty coefficient based on the actual temperature deviation. The calculation formula is:

[0075] ;

[0076] in Take 0.1, Set the temperature to a value such as 24 degrees Celsius. When the actual temperature is higher than the predicted temperature, the exponent term is greater than 1. Increasing the temperature penalty strengthens the overall solution, forcing it to become more conservative; when the actual temperature is lower than the predicted temperature... This reduces energy consumption while allowing for a suitable increase in energy-saving potential. This adaptive mechanism achieves a dynamic balance between thermal security and energy consumption.

[0077] Finally, the computer equipment updates the current time, returns to the data acquisition step, and enters the next control cycle, forming a closed-loop rolling optimization control. The entire control process runs continuously online. As the outdoor environment, server load, and system thermal characteristics change, the various units of model predictive control work together to adjust the cooling system operating parameters in real time.

[0078] The energy consumption control method for data center cooling systems provided in this embodiment integrates physical and data-driven approaches. It utilizes an attention-based long short-term memory network for time-series prediction, employs recursive least squares to update the second-order RC thermal dynamic model online, and utilizes an improved particle swarm optimization algorithm with tabu search and Gaussian mutation to solve constrained nonlinear optimization problems. Furthermore, a performance evaluation unit is introduced to achieve adaptive penalty coefficient adjustment, significantly improving the accuracy, robustness, and adaptability of cooling system energy consumption control. Compared to existing technologies, this method effectively copes with drastic fluctuations in data center load and dynamic changes in outdoor climate. While ensuring server temperature safety, it minimizes the combined energy consumption of primary-side water pumps, secondary-side water pumps, and cooling tower fans, making it suitable for energy-saving operation control of various liquid-cooled data center cooling systems.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] It should be noted that the components mentioned in the above embodiments are all general standard parts or components known to those skilled in the art. Their structures and principles can be learned by those skilled in the art through technical manuals or conventional experimental methods.

[0081] This invention has illustrated its principles and implementation methods using specific examples. The descriptions of these embodiments are merely illustrative of the method and its core ideas; furthermore, those skilled in the art will recognize that modifications may be made to the specific implementation methods and application scope based on the principles of this invention. Therefore, the content of this specification should not be construed as limiting the invention.

Claims

1. An energy consumption control method for a data center cooling system, characterized in that, The data center cooling system includes a cooling tower, a primary water pump, a secondary water pump, and servers. The method includes: The system acquires the current operating status data and attribute data of the data center cooling system. The operating status data includes at least one of the following: current outdoor temperature, current outdoor humidity, and server load power. The attribute data includes at least one of the following: specific heat capacity of the primary fluid, specific heat capacity of the secondary fluid, server safe temperature, minimum outlet water temperature of the cooling tower, reference value of water pump energy consumption power, reference value of water pump flow rate at reference power, reference value of fan energy consumption power, and reference value of fan flow rate at reference power. The operating status data and attribute data are input into the energy consumption optimization model based on model predictive control for processing to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump, and the third target flow rate of the cooling tower fan in the cooling tower. The energy consumption optimization model based on model predictive control includes a heat load prediction unit, a system identification unit, a rolling optimization unit, and a performance evaluation unit. The rolling optimization unit aims to minimize the total energy consumption of the cooling system, solves for the optimal control sequence in the future control time domain under the condition of satisfying temperature safety constraints, and extracts the control quantity of the first control step as the target adjustment parameter. Adjust the current flow rate of the primary water pump to the first target flow rate, adjust the current flow rate of the secondary water pump to the second target flow rate, and adjust the current flow rate of the cooling tower fan to the third target flow rate.

2. The energy consumption control method for a data center cooling system according to claim 1, characterized in that, The step of inputting the operating status data and the attribute data into the energy consumption optimization model based on model predictive control for processing, to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump, and the third target flow rate of the cooling tower fan in the cooling tower, includes: The operating status data is input into the heat load prediction unit, and a long short-term memory network with an attention mechanism is used to predict the server load power and outdoor dry-bulb temperature in the future prediction time domain. The operating status data is input to the system identification unit, and the parameters of the data center thermal dynamic model are identified online using the recursive least squares method; the thermal dynamic model is a second-order RC lumped parameter model, and its state variables include the indoor average temperature and the equivalent temperature of the building envelope; The rolling optimization unit constructs an optimization problem based on the predicted server load power, outdoor dry-bulb temperature, and online identified thermal dynamic model parameters. The problem aims to minimize the total energy consumption of the cooling system and is constrained by ensuring that the indoor temperature does not exceed the safety limit. An improved particle swarm optimization algorithm with tabu search and Gaussian mutation is used to solve the problem and obtain the optimal control sequence in the future control time domain. The performance evaluation unit dynamically adjusts the temperature penalty coefficient in the objective function of the rolling optimization unit based on the deviation between the actual measured temperature and the predicted temperature, and triggers the system identification unit to re-identify parameters when the deviation continues to exceed the threshold.

3. The energy consumption control method for a data center cooling system according to claim 2, characterized in that, The objective function of the optimization problem constructed by the rolling optimization unit is: ； in, To predict the length of the time domain, To control the length of the time domain, For the predicted total power of the cooling system, This is the temperature penalty coefficient. For the predicted average indoor temperature, To the upper limit of temperature safety, To control the smoothing coefficient, The control vector includes chilled water set temperature, cooling water set temperature, water pump frequency, and fan speed.

4. The energy consumption control method for a data center cooling system according to claim 2, characterized in that, The second-order RC lumped-parameter thermal dynamic model used by the system identification unit is as follows: ； ； in, The average indoor temperature. The equivalent temperature of the building envelope. , The heat capacity of indoor air and building envelope. , Thermal resistance from the interior to the building envelope and from the building envelope to the exterior. The heat transfer coefficient is the cooling coefficient. for Total power of equipment Outdoor dry-bulb temperature, For cooling water flow rate, The supply air temperature; the system identification unit uses the recursive least squares method to estimate the parameter vector online. θ =[ , , , , ] T And set the forgetting factor λ=0.

98.

5. The energy consumption control method for a data center cooling system according to claim 2, characterized in that, The position update formula for the improved particle swarm algorithm is: ； ； in, For inertial weights, The speed decreased linearly from 0.9 to 0.4, which is the speed of the previous generation. , The learning factor has a value of 1.

5. , A random number within the range [0,1] For the first The individual best position in the history of each particle. For the first The current position vector of each particle. This is the globally optimal position for the entire population. The updated particle position; the improvement also includes: setting a tabu list, whereby if a particle has not updated its optimal position for 5 consecutive generations, the particle position is subjected to Gaussian mutation with a 20% probability: ； in, , These are the upper and lower bounds for the decision variables.

6. The energy consumption control method for a data center cooling system according to claim 2, characterized in that, The performance evaluation unit dynamically adjusts the temperature penalty coefficient based on the actual temperature deviation. The method is as follows: ； in, This represents the deviation between the actual and predicted temperatures, and is the adjustment step size coefficient, with a value of 0.

1. Set the temperature value; when | When the temperature exceeds 0.5℃ and the duration exceeds two control cycles, the system identification unit is triggered to re-identify the parameters.

7. The energy consumption control method for a data center cooling system according to claim 1, characterized in that, In the optimal control sequence obtained by the rolling optimization unit, except for the control quantity of the first control step used for execution in the current cycle, the control quantities of the remaining control steps are cached as the initial solution for the rolling optimization of the next control cycle.

8. An energy consumption control system for a data center cooling system, characterized in that, The system includes: The data acquisition unit is used to acquire the current operating status data and attribute data of the data center cooling system; the operating status data includes at least one of the current outdoor temperature, current outdoor humidity, and server load power; the attribute data includes at least one of the following: specific heat capacity of the primary fluid, specific heat capacity of the secondary fluid, server safe temperature, minimum outlet water temperature of the cooling tower, reference value of water pump energy consumption power, reference value of water pump flow rate at reference power, reference value of fan energy consumption power, and reference value of fan flow rate at reference power. A heat load prediction unit, connected to the data acquisition unit, is used to predict the server load power and outdoor dry-bulb temperature in the future time domain based on the operating status data. The system identification unit is connected to the data acquisition unit and is used to identify the thermal dynamic model parameters of the data center online. The rolling optimization unit, connected to the heat load prediction unit and the system identification unit, is used to solve the optimal control sequence in the future control time domain with the goal of minimizing the total energy consumption of the cooling system and satisfying the temperature safety constraints, and extract the control quantity of the first control step as the target adjustment parameter to obtain the first target flow rate of the primary side water pump, the second target flow rate of the secondary side water pump and the third target flow rate of the cooling tower fan in the cooling tower. A control execution unit, connected to the rolling optimization unit, is used to adjust the current flow rate of the primary side water pump to the first target flow rate, adjust the current flow rate of the secondary side water pump to the second target flow rate, and adjust the current flow rate of the cooling tower fan to the third target flow rate; The performance evaluation unit, connected to the data acquisition unit, the rolling optimization unit, and the system identification unit, is used to dynamically adjust the temperature penalty coefficient in the objective function of the rolling optimization unit according to the deviation between the actual measured temperature and the predicted temperature, and to trigger the system identification unit to perform parameter re-identification when the deviation continues to exceed a threshold.

9. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the energy consumption control method for the data center cooling system according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the energy consumption control method for the data center cooling system according to any one of claims 1 to 7.

Images