A full-stack joint optimization system for computing power, electricity and heat in data centers

Abstract

The present invention discloses a full-stack joint optimization system for computing power, electricity and heat of a data center, including: a server chip power consumption model, a pump-driven two-phase cooling system model, an optimization scheduling controller and an anti-disturbance controller, wherein there is information interaction between the server chip power consumption model and the pump-driven two-phase cooling system model; the server chip power consumption model is optimized in real time by the optimization scheduling controller to achieve that the computing device in the server chip power consumption model is in a flexible temperature domain; the pump-driven two-phase cooling system model is anti-interference controlled by the anti-disturbance controller to achieve control of the cooling system state quantity in the pump-driven two-phase cooling system model. The present invention achieves global optimization of the system state through flexible control of temperature, thereby achieving that the computing device is always in the flexible temperature domain, that is, maintaining a stable temperature within a reasonable range, and achieving the goal of energy saving and efficiency improvement.

Description

Technical Field

[0001] The present invention belongs to the technical field of integrated energy systems and joint optimization control methods for data centers, and in particular relates to a full-stack joint optimization system for computing power, electricity and heat of a data center. Background Art

[0002] During the normal operation of the data center, when current passes through the chip inside the computing device, the chip resistor will generate heat. In addition, computing devices need to consume energy during operation, and the conversion of energy will cause part of the energy to be converted into heat. The size of this part is related to the load rate of the computing device. The accumulation of heat causes the chip temperature to continue to rise, so the data center needs cooling equipment to take away the generated heat and keep the chip temperature within a suitable range. However, due to the transfer of computing power, the heat generated by the chip will change dramatically, but due to the different heat generation mechanisms, the cooling system will take longer to generate enough cold to reduce the chip temperature. Excessive chip temperature will affect the performance of the computing device and even have the risk of data loss, seriously affecting the safe and stable operation of the data center.

[0003] The prior art discloses a flat control method and control system for an air-cooled data center, which obtains the real-time working data of each device in each grid space in the data center and compares and analyzes the preset target control data in each grid area space to obtain the cooling output adjustment data of each cooling device, and adjusts the working output of each cooling device accordingly, and at the same time coordinates the operation of the cooling devices in each grid area space in the data center and forms the coordinated scheduling data, which is finally remotely monitored and displayed. This method can realize the temperature control of the computing equipment in the data center and the coordinated management of the cooling equipment, but it does not take into account the impact of external interference on the system operation status during operation, and lacks the potential for optimizing the data center.

[0004] The prior art also discloses a data center cooling control method, a cooling system, and a data center, which obtains the temperature difference between the return air or the inlet air of the air supply device and the actual temperature of the liquid in the liquid inlet pipe. According to the temperature difference, the operation mode of the dual cold source cooling unit is selected, wherein the operation mode includes a mechanical refrigeration operation mode, a natural cooling operation mode, and a mixed refrigeration mode. The liquid inlet pipe is connected to the cooling device and the dual cold source cooling unit to realize the circulation of the coolant between the cooling device and the dual cold source cooling unit. The above method realizes the effective utilization of the natural cold source, thereby reducing energy consumption. This control method cannot achieve the optimal control of the data center as a whole, lacks the monitoring of the entire system status and the scheduling of computing resources, and still fails to couple the computing equipment and the cooling equipment.

[0005] In summary, the existing control strategy for data centers only considers the control and optimization of refrigeration equipment. It is impossible to coordinate and regulate the status and changes of computing equipment, and it is impossible to achieve coordinated control of the overall status of the data center and improve the ability to resist external interference, resulting in the problem of local optimization. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a full-stack joint optimization system for computing power, electricity and heat of a data center to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above objectives, the present invention provides a full-stack joint optimization system for computing power, electricity and heat of a data center, including:

[0008] A server chip power consumption model, a pump-driven two-phase cooling system model, an optimization scheduling controller, and an anti-disturbance controller, wherein there is information interaction between the server chip power consumption model and the pump-driven two-phase cooling system model;

[0009] The server chip power consumption model is optimized in real time by the optimization scheduling controller, so as to realize that the computing device in the server chip power consumption model is in a flexible temperature domain;

[0010] The anti-disturbance controller is used to perform anti-disturbance control on the pump-driven two-phase cooling system model, so as to control the cooling system state quantity in the pump-driven two-phase cooling system model.

[0011] Preferably, the process of performing real-time rolling optimization on the server chip power consumption model includes:

[0012] During the normal operation of the data center system, the constraint information of the system is obtained based on the information interaction relationship;

[0013] Based on the constraint information, the optimization objective function established by the system is solved by the optimization scheduling controller to obtain the optimization target state quantity, wherein the optimization scheduling controller is an MPC controller;

[0014] The server chip power consumption model is optimized in real time based on the optimization target state quantity.

[0015] Preferably, the constraint condition information includes: economic temperature operation domain, operation constraints, and power limit information.

[0016] Preferably, the optimization objective function includes: a function for minimizing energy consumption per unit computing power and a function for minimizing energy consumption of a cooling system.

[0017] Preferably, the optimization target state quantities include: ideal load distribution of the server chip power consumption model, cooling system state quantities of the pump-driven two-phase cooling system, wherein the cooling system state quantities include: ideal frequency of the working fluid pump, ideal fan speed, and ideal opening of the expansion valve.

[0018] Preferably, the process of performing anti-interference control on the pump-driven two-phase cooling system model by the anti-interference controller includes:

[0019] The pump-driven two-phase cooling system model is estimated in real time by the anti-disturbance controller to obtain system state information, wherein the anti-disturbance controller is an ADRC controller;

[0020] Based on the external estimation of the disturbance observer, the external disturbance information is obtained;

[0021] Combining the system state information with the external disturbance information to obtain a disturbance compensation term;

[0022] Based on the disturbance compensation term, the influence of external disturbance information on the system is offset, and the execution state quantity of the pump-driven two-phase cooling system model is obtained through feedback control.

[0023] Preferably, the process further includes adjusting the load distribution process in the server chip power consumption model:

[0024] Obtaining a load transfer signal, performing a space transfer operation based on the load transfer signal, distributing the computing task sequence to different servers, and obtaining a first optimization objective function;

[0025] Performing a time transfer operation to obtain analysis results and key information of the computing task, and obtaining a second optimization objective function based on the analysis results and key information;

[0026] Based on the first optimization objective function and the second optimization objective function, the load distribution in the server chip power consumption model is comprehensively optimized to obtain the best resource allocation and task completion strategy.

[0027] Preferably, the analysis results include: deferrable computing tasks and non-deferrable computing tasks.

[0028] Preferably, the first optimization objective function is a function of minimum power consumption per unit computing power, and the second optimization objective function is a function of user satisfaction.

[0029] The present invention also provides a readable storage medium, which includes a full-stack joint optimization system for computing power, electricity and heat of the data center.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] The present invention provides a full-stack joint optimization system for computing power, electricity and heat of a data center, comprising: a server chip power consumption model, a pump-driven two-phase cooling system model, an optimization scheduling controller and an anti-disturbance controller, wherein there is information interaction between the server chip power consumption model and the pump-driven two-phase cooling system model; the server chip power consumption model is optimized in real time by the optimization scheduling controller to achieve that the computing device in the server chip power consumption model is in a flexible temperature domain; the pump-driven two-phase cooling system model is anti-interference controlled by the anti-disturbance controller to achieve control of the cooling system state quantity in the pump-driven two-phase cooling system model.

[0032] The present invention achieves global optimization of the system state through flexible control of temperature, so that the computing device is always in a flexible temperature domain, that is, maintains a stable temperature within a reasonable range, achieving the goal of energy saving and efficiency improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 A schematic diagram of a full-stack joint optimization system according to an embodiment of the present invention;

[0035] Figure 2 A flowchart of a method for a normal operation phase of a data center according to an embodiment of the present invention;

[0036] Figure 3 This is a flow chart of a method for data center load transfer stage according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0038] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] Embodiment 1

[0040] like Figure 1 As shown, in this embodiment, a full-stack joint optimization system for computing power, electricity and heat of a data center is provided, including:

[0041] A server chip power consumption model, a pump-driven two-phase cooling system model, an optimization scheduling controller, and an anti-disturbance controller, wherein there is information interaction between the server chip power consumption model and the pump-driven two-phase cooling system model;

[0042] The server chip power consumption model is optimized in real time by the optimization scheduling controller, so as to realize that the computing device in the server chip power consumption model is in a flexible temperature domain;

[0043] The anti-disturbance controller is used to perform anti-disturbance control on the pump-driven two-phase cooling system model, so as to control the cooling system state quantity in the pump-driven two-phase cooling system model.

[0044] Specifically, the system can be divided into an upper server chip power consumption model and a lower pump-driven two-phase cooling system model. The upper server chip power consumption model and the lower pump-driven two-phase cooling system model in the system are each equipped with a corresponding controller to achieve system optimization and anti-interference functions. In the upper server chip power consumption model, the first optimization scheduling controller includes a first optimization scheduling module and a first control module, as well as a first solution module and a first objective function setting module. These modules work together to enable the system to adjust task allocation and resource utilization according to real-time requirements to improve system efficiency and performance. In the lower pump-driven two-phase cooling system model, the second optimization scheduling controller includes a second optimization scheduling module and a second control module, a second solution module and a second objective function setting module, and also includes a first anti-disturbance controller, a second anti-disturbance controller, and a third anti-disturbance controller, each of which includes components such as a tracking differentiator, a nonlinear control rate, and an extended state observer. These controllers and modules work together to ensure that the cooling system can respond quickly and maintain stable operation under different working conditions, while resisting external interference and improving system reliability and stability.

[0045] In order to achieve the best control effect, it is necessary to consider the characteristics of data center operation such as being affected by external factors and real-time performance. Therefore, when selecting a controller, this embodiment selects an MPC controller with real-time rolling optimization characteristics as the optimization scheduling controller for the upper-layer server chip power consumption model, and selects an ADRC controller with fast response, strong anti-interference ability, and good robustness as the anti-interference controller for the lower-layer pump-driven two-phase cooling system model. Such a combination can effectively optimize system operation and ensure the stability and performance of the data center.

[0046] The process of real-time rolling optimization of the server chip power consumption model includes: in the normal operation stage of the data center system, based on the information interaction relationship, obtaining the system constraint information; based on the constraint information, solving the optimization objective function established by the system through the optimization scheduling controller to obtain the optimization target state quantity, wherein the optimization scheduling controller is an MPC controller; based on the optimization target state quantity, real-time rolling optimization of the server chip power consumption model is performed.

[0047] Specifically, during the normal operation of the system and the load transfer stage, the MPC control method is used to control the system input, predict the system state, and optimize the system performance and efficiency. Specifically, since there is information interaction between the upper server state and the lower cooling system state between the first optimization scheduler and the second optimization scheduler, there is also information interaction between the server chip power consumption model and the pump-driven two-phase cooling system. The overall management platform extracts the constraints of the entire system from the system interaction, including economic temperature operation domain, operation constraints, power limit and other information, and passes the constraint information to the MPC controller. The MPC controller uses optimization methods such as gradient descent or quadratic programming to solve the optimization objective function of the system, and inputs the optimized system state input to the upper server chip power consumption model and the lower pump-driven two-phase cooling system respectively.

[0048] In this embodiment, Figure 2 As shown in the figure, at the initial time t0 of the normal operation stage, the overall management platform extracts the system constraints, such as the economic temperature operation domain, the operation constraints and power limits of the server chip power consumption model, and the operation constraints and power limits of the pump-driven two-phase cooling system, through the system interaction between the server chip model system and the pump-driven two-phase cooling system. These constraints are used as the constraint boundary conditions of the MPC controller optimization solution algorithm to solve the upper system with the lowest unit computing power energy consumption as the objective function and the lower system with the lowest cooling system energy consumption as the objective function. The ideal load distribution u of the upper server chip power consumption model is obtained respectively. 4d , and the ideal frequency u of the working fluid pump in the pump-driven two-phase cooling system 1d 、Ideal fan speed u 2d , ideal opening of expansion valve u 3d , taking this as the final optimization target state quantity, it is uploaded to the overall management platform, and through the overall management platform, the data is transmitted to the first control module of the first optimization scheduling controller and the second control module of the second optimization scheduling controller respectively, and the state information of the optimization target is transmitted to the next level for execution through these two control parts. For the upper-level server chip model, since it is not affected by external interference, it can directly receive and execute the ideal load distribution u transmitted by the first optimization scheduling controller 4d .

[0049] The process of performing anti-disturbance control on the pump-driven two-phase cooling system model through the anti-disturbance controller includes: performing real-time estimation on the pump-driven two-phase cooling system model through the anti-disturbance controller to obtain system state information, wherein the anti-disturbance controller is an ADRC controller; obtaining external disturbance information based on external estimation of a disturbance observer; combining the system state information with the external disturbance information to obtain a disturbance compensation term; based on the disturbance compensation term, offsetting the influence of the external disturbance information on the system, and obtaining the execution state quantity of the pump-driven two-phase cooling system model through feedback control.

[0050] Specifically, in order to achieve anti-interference effect on the uncertainty of outdoor environmental factors, this embodiment adopts three ADRC anti-interference controllers to achieve precise control of the optimized working fluid pump frequency, fan speed, and expansion valve opening.

[0051] For the lower pump-driven two-phase cooling system, its operating state is easily affected by external disturbances such as outdoor temperature and outdoor humidity. Therefore, when transmitting three ideal inputs, it is necessary to use the ADRC controller to eliminate the influence of external disturbances. Specifically, the ADRC controller uses the state observer to estimate the state of the system in real time. Including the estimation of system state variables and unknown disturbances. By combining the state information estimated by the state observer with the external disturbance information estimated by the disturbance observer, the ADRC controller can specifically design disturbance compensation items to offset the impact of external disturbances on the system. In this way, the ADRC controller realizes the anti-interference ability of external disturbances during the control process to ensure that the system can operate stably. Finally, the state quantity of the lower pump-driven two-phase cooling system is accurately calculated and solved again through feedback control, and the final working fluid pump frequency u1, fan theoretical speed u2 and expansion valve opening u3 are obtained. Such a design can effectively improve the stability and robustness of the pump-driven two-phase cooling system and ensure that the system can maintain a good operating state when the external environment changes. In this process, the system state quantity that the pump-driven two-phase system really needs to receive and execute is obtained.

[0052] Subsequently, after the server power consumption model and the pump-driven two-phase cooling system execute these received state quantities, the four actual state quantities U0, including real-time temperature, working fluid pump frequency, fan speed, and expansion valve opening, are measured by sensors. These actual state quantities are compared with the predicted state to achieve feedback correction of the system. At time t1, the overall management platform combines the actual state quantity at time t0 to predict the future state through the server power consumption model and the pump-driven two-phase cooling system model, and passes the information to the corresponding optimization objective function module. Through the MPC optimization solver, the ideal state quantity vector at time t1 is obtained. Subsequently, the above operations are repeated to achieve rolling optimization control and anti-interference control of the entire system. The stability and robustness of the system are guaranteed by the MPC controller and the ADRC controller, while ensuring that the system can maintain a good operating state under various external environmental changes.

[0053] The system also includes a process for adjusting the load distribution in the server chip power consumption model: obtaining a load transfer signal, performing a space transfer operation based on the load transfer signal, distributing a computing task sequence to different servers, and obtaining a first optimization objective function; performing a time transfer operation, obtaining analysis results and key information of the computing task, and obtaining a second optimization objective function based on the analysis results and key information; based on the first optimization objective function and the second optimization objective function, comprehensively optimizing the load distribution in the server chip power consumption model to obtain the best resource allocation and task completion strategy.

[0054] Specifically, during the normal operation of the system, the overall planning platform receives a signal that load transfer can be performed, and implements server load transfer by adjusting the load distribution of the server. Figure 3 As shown in the figure, in order to realize the load transfer of the server, the overall planning platform divides the load transfer into space transfer and time transfer. At time tk, after receiving the load transfer signal, the overall planning platform takes a series of actions to optimize the resource utilization and user satisfaction of the data center. The specific steps are as follows:

[0055] S1. First, the platform performs a spatial transfer operation to distribute the computing task sequence to different servers. The optimization objective function J1 is constructed to minimize the power consumption per unit computing power of the system to ensure optimal resource utilization.

[0056] S2. The platform performs time transfer operations, analyzes the characteristics of computing tasks and extracts key information, including the size of the task, the transfer time, and the delayable processing time (in units of 2 hours). Depending on whether the task can be delayed, the computing tasks are divided into two categories: delayable and non-delayable. Non-delayable tasks need to be calculated and processed immediately, while delayable tasks can flexibly adjust the computing power according to the electricity price of the day to control the progress of the task.

[0057] Specifically, for delayed computing tasks, the platform adjusts computing power to control task progress based on the electricity price of the day. During periods of high electricity prices, task progress is slowed down to reduce energy consumption and costs; while during periods of low electricity prices, task progress is accelerated to improve efficiency. By flexibly adjusting computing power, the platform can strike a balance between economy and performance.

[0058] S3. The platform combines the time and price required to complete the task to construct a user satisfaction function J2 as the optimization target. The user satisfaction function comprehensively considers the time and cost of task completion to ensure that users complete tasks within a reasonable time and reduce the cost burden as much as possible. Finally, all the above optimization targets are integrated into the final optimization target function J, and the optimization operation is performed to obtain the best resource allocation and task completion strategy.

[0059] Through the above logic and operations, the overall planning platform can make immediate responses at the moment of load transfer, and through comprehensive optimization, maximize the utilization of data center resources and optimize user satisfaction.

[0060] The overall optimization process includes the operation of computing load transfer to ensure balanced resource utilization and reasonable load distribution. Once the load transfer operation is completed, the next step is the normal operation phase, and the platform will continue to pay attention to and optimize the state variables of the underlying system. By dynamically optimizing the state variables of the underlying system, the platform can further improve the operating efficiency of the data center, reduce energy consumption, and ensure that the temperature is always in the flexible temperature range.

[0061] In this embodiment, after completing the overall optimization of computing load transfer, the overall planning platform will continue to optimize the underlying system state variables to further improve the performance and efficiency of the data center. In this way, all operations of load transfer are completed and the data center is ensured to maintain an efficient and stable state in actual operation.

[0062] Beneficial effects of this embodiment:

[0063] This embodiment proposes a data center upper computing equipment and lower cooling system optimization modeling method. By establishing a server power consumption model for computing equipment and a lower pump-driven two-phase cooling system model, the overall optimization objective function of the system is established, and by decoupling the strongly coupled temperature, an MPC controller is used to implement rolling optimization of the system state quantity, and an ADRC controller is used to eliminate the interference of external uncertain factors to achieve accurate control of the cooling system state quantity.

[0064] This embodiment proposes a method for solving the joint optimization of computing power, electricity, and heat, and realizes efficient management of the normal operation stage and the load transfer stage of the system through a comprehensive strategy. In the normal operation stage, the overall management platform obtains the optimization constraint boundary conditions by extracting the information of the information interaction between the server power consumption model and the pump-driven two-phase cooling system model, and then realizes real-time optimization of the system through MPC control. In the load transfer stage, the overall management platform analyzes the characteristics of each task in the task sequence to personalize the tasks. After completing the overall optimization of the server task processing, the normal operation stage operations are performed to finally complete the joint optimization scheduling of computing power, electricity, and heat.

[0065] Embodiment 2

[0066] This embodiment also provides a readable storage medium, which includes the full-stack joint optimization system for computing power, electricity, and heat of the data center described in Example 1.

[0067] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A full-stack joint optimization system for computing power, electricity and heat in a data center, characterized in that: include: A server chip power consumption model, a pump-driven two-phase cooling system model, an optimization scheduling controller and an anti-disturbance controller, wherein there is information interaction between the server chip power consumption model and the pump-driven two-phase cooling system model; The server chip power consumption model is optimized in real time by the optimization scheduling controller, so as to realize that the computing device in the server chip power consumption model is in a flexible temperature domain; The anti-disturbance controller performs anti-disturbance control on the pump-driven two-phase cooling system model to achieve control of the cooling system state quantity in the pump-driven two-phase cooling system model; The process of performing real-time rolling optimization on the server chip power consumption model includes: During the normal operation of the data center system, the constraint information of the system is obtained based on the information interaction relationship; Based on the constraint information, the optimization objective function established by the system is solved by the optimization scheduling controller to obtain the optimization target state quantity, wherein the optimization scheduling controller is an MPC controller; Performing real-time rolling optimization on the server chip power consumption model based on the optimization target state quantity; The process of performing anti-interference control on the pump-driven two-phase cooling system model by the anti-interference controller includes: The pump-driven two-phase cooling system model is estimated in real time by the anti-disturbance controller to obtain system state information, wherein the anti-disturbance controller is an ADRC controller; Based on the external estimation of the disturbance observer, the external disturbance information is obtained; Combining the system state information with the external disturbance information to obtain a disturbance compensation term; Based on the disturbance compensation term, the influence of external disturbance information on the system is offset, and the execution state quantity of the pump-driven two-phase cooling system model is obtained through feedback control; Adjust the load distribution process in the server chip power consumption model: Obtaining a load transfer signal, performing a space transfer operation based on the load transfer signal, distributing the computing task sequence to different servers, and obtaining a first optimization objective function; Performing a time transfer operation to obtain analysis results and key information of the computing task, and obtaining a second optimization objective function based on the analysis results and key information; Based on the first optimization objective function and the second optimization objective function, the load distribution in the server chip power consumption model is comprehensively optimized to obtain the best resource allocation and task completion strategy.

2. The full-stack joint optimization system for computing power, electricity and heat of a data center according to claim 1, characterized in that: The constraint condition information includes: economic temperature operation domain, operation constraints, and power limit information.

3. The full-stack joint optimization system for computing power, electricity and heat of a data center according to claim 1, characterized in that: The optimization objective functions include: a function for minimizing energy consumption per unit computing power and a function for minimizing energy consumption of a cooling system.

4. The full-stack joint optimization system for computing power, electricity and heat of a data center according to claim 1, characterized in that: The optimization target state quantities include: the ideal load distribution of the server chip power consumption model, the cooling system state quantities of the pump-driven two-phase cooling system, wherein the cooling system state quantities include: the ideal frequency of the working fluid pump, the ideal speed of the fan, and the ideal opening of the expansion valve.

5. The full-stack joint optimization system for computing power, electricity and heat of a data center according to claim 1, characterized in that: The analysis results include: deferrable computing tasks and non-deferrable computing tasks.

6. The full-stack joint optimization system for computing power, electricity and heat of a data center according to claim 1, characterized in that: The first optimization objective function is a function of minimum power consumption per unit computing power, and the second optimization objective function is a function of user satisfaction.

Images