Management of power distribution to and from data centers

Abstract

This paper discloses a system and method for providing consistent and reliable power to data centers running high-power and variable high-performance computing (HPC) and artificial intelligence (AI) workloads. [Solution] By installing three-phase current sensors in or near the data center or along the power grid before the power reaches the power conversion system (PCS), the power conversion system is configured to respond more quickly to rapidly and irregularly increasing and decreasing workloads. Subsequently, the controller within the PCS is configured to customize the amount of power supplied to the data center from converted power from the power grid and converted power from the battery energy storage system.

Description

Technical Field

[0001] Reference to Related Applications This U.S. Provisional Patent Application claims the benefit and priority of U.S. Provisional Patent Application No. 63 / 719332, filed Nov. 12, 2024, the entire content of which is incorporated herein by reference.

[0002] Technical Field The present disclosure relates to a system for stabilizing power to a data center that is executing a large-scale and variable workload.

Background Art

[0003] Background The rapid development of advanced computing systems driven by high-performance computing (HPC), artificial intelligence (AI), and other advanced computing technologies has changed the power demand patterns of data centers that host computing devices capable of executing these HPC and AI workloads. The amount of energy used to perform simultaneous and more complex operations rapidly (e.g., hundreds of times per minute) changes the power demand of such data centers, creating a "pulse energy" demand. Power providers, such as the public power grid, do not actually have the scale to accommodate the total amount of power required for this type of operation and cannot provide the pulse energy required by these new loads. This situation can lead to instability of the power grid and cause power quality problems, and may even cause more serious problems such as data center providers deploying large-scale HPC and AI workloads being "expelled" from the power grid.

Summary of the Invention

Means for Solving the Problems

[0004] Summary Aspects of the disclosed embodiments include a system for managing high-power and variable workloads in a data center. The system comprises a three-phase current sensor configured to provide current measurements to a power conversion system, the three-phase current sensor being located in close proximity to the data center or along a power grid supplying power to the data center. The power conversion system of the data center includes a controller. The power conversion system is configured to control the rate of power supplied to the data center from the power grid and from a battery energy storage system connected to the power conversion system. To control the rate of power supply, the controller receives an indication of the difference between the current measured in the data center and the current measured in the power grid, at least in part on current measurements provided by the three-phase current sensor, and is configured to adjust one or both of the rate of power supply from the power grid and the rate of power supply from the battery energy storage system, at least in part on the indication of the difference, to reduce the difference to near zero.

[0005] Another aspect of the disclosed embodiments includes a method for managing power to a data center. This method includes receiving a first three-phase current from a power conversion system connected to a battery energy storage system; receiving a second three-phase current from the data center; applying a phase-locked loop to measure the frequency and phase angle of a three-phase voltage from a power grid for a given phase; outputting a first equivalent phase current by performing a first Park-Clark conversion on the first three-phase current and the phase angle of the first three-phase voltage; outputting a second equivalent phase current by performing a second Park-Clark conversion on the second three-phase current and the phase angle of the three-phase voltage; outputting the difference between the three equivalent phase currents by comparing the first and second equivalent phase currents with an average equivalent phase current from the data center; and generating a fourth equivalent phase current that reduces the difference to near zero by outputting a pulse-width modulation command to the power conversion system.

[0006] This disclosure is best understood by reading the following detailed description in conjunction with the attached drawings. Note that, as is customary, various features in the drawings are not depicted to scale. Rather, the dimensions of various features in the drawings have been arbitrarily enlarged or reduced for clarity. [Brief explanation of the drawing]

[0007] [Figure 1] This figure shows a controller within a power conversion system (PCS) configured to manage power entering and leaving a data center from the power grid, at least in part, based on three-phase current sensors located in close proximity to the data center side or the power grid side of the integrated system, according to several embodiments. [Figure 2] This figure shows a visual representation of voltage and current measurements used to manage power entering and leaving a data center in an integrated system, as shown in Figure 1, according to several embodiments. [Figure 3] This figure shows a feedback control loop corresponding to the management of power entering and leaving a data center in an integrated system, as shown in Figure 1, according to several embodiments. [Figure 4A] This figure shows an example of D-phase current measured in a data center, according to several embodiments, from a time corresponding to the start of the AI ​​workload in the data center through the subsequent period managing power between the power grid and the data center. [Figure 4B] This figure shows examples of D-phase current measured on the grid side, according to several embodiments, from a time corresponding to the start of the AI ​​workload in the data center through the subsequent period during which power is managed between the grid and the data center. [Figure 4C] This figure shows an example of D-phase current measured by the PCS, according to several embodiments, from a time corresponding to the start of the AI ​​workload in the data center through the subsequent period managing power between the power grid and the data center. [Figure 5] This flowchart shows the control algorithms executed by the controller within the PCS according to several embodiments. [Figure 6A] This figure shows a current-to-time simulation over a period of exemplary AI workload being run in a data center, according to several embodiments. [Figure 6B] This figure shows a simulation of current versus time measured by a PCS during an exemplary AI workload being run in a data center, according to several embodiments. [Figure 6C] This figure shows input current versus time over a period of exemplary AI workload being performed in a data center, according to several embodiments. [Figure 6D] This figure shows the battery voltage and power measured in BESS during an exemplary AI workload being run in a data center, according to several embodiments. [Figure 6E] This figure shows the battery current measured in BESS during an exemplary AI workload being run in a data center, according to several embodiments. [Figure 6F] This figure shows the current-to-time ratio of the data center AI load current, the grid-side current, and the current measured by the PCS during an exemplary AI workload being performed in a data center, according to several embodiments. [Figure 7] This figure shows another example of a controller within a power conversion system (PCS) configured to manage power flowing in and out of a data center's server racks from a data center's low-voltage AC distribution unit, at least in part on three-phase current sensors located in close proximity to the server racks or low-voltage AC distribution units of the integrated system, according to some embodiments. [Modes for carrying out the invention]

[0008] Detailed explanation The embodiments shown in the attached drawings will be described in detail below. In the attached drawings, similar reference numerals indicate similar elements. Exemplary embodiments may be of different types and should not be construed as being limited to those described herein.

[0009] It should be understood that the terms “includes,” “contains,” “equip,” and / or “equip” as used herein identify the presence of the described features, integers, steps, actions, elements, and / or components, and do not exclude the presence or addition of one or more other features, integers, steps, actions, elements, components, and / or groups thereof.

[0010] Furthermore, it should be understood that, in this specification, terms such as “first,” “second,” and “third” may be used to describe various elements, components, areas, layers, and / or sections, but these elements, components, areas, layers, and / or sections may not be limited by these terms. These terms are used solely to distinguish a particular element, component, area, layer, or section from another element, component, area, layer, or section.

[0011] Furthermore, it should be understood that terms such as “optimize,” “optimal,” and “approximately” as used herein can be used to mean achieving or realizing the most effective or perfect performance possible. However, as any person skilled in the art who reads this specification will understand, perfection is not always achievable. Therefore, these terms can also encompass achieving or realizing the best possible or effective performance in a particular situation, or achieving or realizing performance that is better than what could be achieved using other settings or parameters, or achieving or realizing performance that is better than what would be achieved if the operations, calculations, etc. described herein were not performed.

[0012] As used herein, the terms "and / or" include any and all combinations of one or more related enumeration items. Expressions such as "at least one" preceding a list of elements qualify all elements of the list, not individual elements of the list.

[0013] Various terms are used to refer to specific components of the system. Different terms may refer to components with different names, and this specification is not intended to distinguish components that have the same function even though they have different names.

[0014] Matters that are obvious to those skilled in the art to which these exemplary embodiments belong may not be described in detail herein.

[0015] It should be understood that the exemplary embodiments described herein are to be considered in an illustrative sense only and not for purposes of limitation. The description of features or aspects in each exemplary embodiment may be considered applicable to other similar features or aspects in other exemplary embodiments.

[0016] As described above, computer servers, computing devices, processors, and support devices used for HPC and AI workload processing always require high power demand and variable power demand. Meeting both the high power demand and variable power demand of the data center can be difficult for both the power grid operator side and the data center provider side.

[0017] The present disclosure overcomes this problem by incorporating additional three-phase current sensor measurements into the control algorithm executed by the controller within the PCS. By implementing a three-phase current sensor that periodically provides current measurements from the power grid side or the data center side to the controller within the PCS, the PCS is configured to respond more quickly to the rapidly and highly variable power demands of the HPC and AI workloads being executed in the data center. By incorporating additional three-phase current sensor measurements into the control algorithm managed by the controller within the PCS, the power consumption in the data center is better adapted to the HPC and AI workloads. For example, the systems and methods described herein take into account peak shaving (also referred to as load limiting), frequency stabilization, and the ability to sell excess power back to the power grid.

[0018] Figure 1 shows a controller within a PCS configured to manage power flowing in and out of a data center from the power grid, at least in part, based on three-phase current sensors located in close proximity to either the data center side or the power grid side of the integrated system, according to several embodiments.

[0019] In the integrated system 100 shown in Figure 1, the power grid 102 is configured to supply power to the data center 106. As shown in Figure 1, since the input power from the power grid 102 is typically high-voltage AC power (e.g., 34.5kV), the PCS 108 may be located close to the data center 106 and configured to convert the high-voltage AC power to lower-voltage AC power and further to DC power. Thus, the battery energy storage system (BESS) 110 is connected to the power grid 102 via the PCS 108, and the PCS 108 manages the power supplied from the power grid.

[0020] To improve the response time of the PCS108 and BESS110 when providing high-power and variable-power demands for the data center 106, a three-phase current sensor 104 can be placed on the data center side or the power grid side, as shown in a specific embodiment in Figure 1. In some embodiments, the three-phase current sensor 104 is configured to measure power consumption at the sensor location and provide the measurement value to the controller of the PCS108. An example of the controller of the PCS108 is shown using controller 316 in Figure 3.

[0021] As will be further described below, current measurements from the three-phase current sensor 104 and another current sensor in the PCS 108 may be provided as input to the control algorithm. This control algorithm is configured to rapidly detect changes in the workload in the data center 106 and then respond to the changes in the workload by supplying power to or absorbing power from the power grid. In some embodiments, this detection and response method allows the power grid 102 to supply constant or near-constant power to the data center 106 even if a given AI workload fluctuates between 30% and 100% per second.

[0022] The integrated system 100 is further configured, for example, to track AC voltages from the power grid 102 using a phase-locked loop (PLL), and to convert AC voltages and AC currents into equivalent DC signals using a Park-Clark transformer. Such conversion from AC signals to equivalent DC signals is expressed by equation f qdo =K s ·f abc It is described as follows. In the formula, K s This is expressed by the following formula:

[0023]

number

[0024] The angle θ in the above equation can be derived from the PLL and is equivalent to the instantaneous angle of the power grid voltage.

[0025] Figure 2 shows a visual representation of voltage and current measurements used to manage power entering and leaving a data center of an integrated system as shown in Figure 1, according to several embodiments.

[0026] The integrated system 200 is similar to the integrated system 100 described above. To facilitate the explanation in this specification, and in particular to facilitate the explanation of the feedback control loop shown in Figure 3 below, the figure shows arrows indicating "three-phase power grid voltage," "three-phase PCS current," "BESS voltage," "BESS current," and "three-phase AI data center current." Note that the power grid voltage of the power grid 102 may be measured within the PCS 108 in some embodiments, but for ease of explanation it is shown as shown in the figure.

[0027] Figure 3 shows a feedback control loop corresponding to the management of power entering and leaving a data center of an integrated system as shown in Figure 1, according to several embodiments.

[0028] In some embodiments, the components of the feedback control loop 300 correspond to both the hardware components shown in Figures 1 and 2 and the sensor signals visually shown in Figure 2. For example, the grid voltage 302 points to the arrow labeled "Three-phase grid voltage" in Figure 2, and therefore to the three-phase voltage of the input AC power from the grid 102. Similarly, the PCS current 306 points to the arrow labeled "Three-phase PCS current" in Figure 2, and the AI ​​data center current 308 points to the arrow labeled "Three-phase AI data center current" in Figure 2.

[0029] As shown in Figure 3, the grid voltage 302, also referred to as the three-phase ABC voltage in the drawing, is supplied to the PLL 304, which rapidly and accurately measures the grid voltage frequency and phase angle for a predetermined phase, also referred to as the grid voltage phase angle in the drawing. In some embodiments, phase A is selected as the reference phase, but any phase may be used as the reference phase. In other embodiments, multiple phases may be used as reference phases.

[0030] Next, the PCS current 306, also called the three-phase ABC current in the diagram, is supplied as input to the Park-Clark converter 310 along with the grid voltage phase angle, and the output of the Park-Clark converter 310 is the converted equivalent D-phase current.

[0031] Next, the AI ​​data center current 308, also called the three-phase ABC current in the diagram, is supplied as input to the Park-Clark converter 312 along with the grid voltage phase angle, and the output of the Park-Clark converter 312 is the converted equivalent D-phase current.

[0032] According to some embodiments, when an instantaneous grid voltage phase angle is applied, both D-phase currents become DC signals. Also, for the sake of convenience of this specification, we assume that the values ​​of phase Q and phase D are zero because the voltage and current are balanced and do not have harmonic components. However, it should be understood that the implementation form of the control feedback loop 300 includes additional feedback loops corresponding to these phases to adjust these phases to specified values.

[0033] Continuing the description of the control feedback loop 300, the D-phase currents of the PCS and AI data centers are compared to another signal labeled “Grid Target D-Phase Current” in the diagram, in a step shown as comparison (Σ) 314. The “Grid Target D-Phase Current” signal can be defined as the average value of the AI ​​data center D-phase current plus or minus some value, thereby ensuring that the battery in BESS 110 is neither completely discharged nor overcharged.

[0034] The output of the comparison (Σ) 314 is the difference between each signal, i.e., the "error". This error is provided to the controller 316, for example, a proportional-integral-derivative (PID) controller. The controller 316 is configured to generate pulse-width modulation (PWM) commands. The generated PWM commands are provided to the PCS 318. The PCS 318 generates a current that reduces the error to zero.

[0035] The PCS D-phase current is compared with the AI ​​data center D-phase current to generate a difference (error). This error is provided, for example, to a proportional-integral-derivative (PID) controller. Controller 316 generates a PWM command. The generated PWM command is provided to PCS318. PCS318 generates a current that reduces the error to zero.

[0036] Figures 4A, 4B, and 4C show examples of D-phase current measured at the data center 106, the power grid, and the PCS 108, respectively, from a time corresponding to the start of the AI ​​workload at the data center 106 through the subsequent period of managing power between the power grid 102 and the data center 106, according to several embodiments. Figures 4A, 4B, and 4C should be understood within the context of the operation of the controller 316.

[0037] Each of the three D-phase currents shown in Figures 4A, 4B, and 4C can be assumed to start from zero at time t=0. Alternatively, time t=0 may refer to the point in time when the HPC or AI workload in data center 106 has risen to a certain value. Subsequently, as shown in Figure 4B, the grid current surges to the same value to support the power demand of the new HPC or AI workload. Controller 316 measures the difference, or “error,” shown in Figure 3, and then increases the current level, drawing power from BESS 110 until it reaches the same value as the HPC or AI workload.

[0038] Subsequently, at time t=t1, the HPC or AI workload drops to zero, and the current from BESS110 flows back into the grid 102, as indicated by the negative grid current shown in Figure 4B. The controller 316 then rapidly increases to a negative value close to the average value of the last AI load pulse. The grid current reverses direction, as shown in Figure 4B, and settles as a positive value close to the average value of the last AI load pulse.

[0039] At time t=t2, the AI ​​workload increases, and the controller 316 rapidly reverses the current direction. As a result, the power grid 102 continues to provide average power, while the BESS 110 provides peak power for the AI ​​workload.

[0040] Subsequently, as shown at time t=t3, this process can be repeated so that the power grid 102 supplies average power to the data center 106, the BESS 110 supplies peak power, and the average current is drawn from the power grid to recharge the battery in the BESS 110. The increase and decrease changes after time t=t3 in Figure 4B indicate the response time of the controller 316.

[0041] Figure 5 is a flowchart showing control algorithms executed by the controller in the PCS108 according to several embodiments.

[0042] Process 500 can be executed using both the hardware components and the feedback control loop architecture described above with reference to Figures 1-4C. As shown in Figure 5, in order to obtain the grid target D-phase current shown in block 514, first the average AI current is calculated in block 502. Next, the BESS charge state (SOC) is calculated in block 504. Then, if the battery is discharged, as shown by the diamond-shaped block 506, the controller increases the grid average current by a certain amount (+K), as shown by block 508. If the battery is overcharged, the controller decreases the grid average current by a certain amount (-K), as shown by block 510.

[0043] In other embodiments, the feedback controller can slowly adjust the average target current over a period of time to balance the battery state of charge (SOC).

[0044] In some embodiments, the target D-phase current of the power grid is defined as the average value plus an increment (±K). The calculation 512 for determining the average D-phase current is described as follows.

[0045]

number

[0046] In the formula, T min This is the time when the current HPC or AI workload running in the data center is at a low level. max This represents the time when the current HPC or AI workload running in the data center is at its peak.

[0047] In other embodiments, the average AI phase D current can be obtained by applying other methods, such as analog filtering or digital filtering.

[0048] Figure 6A shows a current-to-time simulation over a period of exemplary AI workload being run in a data center, according to several embodiments.

[0049] Figure 6B shows a simulation of current versus time measured by the PCS during an exemplary AI workload being run in a data center, according to several embodiments.

[0050] Figure 6C shows the input current versus time over a period of exemplary AI workload being performed in a data center, according to several embodiments.

[0051] Figure 6D shows the battery voltage and power measured in the BESS during an exemplary AI workload being run in a data center, according to several embodiments.

[0052] Figure 6E shows the battery current measured in BESS during an exemplary AI workload being run in a data center, according to several embodiments.

[0053] Figure 6F shows the current-to-time ratio of the data center AI load current, the grid-side current, and the current measured by the PCS during an exemplary AI workload being performed in a data center, according to several embodiments.

[0054] Figure 7 shows another example of a controller within a PCS 708 configured to manage power flowing in and out of a data center's server rack 706 from a data center's low-voltage AC distribution unit, at least in part, based on a three-phase current sensor 704 located in close proximity to the server rack 706 or low-voltage AC distribution unit of the integrated system 700, according to several embodiments. An example of a controller in the PCS 708 is shown using controller 316 in Figure 3.

[0055] In such embodiments, a similar current measurement method using a three-phase current sensor 704 located in close proximity to the server rack 706 or the low-voltage AC power distribution unit of the data center can be incorporated into a feedback control loop similar to that shown in Figure 3. The electrical arrangement of the PCS 708 and the energy storage device 710 can be modified accordingly.

[0056] Furthermore, the energy storage device 710 of the integrated system 700 can refer to any energy storage device capable of handling low-voltage AC power distribution, such as electrolytic capacitors, supercapacitors, and lithium-ion batteries.

[0057] While embodiments of the present invention have been described with reference to the drawings, those skilled in the art will understand that various modifications can be made to the form and details without departing from the idea and scope set forth in the claims.

Claims

1. A system for managing power to a data center, wherein the system is A three-phase current sensor is provided, configured to provide current measurements to a power conversion system, and the three-phase current sensor is located in close proximity to the data center. The power conversion system of the data center includes a controller, The power conversion system is configured to control the rate of power supplied to the data center from the power grid and from a battery energy storage system connected to the power conversion system. In order to control the power supply rate, the controller, Based at least in part on the current measurement values ​​provided by the three-phase current sensor, an indicator value of the difference between the current measured at the data center and the current measured at the power grid is received. A system configured to reduce the difference to approximately zero by adjusting one or both of the power supply rate from the power grid and the power supply rate from the battery energy storage system, based at least in part on the indicated value of the difference.

2. The system according to claim 1, wherein the controller is a proportional-integral-derivative controller.

3. The system according to claim 1, wherein the controller is further configured to calculate the charge state of the battery energy storage system.

4. The aforementioned controller, Based on the calculated charge state, it is determined that the battery energy storage system is being discharged. The system according to claim 3, further configured to increase the power supply rate from the battery energy storage system and to rebalance the charge state of the battery energy storage system.

5. The aforementioned controller, Based on the calculated charge state, it is determined that the battery energy storage system is overcharged. The system according to claim 3, further configured to reduce the power supply rate from the battery energy storage system and to rebalance the charge state of the battery energy storage system.

6. The battery energy storage system is the system according to claim 1, wherein the battery energy storage system includes an electrolytic capacitor capable of handling low-voltage alternating current (AC) power distribution.

7. The battery energy storage system according to claim 1, further comprising a supercapacitor capable of handling low-voltage alternating current (AC) power distribution.

8. The battery energy storage system according to claim 1, comprising a lithium-ion battery capable of handling low-voltage alternating current (AC) power distribution.

9. A system for managing power to a data center, wherein the system is A three-phase current sensor is provided, configured to provide current measurements to a power conversion system, and the three-phase current sensor is located along a power grid supplying power to the data center. The power conversion system of the data center includes a controller, The power conversion system is configured to control the rate of power supplied to the data center from the power grid and from a battery energy storage system connected to the power conversion system. In order to control the power supply rate, the controller, Based at least in part on the current measurement values ​​provided by the three-phase current sensor, an indicator value of the difference between the current measured at the data center and the current measured at the power grid is received. A system configured to reduce the difference to approximately zero by adjusting one or both of the power supply rate from the power grid and the power supply rate from the battery energy storage system, based at least in part on the indicated value of the difference.

10. The system according to claim 9, wherein the controller is a proportional-integral-derivative controller.

11. The system according to claim 9, wherein the controller is further configured to calculate the charge state of the battery energy storage system.

12. The aforementioned controller, Based on the calculated charge state, it is determined that the battery energy storage system is being discharged. The system according to claim 11, further configured to increase the power supply rate from the battery energy storage system and to rebalance the charge state of the battery energy storage system.

13. The aforementioned controller, Based on the calculated charge state, it is determined that the battery energy storage system is overcharged. The system according to claim 11, further configured to reduce the power supply rate from the battery energy storage system and to rebalance the charge state of the battery energy storage system.

14. The battery energy storage system according to claim 9, further comprising an electrolytic capacitor capable of handling low-voltage alternating current (AC) power distribution.

15. The battery energy storage system according to claim 9, wherein the battery energy storage system includes a supercapacitor capable of handling low-voltage alternating current (AC) power distribution.

16. The battery energy storage system according to claim 9, comprising a lithium-ion battery capable of handling low-voltage alternating current (AC) power distribution.

17. A method for managing power to a data center, wherein the method is Receiving three-phase voltage from the power grid, Receiving a first three-phase current from a power conversion system connected to a battery energy storage system, Receiving a second three-phase current from the data center, Applying a phase-locked loop, the frequency and phase angle of the three-phase voltage from the power grid are measured for a predetermined phase, By performing a first Park-Clark transformation on the phase angle of the first three-phase current and the first three-phase voltage, a first equivalent phase current is output. By performing a second Park-Clark transformation on the second three-phase current and the phase angle of the three-phase voltage, a second equivalent phase current is output. The difference between the three equivalent phase currents is output by comparing the first equivalent phase current and the second equivalent phase current with the average equivalent phase current from the data center. A method comprising generating a fourth equivalent phase current that reduces the difference to near zero by outputting a pulse width modulation command to the power conversion system.

18. The method according to claim 17, further comprising calculating the charge state of the battery energy storage system.

19. Based on the calculated charge state, the battery energy storage system is determined to be discharging, The method according to claim 17, further comprising increasing the power supply rate from the battery energy storage system and re-equalizing the charge state of the battery energy storage system.

20. Based on the calculated charge state, it is determined that the battery energy storage system is overcharged, The method according to claim 17, further comprising reducing the power supply rate from the battery energy storage system and re-equalizing the charge state of the battery energy storage system.

Images