Power supply system, power supply method, and entire cabinet
Through the coordinated work of the energy management unit, power supply unit, and peak power supply unit, the current is dynamically monitored and compensated, solving the problem of low power density of the entire AI cabinet and achieving efficient power distribution and stable equipment operation.
Patent Information
- Application Number
- CN202511067295.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2025-10-21
- Estimated Expiration
- 2045-07-31
AI Technical Summary
Existing AI cabinet power supply units are unable to meet power density requirements of several times the peak power within the same space constraints, resulting in low power density.
The energy management unit works in conjunction with multiple power supply units and peak power supply units to obtain the output current through the current state bus, dynamically monitor and trigger current compensation instructions when necessary, and use the peak power supply unit to meet the peak power requirements of the equipment.
It improves the power density and power supply reliability of the entire cabinet, and can provide sufficient compensation current when the power demand of the equipment increases sharply, solving the problem of low power density of the entire cabinet.
Smart Images

Figure CN120566437B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a power supply system, a power supply method, and a complete cabinet. Background Art
[0002] With the rapid development of artificial intelligence (AI) technology, demand for AI computing power has exploded, driving the rapid expansion of data centers. Simultaneously, the per-chip power consumption (TDP) of computing chips is rapidly increasing, significantly increasing the power density requirements for a single cabinet, from the traditional kilowatt level to the megawatt level.
[0003] In existing technologies, the peak power demand of GPUs in AI cabinets is typically several times their rated power. However, existing AI cabinet power supply unit designs often struggle to meet the power density requirements of several times the peak power within the same space constraints. Consequently, related technical solutions suffer from low cabinet power density.
[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention
[0005] The embodiments of the present application provide a power supply system, a power supply method and a whole cabinet to at least solve the technical problem of low power density in related power supply methods.
[0006] The present application provides a power supply system, which is applied to an entire cabinet and includes an energy management unit, multiple power supply units, multiple peak power supply units, a current state bus, and a power supply bus;
[0007] The plurality of power supply units are connected to the power supply bus and are used to supply power to at least one electrical device via the power supply bus;
[0008] The plurality of peak power supply units are connected to the power supply bus and are configured to supply power to at least one electrical device through the power supply bus upon receiving a power supply compensation instruction sent by the energy management unit;
[0009] The above-mentioned energy management unit is connected to the multiple power supply units through a current state bus. The above-mentioned current state bus is used to obtain the output current corresponding to each of the multiple power supply units. The above-mentioned energy management unit is used to send the above-mentioned power supply compensation instructions to the multiple peak power supply units when the sum of the above-mentioned output currents corresponding to each of the multiple power supply units meets the peak power compensation condition.
[0010] Optionally, the plurality of peak power supply units are connected via a first current-sharing bus;
[0011] The energy management unit is configured to determine a target compensation current based on a current difference between the sum of the output currents and the target rated current; and send the power supply compensation instruction to the plurality of peak power supply units based on the target compensation current;
[0012] The first current-sharing bus is used to determine the unit output currents corresponding to the plurality of peak power supply units according to the target compensation current.
[0013] Optionally, the first current sharing bus is further used to:
[0014] Obtain the number m of the peak power supply units in the standby state, where m is an integer greater than 1 and less than or equal to N+L, L is an integer greater than or equal to 1, and N is an integer greater than or equal to 1. The current entire cabinet includes N+L peak power supply units.
[0015] When m peak power supply units in standby state are switched to running state, the unit output current corresponding to each of the m peak power supply units in running state is determined according to the ratio between the target compensation current and the number of units m.
[0016] Optionally, the energy management unit is configured to obtain energy states corresponding to each of the plurality of peak power supply units according to the communication connection;
[0017] The energy management unit is further configured to:
[0018] When the stored electrical energy corresponding to each of the plurality of peak power supply units is less than or equal to a first threshold, sending a first control instruction, wherein the first control instruction is used to reduce the output power corresponding to each of the plurality of power supply units;
[0019] When the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, a second control instruction is sent, wherein the second control instruction is used to charge the plurality of peak power supply units.
[0020] Optionally, when the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, the energy management unit is configured to:
[0021] When the current difference between the sum of the output currents and the target rated current is greater than or equal to the reference charging current, determining the unit charging current corresponding to each of the m peak power supply units in operation according to the ratio of the reference charging current to m;
[0022] When the current difference between the sum of the above-mentioned output currents and the above-mentioned target rated current is less than the above-mentioned reference charging current, the above-mentioned unit charging current corresponding to each of the m above-mentioned peak power supply units in operation is determined according to the ratio of the above-mentioned current difference to the above-mentioned m.
[0023] Optionally, the energy management unit is configured to:
[0024] Obtaining the target rated current, wherein the target rated current is determined according to the unit rated currents of the plurality of power supply units;
[0025] In a case where the sum of the output currents is greater than or equal to a reference rated current, determining that the sum of the output currents satisfies the peak power compensation condition, wherein the reference rated current is determined based on the target rated current;
[0026] The current difference between the sum of the output currents and the target rated current is determined as the target compensation current.
[0027] Optionally, the plurality of peak power supply units are connected via a second current sharing bus;
[0028] The above-mentioned current state bus is used to obtain the state voltage detected on the above-mentioned current sharing bus; and determine the above-mentioned output current corresponding to each of the multiple power supply units according to the above-mentioned state voltage and a preset mapping relationship.
[0029] Optionally, the plurality of peak power supply units are connected via a second current sharing bus;
[0030] The energy management unit is used to determine the number n of the power supply units currently in working state based on the status information fed back by the current status bus, wherein n is an integer greater than 1 and less than or equal to M*N, M and N are integers greater than or equal to 1, and the current entire cabinet includes M distribution units, the first end of the distribution unit is connected to the input power bus, and the second end of the distribution unit is connected to N power supply units; the sum of the output currents is determined based on the product of the number of units n and the output current.
[0031] Optionally, multiple power supply units are arranged in the first area of the entire cabinet, multiple power supply units are connected in parallel, and the first ends of each of the multiple power supply units are connected to the power supply bus; multiple peak power supply units are arranged in the second area of the entire cabinet.
[0032] Optionally, a communication connection is established between the energy management unit and at least one server device;
[0033] The energy management unit is further configured to:
[0034] Determining target power parameters of the plurality of power supply units within the target time period based on task parameters corresponding to at least one of the server devices within the target time period;
[0035] Determine the target output current corresponding to each of the plurality of power supply units according to the target power parameter;
[0036] When the sum of the target output currents corresponding to the plurality of power supply units satisfies the peak power compensation condition, charging the plurality of peak power supply units is performed;
[0037] The voltages of the plurality of peak power supply units are adjusted to a reference voltage, wherein the reference voltage is lower than a target voltage corresponding to the power supply bus.
[0038] The present application also provides a power supply method, comprising:
[0039] Obtaining the output current corresponding to each of the multiple power supply units in the current entire cabinet through the current state bus, wherein the multiple power supply units are connected to the current state bus, and the multiple power supply units are connected to the power supply bus, and the power supply bus is used to supply power to at least one electrical device;
[0040] When the sum of the output currents corresponding to the plurality of power supply units meets the peak power compensation condition, a power supply compensation instruction is sent to the plurality of peak power supply units in the current entire cabinet;
[0041] A plurality of the above-mentioned peak power supply units are operated, wherein the plurality of the above-mentioned peak power supply units are connected to the above-mentioned power supply bus, and the plurality of the above-mentioned peak power supply units supply power to at least one electrical device through the above-mentioned power supply bus.
[0042] The present application also provides a whole cabinet, comprising the above-mentioned power supply system, wherein the above-mentioned power supply system is used to supply power to at least one of the above-mentioned server devices through a power supply bus.
[0043] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the above power supply methods when executing the computer program.
[0044] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above power supply methods are implemented.
[0045] The present application also provides a computer program product, including a computer program, which implements the steps of any of the above power supply methods when executed by a processor.
[0046] Through the above-described embodiment of the present application, the output current corresponding to each of the multiple parallel-connected power supply units is first obtained through the current state bus in the entire cabinet. Since the power supply units and the multiple peak power supply units are all connected to the power supply bus, and the power supply bus is used to power at least one electrical device, when the power demand of the electrical device suddenly increases, a current compensation instruction can be triggered by the energy management unit, and the peak power demand of the device can be met by the connected peak power supply unit.
[0047] In the above-described embodiments of the present application, the coordinated operation, dynamic monitoring and calculation of parallel power supply units, and the timely intervention of peak power supply units effectively address the low power density of the entire cabinet in the prior art. This not only meets the power requirements of the equipment during normal operation but also provides sufficient compensation current when the equipment power demand increases sharply, significantly improving the power density and power supply reliability of the entire cabinet, and resolving the technical problem of low power density that exists in prior art power supply methods based on entire cabinets. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0049] Figure 1 This is a hardware structure block diagram of a server device according to an embodiment of the present application;
[0050] Figure 2 is a schematic diagram of a power supply system according to an embodiment of the present application;
[0051] Figure 3 is a schematic diagram of an entire cabinet according to an embodiment of the present application;
[0052] Figure 4 is a schematic diagram of another power supply system according to an embodiment of the present application;
[0053] Figure 5 is a schematic diagram of another whole cabinet according to an embodiment of the present application;
[0054] Figure 6 is a schematic diagram of another power supply system according to an embodiment of the present application;
[0055] Figure 7is a flow chart of a power supply method according to an embodiment of the present application;
[0056] Figure 8 is a flow chart of another power supply method according to an embodiment of the present application;
[0057] Figure 9 It is a structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0058] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0059] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.
[0060] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0061] The method embodiments provided in the embodiments of the present application can be used to power server devices or similar computing devices. For example, an AI (Artificial Intelligence) server is an example of a power-consuming device. An AI server is a high-performance computing server designed specifically for artificial intelligence (AI) applications, primarily used to handle complex AI-related computing tasks such as machine learning, deep learning, data analysis, and model training. Due to varying degrees of concentration of related machine learning and model training tasks within the AI server, peak power can easily exceed rated power.
[0062] Figure 1 This is a hardware structure diagram of an AI server device according to an embodiment of the present application. Figure 1 As shown, the server 102 may include multiple ( Figure 1Only three are shown in the figure) of the graphics processing unit, such as a first graphics processing unit 104, a second graphics processing unit 106, and a third graphics processing unit 108. The multiple graphics processing units are connected to a graphics processing unit (GPU) chip power supply unit 110. The GPU chip power supply unit 110 is connected to an external circuit, for example, it can be connected to the power bus of the entire cabinet to obtain power from the entire cabinet.
[0063] It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above server device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.
[0064] In addition to including multiple image processing units, server 102 may also be equipped with a high-performance central processing unit (CPU) or a multi-core tensor processing unit (TPU). GPUs and TPUs are used to handle parallel computing tasks and support simultaneous processing of multiple tasks through a multi-core architecture.
[0065] The server 102 may further include a memory, which may be used to store computer programs, for example, software programs and modules of application software, such as computer programs corresponding to the data processing methods of the memory in the embodiments of the present application, and the central processing unit executes various functional applications and data processing by running the computer programs stored in the memory. The memory may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the server device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0066] Server 102 may further include a transmission device configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the server device's communications provider. In one embodiment, the transmission device includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the internet. In another embodiment, the transmission device may be a radio frequency (RF) module configured to communicate with the internet wirelessly.
[0067] The embodiment of the present application provides a power supply system for an entire cabinet, such as Figure 2 As shown, it includes an energy management unit 208, multiple power supply units, multiple peak power supply units, a power supply bus 210 and a current state bus 212;
[0068] Multiple power supply units, connected to the power supply bus 210, for supplying power to at least one power-consuming device through the power supply bus 210;
[0069] a plurality of peak power supply units connected to the power supply bus 210 and configured to supply power to at least one electrical device through the power supply bus 210 upon receiving a power supply compensation instruction sent by the energy management unit 208;
[0070] The energy management unit 208 is connected to multiple power supply units through the current state bus 212. The current state bus 212 is used to obtain the output current corresponding to each of the multiple power supply units. The energy management unit 208 is used to send power supply compensation instructions to the multiple peak power supply units when the sum of the output currents corresponding to the multiple power supply units meets the peak power compensation conditions.
[0071] It should be noted that in the above power supply system, multiple power supply units are arranged in the first area of the entire cabinet 200, the multiple power supply units are connected in parallel, and the first ends of the multiple power supply units (PSU, Power Supply Unit) are each connected to the power supply bus 210.
[0072] It is understandable that if Figure 2 As shown, the second end of each of the multiple power supply units can be connected to a power distribution unit (PDU), respectively, and the external power supply is then distributed to each power supply unit through the power distribution unit. The above-mentioned external power supply may include, but is not limited to, a power supply provided for an external power grid, and may also be a distributed power supply. The above-mentioned distributed power supply may be a power generation equipment distributed at a load center or a location close to a load center. The types of distributed power supplies may include solar photovoltaic power generation systems, wind power generation systems, fuel cell systems, natural gas power generation systems, hydropower generation systems, and biomass power generation systems, etc. The above-mentioned power distribution units and external power supplies are for illustrative purposes only and do not limit the specific units or power supply types used in the embodiments of the present application.
[0073] In the above power supply system, multiple peak load supply units (PLS) are arranged in the second area of the entire cabinet 200, and the multiple peak load supply units are connected to the power supply bus 210;
[0074] Optionally, the peak power supply unit can be used to provide additional power to devices during periods of peak power demand. Specifically, it is used to provide additional power support when the normal power supply unit cannot meet the system's full power demand, ensuring stable system operation and continuity of power supply. The PLS typically activates when power demand exceeds the capacity of the primary power supply to fill the power gap.
[0075] Specifically, the peak power supply unit may include but is not limited to one of a battery energy storage system, a generator power supply system, and a capacitor, and multiple peak power supply units may be a combination of one or more of the above systems or physical objects.
[0076] In the above power supply system, the power supply unit and the peak power supply unit are in different power supply domains in the entire cabinet, that is, the power supply unit and the peak power supply unit are separated in different structural units, and their outputs are uniformly connected to the power supply bus 210 of the entire cabinet, wherein all servers in the entire cabinet draw power from the power supply bus 210 to provide power for the internal load represented by the GPU.
[0077] In the above power supply system, the power bus 210 is used to supply power to at least one electrical device.
[0078] It should be noted that the power bus 210 can be a device used for power distribution, used to transmit power from a power source (e.g., a power supply unit) to a load (e.g., a server, computer, or other power-consuming device). The bus design allows for the parallel connection of multiple power supply units, thereby improving system reliability and flexibility. It can also be used to balance the load between the power supply units, ensuring even power distribution.
[0079] exist Figure 2 The power supply system shown can provide power to a first server 202, a second server 204, and a third server 206 via a power bus 210. The first server 202, the second server 204, and the third server 206 can be specifically an AI server for processing complex computing tasks related to artificial intelligence, such as machine learning, deep learning, data analysis, and model training.
[0080] In the above power supply system, the energy management unit 208 can establish communication connections with multiple peak power supply units and multiple power supply units. Specifically, the energy management unit 208 can establish communication connections with multiple power supply units through the current state bus 212. Then, the power supply control method is executed by the energy management unit 208, including: obtaining the output current corresponding to each of the multiple power supply units in the entire cabinet, and when the sum of the output currents corresponding to the multiple power supply units meets the peak power compensation condition, determining the target compensation current according to the current difference between the sum of the output currents and the target rated current; and operating the multiple peak power supply units according to the target compensation current, wherein the total output current corresponding to the multiple peak power supply units is the target compensation current.
[0081] It should be noted that the above peak power compensation condition can be used to indicate that in a power supply system, when the sum of the output currents of multiple power supply units meets a specific threshold condition, the system needs additional power support to meet higher power requirements.
[0082] Optionally, the first peak power compensation condition may be: the sum of the output currents reaches or exceeds the rated current. That is, when the sum of the output currents of the multiple power supply units reaches or exceeds the rated current of the power supply system, it is determined that the peak power compensation condition is met.
[0083] Optionally, the second peak power compensation condition may be that the fluctuation level of the sum of the output currents is greater than or equal to a target fluctuation parameter. This fluctuation level can then indicate unstable external grid power supply or unstable operation of the electrical equipment. When the external grid power supply is unstable or voltage fluctuates significantly, the system enters peak power compensation mode to ensure stable operation of the internal electrical equipment. In a specific embodiment, the fluctuation level of the sum of the output currents can be characterized by the current change rate or the current standard deviation, and the corresponding peak power compensation conditions are then determined using corresponding thresholds.
[0084] It should be noted that the first peak power compensation condition and the second peak power compensation condition can be applied to the power supply system selectively or together.
[0085] In the above-described embodiments of the present application, the coordinated operation, dynamic monitoring and calculation of parallel power supply units, and the timely intervention of peak power supply units effectively address the low power density of the entire cabinet in the prior art. This not only meets the power requirements of the equipment during normal operation but also provides sufficient compensation current when the equipment power demand increases sharply, significantly improving the power density and power supply reliability of the entire cabinet, and resolving the technical problem of low power density that exists in prior art power supply methods based on entire cabinets.
[0086] In an optional embodiment, the multiple peak power supply units are connected via a first current sharing bus;
[0087] an energy management unit, configured to determine a target compensation current based on a current difference between the sum of the output currents and the target rated current; and send power supply compensation instructions to the plurality of peak power supply units based on the target compensation current;
[0088] The first current sharing bus is used to determine the unit output current corresponding to each of the multiple peak power supply units according to the target compensation current.
[0089] It should be noted that, in the above embodiment, in the above power supply system, the first current sharing bus can evenly distribute the current to the power supply branches corresponding to each peak power supply unit. Figure 2 In the parallel power supply system shown, multiple peak power supply units are connected through a current-sharing bus to ensure that the current output by each peak power supply unit is approximately equal, thereby preventing any peak power supply unit from being overloaded.
[0090] It will be appreciated that in the above embodiment, the energy management unit can detect the sum of the output currents of the entire system via the current status bus and compare it with the target rated current. If there is a difference between the sum of the output currents and the target rated current (i.e., a current difference), the energy management unit can further calculate the target compensation current to be compensated.
[0091] Furthermore, based on the calculated target compensation current, the energy management unit sends power compensation instructions to multiple peak power supply units. These instructions are used to adjust the output current of each peak power supply unit to ensure that the current output of the entire system reaches the target rated current.
[0092] Through the above-described embodiments of the present application, the first current-sharing busbar can dynamically adjust the output current of each peak power supply unit in response to the energy management unit's instructions. This enables the power supply system to dynamically respond to load changes, optimize current distribution, and improve system efficiency and reliability. Furthermore, the first current-sharing busbar can prevent damage to a peak power supply unit due to overload, thereby ensuring stable operation of the entire system.
[0093] The following describes a method for the first current sharing bus to determine the unit output current corresponding to each peak power supply unit. In an optional embodiment, the first current sharing bus is further used to:
[0094] S1. Obtain the number m of peak power supply units in standby mode, where m is an integer greater than 1 and less than or equal to N+L, L is an integer greater than or equal to 1, and N is an integer greater than or equal to 1. The entire cabinet currently includes N+L peak power supply units.
[0095] S2, when m peak power supply units in standby state are switched to running state, determining the unit output current corresponding to each of the m peak power supply units in running state according to the ratio between the target compensation current and the number of units m.
[0096] Optionally, in the above embodiment, the peak power compensation unit utilizes an N+L redundant power supply. In this embodiment, the peak power compensation unit can be designed to operate without an input power bus (e.g., using capacitors). This eliminates the possibility of an input bus power failure, thereby reducing redundancy. Optionally, if system reliability needs to be further enhanced, the system can be adjusted to N+1, N+2, or N+L, depending on system requirements.
[0097] Through the above-mentioned implementation of the present application, when the space of the entire cabinet is limited, an N+L redundant peak power supply unit architecture can be adopted to ensure high reliability of the input power supply of the entire cabinet under peak conditions.
[0098] In an optional embodiment, a communication connection is established between the energy management unit and the first current sharing bus, and the energy management unit is configured to obtain energy states corresponding to each of the multiple peak power supply units according to the communication connection;
[0099] The energy management unit is further configured to:
[0100] Control mode 1: When the stored electrical energy corresponding to each of the multiple peak power supply units is less than or equal to a first threshold, sending a first control instruction, wherein the first control instruction is used to reduce the output power corresponding to each of the multiple power supply units;
[0101] Control mode 2: When the stored electric energy corresponding to each of the multiple peak power supply units is less than or equal to the second threshold, a second control instruction is sent, wherein the second control instruction is used to charge the multiple peak power supply units.
[0102] In the above embodiment, after the redundant power supply operation is performed by the plurality of peak power supply units, the peak power supply units may be further charged by the at least two control methods described above.
[0103] In the first control method described above, when the stored energy of multiple peak power supply units is less than or equal to a first threshold, a first control instruction can be sent through the power supply system to reduce the output power of the multiple power supply units (PSUs). This can prevent excessive discharge of the PSUs.
[0104] For example, a first threshold (e.g., 30% stored power) can be set based on PLS specifications and system requirements. The PLS's stored power is then monitored in real time, and a control instruction is triggered when the power drops below the first threshold. Finally, the energy management unit (EMU) sends a control instruction to the GPU working units throughout the rack, reducing their output power.
[0105] In the above-mentioned control method 2, when the stored electrical energy of multiple peak power supply units is less than or equal to the second threshold, a second control instruction can be sent to charge the multiple peak power supply units, thereby ensuring that the PLS has sufficient electrical energy reserves to cope with future peak power demand.
[0106] For example, a second threshold (e.g., 10% of stored energy) can be set based on the PLS specifications and system requirements. Combined with real-time monitoring of the PLS's stored energy, a control command is triggered when the energy drops to the second threshold. Finally, the control command is sent to the charging device to charge the PLS.
[0107] In another optional embodiment, the above two control methods can also be combined to implement. In the combined implementation, it can be implemented in the following ways:
[0108] If the output current is detected to be greater than the preset maximum value multiple times, the PLS will immediately jump to the power compensation state; if power compensation occurs multiple times in succession, but the internal energy storage value of the PLS is insufficient, it will then communicate at high speed to the energy management unit of the entire cabinet, and further inform the GPU working unit of the entire cabinet to perform short-term power reduction and other working states. When the PLS state recovers, it will return to the peak power compensation state.
[0109] In an optional embodiment, when the stored electric energy corresponding to each of the multiple peak power supply units is less than or equal to the second threshold, the energy management unit is configured to:
[0110] When the current difference between the sum of the output currents and the target rated current is greater than or equal to the reference charging current, determining the unit charging current corresponding to each of the m peak power supply units in operation according to the ratio of the reference charging current to m;
[0111] When the current difference between the sum of the output currents and the target rated current is less than the reference charging current, the unit charging current corresponding to each of the m peak power supply units in operation is determined according to the ratio of the current difference to m.
[0112] The following combination Figure 3 The method for determining the above charging current is explained below. Figure 3As shown in FIG, after the peak power supply unit (PLS) is discharged, the total output current B of the power supply unit (PSU) continues to be monitored in real time. At the same time, the maximum charging current value C of the PLS is preset within the system.
[0113] The specific charging current is determined as follows: when the difference (AB) between the preset peak power compensation baseline value A and the total PSU output current B is greater than the preset maximum charging current value C, the total charging current of the PLS will be set to C. In this case, the charging current of each PLS is the total charging current C divided by the total number of PLSs n+1, that is, the charging current of each PLS is C / (n+1);
[0114] When the difference (AB) between the preset peak power compensation baseline value A and the total PSU output current B is less than the preset maximum charge current value C, the total charge current of the PLS will be set to AB. In this case, the charge current of each PLS is the total charge current (AB) divided by the total number of PLSs (n+1), that is, the charge current of each PLS is (AB) / (n+1).
[0115] like Figure 3 As shown in the figure, the PLS consists of n+1 power modules, each of which has dynamic current sharing control. The n+1 power modules are connected via a current sharing bus. The PLSs can set their internal current states based on information on the current sharing bus, thus achieving dynamic current sharing control between the PLSs. Simultaneously, the status information of the current sharing bus is transmitted to the energy management unit (EMU) of the entire cabinet via a communication signal.
[0116] The EMU analyzes the real-time total output current of the PLS based on the voltage status of the current-sharing bus and the number of currently operating PLSs. Furthermore, by monitoring the voltage of the voltage-sharing bus, the EMU can calculate the charging and discharging power of the PLSs. Based on the difference in charge and discharge power, the EMU can predict the next power compensation requirement, optimizing energy use and improving power supply system efficiency.
[0117] Through the above implementation, the charging current can be dynamically adjusted according to the actual power demand and the charging capacity of the PLS, ensuring that the PLS is charged under safe and efficient conditions.
[0118] In an optional embodiment, the energy management unit is configured to:
[0119] Obtaining a target rated current, wherein the target rated current is determined according to the unit rated currents of the plurality of power supply units;
[0120] When the sum of the output currents is greater than or equal to a reference rated current, determining that the sum of the output currents meets a peak power compensation condition, wherein the reference rated current is determined according to the target rated current;
[0121] A current difference between the sum of the output currents and the target rated current is determined as a target compensation current.
[0122] In the above embodiment, the target rated current can be determined based on the rated current of each of the multiple power supply units. In other words, the target rated current is the current level anticipated during the design of the power supply system and reflects the total current required by the entire power supply system under normal operating conditions. The target rated current is set by taking into account the sum of the rated currents of all power supply units to ensure that the power supply system can meet the power requirements of all powered devices.
[0123] It should be further explained that the reference rated current is determined based on the target rated current and is used to determine whether the sum of the output currents of the current power supply units meets the peak power compensation requirements. When the sum of the output currents is greater than or equal to the reference rated current, it is considered that the current power supply system needs to trigger peak power compensation, and the peak power supply unit is activated to provide additional power support.
[0124] There are many ways to determine the reference rated current based on the target rated current. For example, the reference rated current can be the same as the target rated current. In addition, the reference rated current corresponding to the target rated current can be determined by the following possible methods:
[0125] Determination method 1: Fixed ratio method. The reference rated current can be set as a fixed percentage of the target rated current. For example, if the target rated current is 100A, the reference rated current can be set to 90A, which is 90% of the target rated current.
[0126] Method 2: Dynamic Adjustment. The reference rated current can be dynamically adjusted based on the system's real-time operating status. For example, if the system detects an increase in power demand, the reference rated current can be appropriately reduced to more quickly respond to peak power demands.
[0127] Determination method three: Historical data analysis. A reasonable reference rated current is determined by analyzing historical power demand data. This historical data analysis method further considers information sources such as the operating cycle of electrical equipment and, combined with system load conditions, dynamically determines the reference rated current corresponding to each time point.
[0128] The following is a specific example of the second determination method:
[0129] In this embodiment, dynamic adjustment of the reference rated current can be achieved through the following steps, specifically including: S1, setting adjustment rules: Based on historical data and system design, the reference rated current adjustment rules are set when power demand increases. For example, for every 10% increase in power demand, the reference rated current is reduced by 5%, thereby improving the compensation response speed.
[0130] S2, monitoring power demand: The power supply system needs to monitor the power consumption of each electrical device in real time. For example, data can be collected through sensors or smart meters to analyze the overall power demand trend.
[0131] S3, implements adjustment; when an increase in power demand is detected, the reference rated current is automatically adjusted. The adjusted reference rated current should ensure that the system can respond to peak power demand in a timely manner while avoiding overcompensation.
[0132] S4, Feedback and Optimization: The power supply system can provide feedback based on the adjusted results and optimize the adjustment rules. For example, through machine learning algorithms, the system can learn the optimal adjustment strategy, improving response speed and accuracy.
[0133] Through the above-described implementation of this application, the power supply system can flexibly adapt to changes in power demand, ensuring timely provision of additional power during peak demand periods while avoiding unnecessary power waste when demand decreases. This dynamic adjustment method improves the system's responsiveness and operational efficiency, helping to optimize energy use and reduce operating costs.
[0134] In an optional embodiment, the multiple peak power supply units are connected via a second current sharing bus;
[0135] The current state bus is used to obtain the state voltage detected on the current sharing bus; and determine the output current corresponding to each of the multiple power supply units according to the state voltage and a preset mapping relationship.
[0136] In the above embodiment, the plurality of power supply units are respectively connected to the second current sharing bus, which can collect the output currents of all power supply units and distribute them to the electrical devices in the system.
[0137] Then, the state voltage can be obtained through the current state bus at the second current sharing bus or near the bus. Then, the output current can be determined based on the mapping relationship between the preset state voltage and current. The above mapping relationship is established based on the resistance characteristics of the current sharing bus and the output characteristics of the power supply unit. For example, if the resistance of the current sharing bus is 1 milliohm, then for every 1 ampere of current passing through, a voltage drop of 1 millivolt will be generated across the bus.
[0138] To determine the output current, the system calculates the total current flowing through the current-sharing bus based on the detected state voltage and a preset mapping relationship. It then determines the output current of each power supply unit by analyzing the output characteristics of each power supply unit and their configuration within the system.
[0139] Specifically, assuming that the resistance of the second current-sharing bus is 1 milliohm, the preset mapping relationship indicates that every 1 millivolt voltage drop corresponds to 1 ampere of current. If the state voltage detected in step S1 is 10 millivolts, then according to the mapping relationship, the total current flowing through the second current-sharing bus is 10 amperes. If there are 5 power supply units in the system and their output characteristics are the same, then the output current of each unit can be assumed to be equal. Therefore, the output current of each power supply unit is the total current divided by the number of units, that is, 10A / 5 = 2A.
[0140] Through the above-mentioned implementation of the present application, the output current of each power supply unit can be accurately monitored and controlled to ensure the uniformity of current distribution and the stability of the system.
[0141] In an optional embodiment, the multiple peak power supply units are connected via a second current sharing bus;
[0142] The energy management unit is used to determine the number n of power supply units currently in operation based on the status information fed back by the current status bus, where n is an integer greater than 1 and less than or equal to M*N, M and N are integers greater than or equal to 1, and the entire cabinet currently includes M distribution units, the first end of the distribution unit is connected to the input power bus, and the second end of the distribution unit is connected to N power supply units; the sum of the output currents is determined based on the product between the number of units n and the output current.
[0143] It can be understood that in the above-mentioned embodiment of the present application, an M*N redundant power supply unit power supply plus an N+L redundant peak power supply unit architecture can be further adopted to ensure high reliability of the input power supply of the entire cabinet.
[0144] When M is specifically 2 and L is specifically 1, if the space of the entire cabinet is limited, an N+N redundant power supply unit plus an N+1 redundant peak power supply unit architecture can be adopted to ensure high reliability of the input power supply of the entire cabinet.
[0145] The following combination Figure 4 The above power supply system and power supply method are further explained.
[0146] exist Figure 4The entire cabinet shown uses an N+N redundant power supply unit and an N+1 redundant peak power supply unit architecture to ensure high reliability of the input power supply of the entire cabinet.
[0147] The entire cabinet adopts a dual-input bus power supply architecture, and is compatible with more data center power supply bus voltage architectures. The input bus can provide redundant power supply for two HVAC (High-Voltage Alternating Current) buses, two HVDC (High-Voltage Direct Current) buses, or a mutually redundant power supply design for one HVAC and HVDC bus.
[0148] Furthermore, each input power bus is input to the power supply unit (PSU) unit through the power distribution unit, and physical isolation is adopted between the two power distribution units to ensure that a failure in one line will not affect the failure in the other line.
[0149] Furthermore, the power supply unit adopts an N+N redundant architecture. Two groups of N PSUs input into independent power supply networks, and then output into a common output network. Each group contains N PSUs, and each power supply network is connected to different mains power lines in the data center. The use of a physically isolated busbar architecture and a common current sharing busbar control design ensures that if a mains power line in the data center fails or N PSUs in the architecture fail simultaneously, the remaining N PSUs in the architecture can continue to ensure the reliability of the power supply for the entire AI cabinet, thereby improving the fault tolerance of the power supply for the entire AI cabinet system.
[0150] Furthermore, the peak power compensation unit utilizes an N+1 redundant power supply. In this embodiment, the peak power compensation unit can be designed without an input power bus (e.g., using capacitors), eliminating the possibility of input bus power failures and thus reducing redundancy. Optionally, if system reliability needs to be further enhanced, the system can be adjusted to N+2 or N+L based on system requirements.
[0151] At the same time, the power supply unit and the peak power supply unit are located in separate power supply domains of the entire cabinet, and their outputs are uniformly connected to the power supply bus of the entire cabinet. All servers in the entire cabinet draw power from this power supply bus to provide power for internal loads represented by GPUs.
[0152] The following combination Figure 5 , further explains the method for determining the output current corresponding to the above power supply system.
[0153] like Figure 5As shown in the figure, since the entire cabinet is powered by N+N redundant power supply units, the 2N power supply units share the cabinet power bus. Therefore, the overall power load of the power supply units uses an average current configuration for each power supply unit. For example, if the total power is 20kW, the 2N power supply units specifically have 20 power supply units, and each PSU has a power of 1kW.
[0154] Furthermore, the average current control of each power supply unit adopts the method of interconnecting the current state bus. Each PSU interconnects the current state bus together. When controlling each power supply unit, the internal output voltage and output current control will be realized according to the state of the current sharing bus.
[0155] It is understandable that the output current status of each power supply unit will be reflected on the current status bus. The current status of all power supply units is connected to the current status bus, which further ensures the balance of the output power of each power supply unit connected to the bus. By measuring the voltage value of the current status bus, the output current value of each PSU can be calculated. At the same time, by collecting the voltage value of the power bus of the entire cabinet, the power load status of the PSU can be understood, including the number of PSUs currently working. Multiplying this number by the power value of a single PSU can obtain the total power of all PSUs.
[0156] Furthermore, by feeding this information back to the entire cabinet's energy management unit (EMU) through communication, more precise power management and optimization can be achieved. For example, as shown in Table 1 below, if a bus voltage of 1V corresponds to an output current of 100A, and the voltage platform of the entire cabinet is 50V, then by analyzing the power output and current output of each PSU, the total output current and total output power can be calculated based on the number of PSUs.
[0157] Table 1
[0158]
[0159] In an optional embodiment, a communication connection is established between the energy management unit and at least one server device;
[0160] The energy management unit is also used to:
[0161] Determining target power parameters of multiple power supply units within the target time period based on task parameters corresponding to at least one server device within the target time period;
[0162] Determining target output currents corresponding to the multiple power supply units according to the target power parameters;
[0163] When the sum of the target output currents corresponding to the multiple power supply units meets the peak power compensation condition, performing a charging operation on the multiple peak power supply units;
[0164] The voltages of the plurality of peak power supply units are adjusted to a reference voltage, wherein the reference voltage is less than a target voltage corresponding to the power supply bus.
[0165] It is understood that, if the power-consuming device is a server, the energy management unit can determine target power parameters for multiple power supply units (PSUs) within a target time period based on the task parameters corresponding to at least one server device within the target time period. This can be achieved by analyzing the server's task requirements, such as the computing requirements of a GPU model, to predict the required power.
[0166] Based on the target power parameters, the cabinet energy management unit can determine the target output current corresponding to each of the multiple PSUs, thereby ensuring that each PSU can provide the required current to meet the power requirements of the server.
[0167] In the above S4, the reference voltage is lower than the target voltage corresponding to the power supply bus so that the PLS can respond quickly and provide additional power when power supply compensation is required, while avoiding excessive voltage shock to the power supply system.
[0168] In a specific service architecture, such as Figure 6 As shown, the whole-cabinet energy management unit can establish communication connections with each server unit, PSU, and PLS, allowing the EMU to pre-estimate the computing needs of the GPU model within the server, predict power consumption, and allocate power units and compensation units in advance. This communication connection also allows the whole-cabinet energy management unit to comprehensively analyze power demand, voltage and current status, and the operating status of the GPU within the server over a period of time, thereby predicting the next peak power state in advance.
[0169] To achieve highly dynamic high-power compensation response, the cabinet energy management unit can instruct the PLS to precharge the output port voltage to a value lower than the cabinet power bus voltage as a preparatory process. When peak power demand is detected, the PLS can respond quickly and provide the required additional power.
[0170] Furthermore, the PSU communicates with the PLS through the rack-wide energy management unit, transmitting its rated load capacity and peak power capability to the PLS before power is applied. This allows the PLS to pre-set the PSU's rated power capability and initiate power compensation if the rated power capability is exceeded.
[0171] Through the above-described implementation of the present application, information exchange between the entire cabinet energy management unit and the server unit allows for charging and discharging power assessments based on voltage and current, and the difference can be used to predict the next step of peak power compensation. Furthermore, based on the GPU operating status and the model's estimated power demand, information can be communicated to the PLS unit in advance, prompting the PLS to perform preprocessing and achieve faster dynamic response capabilities.
[0172] The embodiment of the present application also provides a power supply method applied to the above power supply system, such as Figure 7 As shown, the following steps are included:
[0173] S702: Obtain output currents corresponding to respective power supply units in the entire cabinet via a current status bus, wherein the multiple power supply units are connected to the current status bus, and the multiple power supply units are connected to a power supply bus, which is used to supply power to at least one electrical device.
[0174] S704: When the sum of the output currents corresponding to the multiple power supply units meets the peak power compensation condition, a power supply compensation instruction is sent to the multiple peak power supply units in the current entire cabinet;
[0175] S706: Operate multiple peak power supply units, wherein the multiple peak power supply units are connected to a power supply bus, and the multiple peak power supply units supply power to at least one electrical device through the power supply bus.
[0176] It should be noted that the peak power compensation condition in the above S704 can be used to indicate that in a power supply system, when the sum of the output currents of multiple power supply units meets a specific threshold condition, the system needs additional power support to meet higher power requirements.
[0177] Optionally, the first peak power compensation condition may be: the sum of the output currents reaches or exceeds the rated current. That is, when the sum of the output currents of the multiple power supply units reaches or exceeds the rated current of the power supply system, it is determined that the peak power compensation condition is met.
[0178] Optionally, the second peak power compensation condition may be that the fluctuation level of the sum of the output currents is greater than or equal to a target fluctuation parameter. This fluctuation level can then indicate unstable external grid power supply or unstable operation of the electrical equipment. When the external grid power supply is unstable or voltage fluctuates significantly, the system enters peak power compensation mode to ensure stable operation of the internal electrical equipment. In a specific embodiment, the fluctuation level of the sum of the output currents can be characterized by the current change rate or the current standard deviation, and the corresponding peak power compensation conditions are then determined using corresponding thresholds.
[0179] It should be noted that the first peak power compensation condition and the second peak power compensation condition can be applied to the power supply system selectively or together.
[0180] In the above-described embodiments of the present application, the coordinated operation, dynamic monitoring and calculation of parallel power supply units, and the timely intervention of peak power supply units effectively address the low power density of the entire cabinet in the prior art. This not only meets the power requirements of the equipment during normal operation but also provides sufficient compensation current when the equipment power demand increases sharply, significantly improving the power density and power supply reliability of the entire cabinet, and resolving the technical problem of low power density that exists in prior art power supply methods based on entire cabinets.
[0181] In an optional embodiment, when the sum of the output currents corresponding to the plurality of power supply units satisfies the peak power compensation condition, determining the target compensation current according to the current difference between the sum of the output currents and the target rated current includes:
[0182] S1, obtaining a target rated current, wherein the target rated current is determined according to the unit rated currents of the plurality of power supply units;
[0183] S2, when the sum of the output currents is greater than or equal to a reference rated current, determining whether the sum of the output currents satisfies a peak power compensation condition, wherein the reference rated current is determined according to the target rated current;
[0184] S3, determining the current difference between the sum of the output currents and the target rated current as the target compensation current.
[0185] In the above embodiment, the target rated current can be determined based on the rated current of each of the multiple power supply units. In other words, the target rated current is the current level anticipated during the design of the power supply system and reflects the total current required by the entire power supply system under normal operating conditions. The target rated current is set by taking into account the sum of the rated currents of all power supply units to ensure that the power supply system can meet the power requirements of all powered devices.
[0186] It should be further explained that the reference rated current is determined based on the target rated current and is used to determine whether the sum of the output currents of the current power supply units meets the peak power compensation requirements. When the sum of the output currents is greater than or equal to the reference rated current, it is considered that the current power supply system needs to trigger peak power compensation, and the peak power supply unit is activated to provide additional power support.
[0187] Through the above-described implementation of this application, the power supply system can flexibly adapt to changes in power demand, ensuring timely provision of additional power during peak demand periods while avoiding unnecessary power waste when demand decreases. This dynamic adjustment method improves the system's responsiveness and operational efficiency, helping to optimize energy use and reduce operating costs.
[0188] In an optional embodiment, the obtaining of the output current corresponding to each of the multiple power supply units in the current entire cabinet includes:
[0189] S1, when multiple power supply units are respectively connected to the current sharing bus, obtaining the state voltage detected on the current sharing bus through the current state bus;
[0190] S2, determining the output current corresponding to each of the plurality of power supply units according to the state voltage and a preset mapping relationship.
[0191] In the above embodiment, the multiple power supply units are respectively connected to a current sharing bus bar, which can collect the output currents of all power supply units and distribute them to the electrical devices in the system.
[0192] Then, the state voltage can be obtained through the current state bus at or near the current sharing bus. Then, the output current can be determined based on the pre-set mapping relationship between the state voltage and the current. The above mapping relationship is established based on the resistance characteristics of the current sharing bus and the output characteristics of the power supply unit. For example, if the resistance of the current sharing bus is 1 milliohm, then for every 1 ampere of current passing through, a voltage drop of 1 millivolt will be generated across the bus.
[0193] To determine the output current, the system calculates the total current flowing through the current-sharing bus based on the detected state voltage and a preset mapping relationship. It then determines the output current of each power supply unit by analyzing the output characteristics of each power supply unit and their configuration within the system.
[0194] Through the above-mentioned implementation of the present application, the output current of each power supply unit can be accurately monitored and controlled to ensure the uniformity of current distribution and the stability of the system.
[0195] In an optional embodiment, when the sum of the output currents corresponding to the plurality of power supply units satisfies the peak power compensation condition, before determining the target compensation current according to the current difference between the sum of the output currents and the target rated current, the method further includes:
[0196] S1. When multiple power supply units are respectively connected to the current sharing bus, determine the number n of power supply units currently in operation, where n is an integer greater than 1 and less than or equal to M*N, M and N are integers greater than or equal to 1, and the entire cabinet currently includes M power distribution units, a first end of the power distribution unit is connected to the input power bus, and a second end of the power distribution unit is connected to N power supply units;
[0197] S2, determining the sum of the output currents according to the product between the number of units n and the output current;
[0198] Run multiple peak power supply units in the current cabinet, including:
[0199] S1: Obtain the number m of peak power supply units in standby mode, where m is an integer greater than 1 and less than or equal to N+L, and L is an integer greater than or equal to 1. The entire cabinet currently includes N+L peak power supply units.
[0200] S2, switching the m peak power supply units in the standby state to the running state, and determining the unit output current corresponding to each of the m peak power supply units in the running state according to the ratio between the target compensation current and the number of units m.
[0201] The following combination Figure 4 The above power supply system and power supply method are further explained.
[0202] exist Figure 4 The entire cabinet shown uses an N+N redundant power supply unit and an N+1 redundant peak power supply unit architecture to ensure high reliability of the input power supply of the entire cabinet.
[0203] The entire cabinet adopts a dual-input bus power supply architecture, and is compatible with more data center power supply bus voltage architectures. The input bus can provide redundant power supply for two HVAC (High-Voltage Alternating Current) buses, two HVDC (High-Voltage Direct Current) buses, or a mutually redundant power supply design for one HVAC and HVDC bus.
[0204] Furthermore, each input power bus is input to the power supply unit (PSU) unit through the power distribution unit, and physical isolation is adopted between the two power distribution units to ensure that a failure in one line will not affect the failure in the other line.
[0205] Furthermore, the power supply unit adopts an N+N redundant architecture. Two groups of N PSUs input into independent power supply networks, and then output into a common output network. Each group contains N PSUs, and each power supply network is connected to different mains power lines in the data center. The use of a physically isolated busbar architecture and a common current sharing busbar control design ensures that if a mains power line in the data center fails or N PSUs in the architecture fail simultaneously, the remaining N PSUs in the architecture can continue to ensure the reliability of the power supply for the entire AI cabinet, thereby improving the fault tolerance of the power supply for the entire AI cabinet system.
[0206] Furthermore, the peak power compensation unit utilizes an N+1 redundant power supply. In this embodiment, the peak power compensation unit can be designed without an input power bus (e.g., using capacitors), eliminating the possibility of input bus power failures and thus reducing redundancy. Optionally, if system reliability needs to be further enhanced, the system can be adjusted to N+2 or N+L based on system requirements.
[0207] At the same time, the power supply unit and the peak power supply unit are located in separate power supply domains of the entire cabinet, and their outputs are uniformly connected to the power supply bus of the entire cabinet. All servers in the entire cabinet draw power from this power supply bus to provide power for internal loads represented by GPUs.
[0208] The following combination Figure 8 , the complete process of the current compensation method corresponding to the above power supply system is further explained.
[0209] S802: Power on and the energy management unit of the entire cabinet performs a self-test.
[0210] S804: The system is normal. The PLS internal energy storage unit is pre-charged and reports to the cabinet energy management unit after charging is complete.
[0211] S806: After the PLS is fully charged, it is always in standby mode to detect the voltage of the power bus of the entire cabinet and the voltage of the PSU current sharing bus.
[0212] S810, the PLS analyzes the PSU output current b and the total current B according to the PSU current sharing bus voltage status;
[0213] S812: The PSU communicates with the PLS through the cabinet energy management unit. The PSU transmits its rated load capacity and peak power capacity to the PLS.
[0214] S814, PLS presets a peak power compensation baseline value, for example, the preset value is A;
[0215] S816, when the PLS responds to the total PSU output current B being greater than the preset rated value A, it turns on the PLS, releases the internal stored energy, and achieves peak power compensation. The compensation current is BA, and then the compensation current of a single PLS is (BA) / n+1, where n+1 is the number of PLSs.
[0216] It should be noted that, when all configured n+1 PLSs are in normal working condition, the compensation current corresponding to each PLS can be determined using the above method. If there is a failed PLS among the n+1 PLSs, the compensation current corresponding to each PLS can be determined based on (BA) / m, where m is the number of PLSs actually working.
[0217] S818, after the PLS is discharged, the total PSU output current B is further detected, and the maximum PLS charging current value C is preset internally;
[0218] S820, when the value of BA is greater than C, the total charging current of the PLS is C, and the charging current of a single PLS is C / n+1; when the value of BA is less than C, the total charging current of the PLS is BA, and the charging current of a single PLS is (BA) / n+1.
[0219] Through the above-described embodiments of the present application, the self-checking of the cabinet's Energy Management Unit (EMU) and the pre-charging mechanism of the Peak Load Supply Unit (PLS) ensure stable system operation upon power-up. Once fully charged, the PLS remains in a standby state, ready to respond to peak power demand fluctuations, thereby improving the reliability of the entire power supply system.
[0220] Furthermore, the PLS monitors the voltage of the entire cabinet's power bus and the voltage of the PSU (Power Supply Unit) current-sharing bus in real time, accurately analyzing the PSU's output current and total current. This precise monitoring and control helps optimize power distribution and ensure the stability and efficiency of the power supply.
[0221] Furthermore, the solution can flexibly adjust the compensation current of each PLS if the number and operating status of the PLSs change. Whether all PLSs are operating normally or some are failing, the system can determine the compensation current for each PLS based on the actual number of operating PLSs (m), ensuring the accuracy and effectiveness of power compensation.
[0222] After discharging, the PLS further detects the total PSU output current (B) and intelligently manages the PLS charging process based on a preset maximum charge current (C). This intelligent charging management not only improves charging efficiency but also helps extend the life of the PLS.
[0223] Through the above implementation methods, accurate current monitoring, dynamic power compensation, flexible PLS management and intelligent charging management are used to significantly improve the reliability, stability and energy utilization efficiency of the power supply system, while also enhancing the adaptability and flexibility of the power supply system.
[0224] In an optional embodiment, after the above-mentioned operation of multiple peak power supply units in the current entire cabinet, at least one of the following is further included:
[0225] Control mode 1: When the stored electrical energy corresponding to each of the multiple peak power supply units is less than or equal to a first threshold, sending a first control instruction, wherein the first control instruction is used to reduce the output power corresponding to each of the multiple power supply units;
[0226] In the first control method described above, when the stored energy of multiple peak power supply units is less than or equal to a first threshold, a first control instruction can be sent through the power supply system to reduce the output power of the multiple power supply units (PSUs). This can prevent excessive discharge of the PSUs.
[0227] For example, a first threshold (e.g., 30% stored power) can be set based on PLS specifications and system requirements. The PLS's stored power is then monitored in real time, and a control instruction is triggered when the power drops below the first threshold. Finally, the energy management unit (EMU) sends a control instruction to the GPU working units throughout the rack, reducing their output power.
[0228] Control mode 2: When the stored electric energy corresponding to each of the multiple peak power supply units is less than or equal to the second threshold, a second control instruction may be sent, wherein the second control instruction is used to charge the multiple peak power supply units.
[0229] In the above-mentioned control method 2, when the stored electrical energy of multiple peak power supply units is less than or equal to the second threshold, a second control instruction can be sent to charge the multiple peak power supply units, thereby ensuring that the PLS has sufficient electrical energy reserves to cope with future peak power demand.
[0230] For example, a second threshold (e.g., 10% of stored energy) can be set based on the PLS specifications and system requirements. Combined with real-time monitoring of the PLS's stored energy, a control command is triggered when the energy drops to the second threshold. Finally, the control command is sent to the charging device to charge the PLS.
[0231] In another optional embodiment, the above two control methods can also be combined to implement. In the combined implementation, it can be implemented in the following ways:
[0232] If the output current is detected to be greater than the preset maximum value multiple times, the PLS will immediately jump to the power compensation state; if power compensation occurs multiple times in succession, but the internal energy storage value of the PLS is insufficient, it will then communicate at high speed to the energy management unit of the entire cabinet, and further inform the GPU working unit of the entire cabinet to perform short-term power reduction and other working states. When the PLS state recovers, it will return to the peak power compensation state.
[0233] In an optional embodiment, when the stored electrical energy corresponding to each of the plurality of peak power supply units is less than or equal to the second threshold, sending the second control instruction includes:
[0234] When the current difference between the sum of the output currents and the target rated current is greater than or equal to the reference charging current, determining the unit charging current corresponding to each of the m peak power supply units in operation according to the ratio of the reference charging current to m;
[0235] When the current difference between the sum of the output currents and the target rated current is less than the reference charging current, the unit charging current corresponding to each of the m peak power supply units in operation is determined according to the ratio of the current difference to m.
[0236] The following combination Figure 3 The method for determining the above charging current is explained below. Figure 3 As shown in FIG, after the peak power supply unit (PLS) is discharged, the total output current B of the power supply unit (PSU) continues to be monitored in real time. At the same time, the maximum charging current value C of the PLS is preset within the system.
[0237] The specific charging current is determined as follows: when the difference (AB) between the preset peak power compensation baseline value A and the total PSU output current B is greater than the preset maximum charging current value C, the total charging current of the PLS will be set to C. In this case, the charging current of each PLS is the total charging current C divided by the total number of PLSs n+1, that is, the charging current of each PLS is C / (n+1);
[0238] When the difference (AB) between the preset peak power compensation baseline value A and the total PSU output current B is less than the preset maximum charge current value C, the total charge current of the PLS is set to AB. In this case, the charge current of each PLS is the total charge current (AB) divided by the total number of PLSs (n+1), that is, the charge current of each PLS is (A-B) / (n+1).
[0239] like Figure 3As shown in the figure, the PLS consists of n+1 power modules, each of which has dynamic current sharing control. The n+1 power modules are connected via a current sharing bus. The PLSs can set their internal current states based on information on the current sharing bus, thus achieving dynamic current sharing control between the PLSs. Simultaneously, the status information of the current sharing bus is transmitted to the energy management unit (EMU) of the entire cabinet via a communication signal.
[0240] The EMU analyzes the real-time total output current of the PLS based on the voltage status of the current-sharing bus and the number of currently operating PLSs. Furthermore, by monitoring the voltage of the voltage-sharing bus, the EMU can calculate the charging and discharging power of the PLSs. Based on the difference in charge and discharge power, the EMU can predict the next power compensation requirement, optimizing energy use and improving power supply system efficiency.
[0241] Through the above implementation, the charging current can be dynamically adjusted according to the actual power demand and the charging capacity of the PLS, ensuring that the PLS is charged under safe and efficient conditions.
[0242] In an optional embodiment, before operating multiple peak power supply units in the current entire cabinet, the above further includes:
[0243] S1, when the power-consuming device is a server device, determining target power parameters of multiple power supply units within a target time period according to task parameters corresponding to at least one server device within a target time period;
[0244] It is understood that if the powered device is a server, the EMU can determine target power parameters for multiple power supply units (PSUs) within a target time period based on the task parameters corresponding to at least one server device within the target time period. This allows the required power to be predicted by analyzing the server's task requirements, such as the computing requirements of a GPU model.
[0245] S2, determining target output currents corresponding to the multiple power supply units according to the target power parameters;
[0246] Based on the target power parameters, the EMU can determine the target output current corresponding to each of the multiple PSUs, thereby ensuring that each PSU can provide the required current to meet the power requirements of the server.
[0247] In step S3, if the sum of the target output currents of the multiple power supply units meets the peak power compensation conditions, the multiple peak power supply units are charged. If the sum of the target output currents of the multiple PSUs meets the peak power compensation conditions, the EMU will charge the multiple PLSs. This ensures that the PLSs have sufficient energy reserves to cope with possible power demand peaks.
[0248] S4, adjusting the voltages of the multiple peak power supply units to a reference voltage, wherein the reference voltage is lower than a target voltage corresponding to the power supply bus.
[0249] In the above S4, the reference voltage is lower than the target voltage corresponding to the power supply bus so that the PLS can respond quickly and provide additional power when power supply compensation is required, while avoiding excessive voltage shock to the power supply system.
[0250] In a specific service architecture, such as Figure 6 As shown, the EMU can establish communication connections with each server unit, PSU, and PLS, allowing the EMU to pre-estimate the computing needs of the GPU model within the server, predict power consumption, and allocate power units and compensation units in advance. This communication connection also allows the EMU to comprehensively analyze power demand, voltage and current status, and the operating status of the GPU within the server over a period of time, thereby predicting the next peak power state in advance.
[0251] To achieve highly dynamic high-power compensation response, the EMU can instruct the PLS to precharge the output port voltage to a value lower than the voltage of the entire cabinet power bus as a preparatory process. When peak power demand is detected, the PLS can respond quickly and provide the required additional power.
[0252] In addition, the PSU communicates with the PLS through the EMU, transmitting its rated load capacity and peak power capability to the PLS before power is turned on. This allows the PLS to preset the PSU's rated power capability and initiate power compensation when the rated power capability is exceeded.
[0253] Through the above-described implementation of the present application, information exchange between the entire cabinet energy management unit and the server unit allows for charging and discharging power assessments based on voltage and current, and the difference can be used to predict the next step of peak power compensation. Furthermore, based on the GPU operating status and the model's estimated power demand, information can be communicated to the PLS unit in advance, prompting the PLS to perform preprocessing and achieve faster dynamic response capabilities.
[0254] The embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned power supply method embodiments. The electronic device may be a terminal device or a server. This embodiment is described by taking the electronic device as a mobile phone or a computer as an example. Figure 9As shown, the electronic device includes a memory 902 and a processor 904. The memory 902 stores a computer program, and the processor 904 is configured to execute the steps in any of the above method embodiments through the computer program.
[0255] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.
[0256] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:
[0257] S1, obtaining the output current corresponding to each of multiple power supply units in the current entire cabinet, wherein the multiple power supply units are connected in parallel, and the first end of each of the multiple power supply units is connected to a power bus, which is used to power at least one electrical device;
[0258] S2, when the sum of the output currents corresponding to the multiple power supply units meets the peak power compensation condition, determining the target compensation current according to the current difference between the sum of the output currents and the target rated current;
[0259] S3: Operate multiple peak power supply units in the current entire cabinet, wherein the multiple peak power supply units are connected to the power supply bus, and the total output current corresponding to the multiple peak power supply units is the target compensation current.
[0260] Alternatively, those skilled in the art will appreciate that Figure 9 The structure shown is for illustration only, and the electronic device may also be a terminal device such as a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), or a PAD. Figure 9 It does not limit the structure of the above electronic device. For example, the electronic device may also include Figure 9 More or fewer components (such as network interfaces, etc.) as shown in, or with Figure 9 Different configurations shown.
[0261] Among them, the memory 902 can be used to store software programs and modules, such as the program instructions / modules corresponding to the power supply method and device in the embodiment of the present application. The processor 904 executes various functional applications and data processing by running the software programs and modules stored in the memory 902, that is, realizing the above-mentioned power supply method. The memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 902 may further include a memory remotely arranged relative to the processor 904, and these remote memories can be connected to the terminal via a network. Examples of the above-mentioned networks include but are not limited to the Internet, corporate intranets, local area networks, mobile communication networks and combinations thereof. This example will not be repeated.
[0262] Optionally, the transmission device 906 is configured to receive or transmit data via a network. Specific examples of the aforementioned network may include wired networks and wireless networks. In one embodiment, the transmission device 906 includes a network interface controller (NIC), which can be connected to other network devices and a router via a network cable to enable communication with the Internet or a local area network. In one embodiment, the transmission device 906 is a radio frequency (RF) module configured to communicate with the Internet wirelessly.
[0263] In addition, the electronic device further includes: a display 908 for displaying an operation interface of the power supply system; and a connection bus 910 for connecting various module components in the electronic device.
[0264] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps of any of the above power supply method embodiments when running.
[0265] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.
[0266] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any of the above power supply method embodiments are implemented.
[0267] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0268] The above is a detailed introduction to a power supply system, a power supply method and a whole cabinet provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.
Claims
1. A power supply system, characterized in that: Applied to the entire cabinet, including energy management unit, multiple power supply units, multiple peak power supply units, current state bus and power supply bus; A plurality of power supply units are connected to the power supply bus and are used to supply power to at least one electrical device through the power supply bus; The plurality of peak power supply units are connected to the power supply bus and are configured to supply power to at least one electrical device through the power supply bus upon receiving a power supply compensation instruction sent by the energy management unit. The plurality of peak power supply units are further connected through a first current sharing bus, which is configured to implement dynamic current sharing control among the plurality of peak power supply units. The energy management unit is connected to the multiple power supply units via the current state bus, the current state bus is used to obtain the output current corresponding to each of the multiple power supply units, and the energy management unit is used to send the power supply compensation instruction to the multiple peak power supply units when the sum of the output currents corresponding to each of the multiple power supply units meets the peak power compensation condition; The energy management unit is further configured to determine, based on status information fed back by the current status bus, the number n of the power supply units currently in operation, where n is an integer greater than 1 and less than or equal to M*N, M and N are integers greater than or equal to 1, the entire cabinet includes M power distribution units, a first end of each power distribution unit is connected to the input power bus, and a second end of each power distribution unit is connected to N power supply units; and the sum of the output currents is determined based on the product of the number n of units and the output currents; The first current-sharing bus is also used to obtain the unit number m of the peak power supply units in the standby state, wherein m is an integer greater than 1 and less than or equal to N+L, L is an integer greater than or equal to 1, and the entire cabinet includes N+L peak power supply units; when the m peak power supply units in the standby state are switched to the running state, the unit output current corresponding to each of the m peak power supply units in the running state is determined according to the ratio between the target compensation current and the unit number m.
2. The power supply system according to claim 1, characterized in that: The energy management unit is configured to determine a target compensation current according to a current difference between the sum of the output currents and a target rated current; and send the power supply compensation instruction to the plurality of peak power supply units according to the target compensation current; The first current-sharing bus is used to determine the unit output current corresponding to each of the multiple peak power supply units according to the target compensation current.
3. The power supply system according to claim 2, characterized in that: A communication connection is established between the energy management unit and the first current sharing bus, and the energy management unit is used to obtain the energy status corresponding to each of the multiple peak power supply units according to the communication connection; The energy management unit is further configured to: When the stored electrical energy corresponding to each of the plurality of peak power supply units is less than or equal to a first threshold, sending a first control instruction, wherein the first control instruction is used to reduce the output power corresponding to each of the plurality of power supply units; When the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, a second control instruction is sent, wherein the second control instruction is used to charge the plurality of peak power supply units.
4. The power supply system according to claim 3, characterized in that: When the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, the energy management unit is configured to: When the current difference between the sum of the output currents and the target rated current is greater than or equal to a reference charging current, determining the unit charging current corresponding to each of the m peak power supply units in operation according to a ratio of the reference charging current to m; When the current difference between the sum of the output currents and the target rated current is less than the reference charging current, the unit charging current corresponding to each of the m peak power supply units in operation is determined according to the ratio of the current difference to m.
5. The power supply system according to claim 2, characterized in that: The energy management unit is configured to: Obtaining the target rated current, wherein the target rated current is determined according to the unit rated currents of the plurality of power supply units; In a case where the sum of the output currents is greater than or equal to a reference rated current, determining that the sum of the output currents satisfies the peak power compensation condition, wherein the reference rated current is determined according to the target rated current; The current difference between the sum of the output currents and the target rated current is determined as the target compensation current.
6. The power supply system according to claim 5, characterized in that: The plurality of power supply units are connected via a second current sharing bus; The current state bus is used to obtain the state voltage detected for the second current sharing bus; and determine the output current corresponding to each of the multiple power supply units according to the state voltage and a preset mapping relationship.
7. The power supply system according to any one of claims 1 to 6, characterized in that: Multiple power supply units are arranged in the first area of the entire cabinet, multiple power supply units are connected in parallel, and the first ends of each of the multiple power supply units are connected to the power supply bus; multiple peak power supply units are arranged in the second area of the entire cabinet.
8. The power supply system according to any one of claims 1 to 6, characterized in that: A communication connection is established between the energy management unit and at least one server device; The energy management unit is further configured to: determining target power parameters of the plurality of power supply units within the target time period according to a task parameter corresponding to at least one of the server devices within the target time period; Determining target output currents corresponding to each of the plurality of power supply units according to the target power parameter; When the sum of the target output currents corresponding to the plurality of power supply units satisfies the peak power compensation condition, performing a charging operation on the plurality of peak power supply units; The voltages of the plurality of peak power supply units are adjusted to a reference voltage, wherein the reference voltage is lower than a target voltage corresponding to the power supply bus.
9. A power supply method, characterized in that: include: Obtaining the output current corresponding to each of the multiple power supply units in the current entire cabinet through the current state bus, wherein the multiple power supply units are connected to the current state bus, and the multiple power supply units are connected to the power supply bus, the power supply bus is used to supply power to at least one electrical device, and the multiple power supply units are connected through the second current sharing bus; Determining the number n of the power supply units currently in operation based on the state information fed back by the current state bus, where n is an integer greater than 1 and less than or equal to M*N, M and N are integers greater than or equal to 1, the entire cabinet currently includes M power distribution units, a first end of the power distribution unit is connected to the input power bus, and a second end of the power distribution unit is connected to N power supply units; determining the sum of the output currents based on the product of the number n of units and the output current; When the sum of the output currents corresponding to the plurality of power supply units meets the peak power compensation condition, sending a power supply compensation instruction to the plurality of peak power supply units in the current entire cabinet; Running a plurality of the peak power supply units, wherein the plurality of the peak power supply units are connected to the power supply bus, the plurality of the peak power supply units supply power to at least one electrical device through the power supply bus, and the plurality of the peak power supply units are further connected to each other through a first current sharing bus, the first current sharing bus being used to determine the unit output current corresponding to each of the plurality of peak power supply units according to the target compensation current; Determining the unit output current corresponding to each of the multiple peak power supply units according to the target compensation current, including: obtaining the unit number m of the peak power supply units in the standby state, wherein m is an integer greater than 1 and less than or equal to N+L, L is an integer greater than or equal to 1, and the current entire cabinet includes N+L peak power supply units; when the m peak power supply units in the standby state are switched to the running state, determining the unit output current corresponding to each of the m peak power supply units in the running state according to the ratio between the target compensation current and the unit number m.
10. The method according to claim 9, characterized in that The sending of power supply compensation instructions to the multiple peak power supply units in the current entire cabinet includes: determining a target compensation current according to a current difference between the sum of the output currents and a target rated current; The power supply compensation instruction is sent to the plurality of peak power supply units according to the target compensation current.
11. The method according to claim 10, characterized in that After running a plurality of the peak power supply units, the method further includes at least one of the following: When the stored electrical energy corresponding to each of the plurality of peak power supply units is less than or equal to a first threshold, sending a first control instruction, wherein the first control instruction is used to reduce the output power corresponding to each of the plurality of power supply units; In a case where the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, a second control instruction is sent, wherein the second control instruction is used to charge the plurality of peak power supply units.
12. The method according to claim 11, characterized in that When the stored electric energy corresponding to each of the plurality of peak power supply units is less than or equal to a second threshold, the method includes: When the current difference between the sum of the output currents and the target rated current is greater than or equal to a reference charging current, determining the unit charging current corresponding to each of the m peak power supply units in operation according to a ratio of the reference charging current to m; When the current difference between the sum of the output currents and the target rated current is less than the reference charging current, the unit charging current corresponding to each of the m peak power supply units in operation is determined according to the ratio of the current difference to m.
13. The method according to claim 10, characterized in that The determining the target compensation current according to the current difference between the sum of the output currents and the target rated current includes: Obtaining the target rated current, wherein the target rated current is determined according to the unit rated currents of the plurality of power supply units; In a case where the sum of the output currents is greater than or equal to a reference rated current, determining that the sum of the output currents satisfies the peak power compensation condition, wherein the reference rated current is determined according to the target rated current; The current difference between the sum of the output currents and the target rated current is determined as the target compensation current.
14. The method according to claim 13, characterized in that Before determining the target compensation current according to the current difference between the sum of the output currents and the target rated current, the method further includes: When the plurality of power supply units are connected via a second current sharing bus, obtaining a state voltage detected on the current sharing bus according to the current state bus; The output current corresponding to each of the plurality of power supply units is determined according to the state voltage and a preset mapping relationship.
15. The method according to any one of claims 9 to 14, characterized in that Before obtaining the output currents corresponding to the multiple power supply units in the entire cabinet through the current status bus, the following steps are also included: In a case where at least one of the power-consuming devices is at least one server device, establishing a communication connection with the at least one server device; Acquiring, through the communication connection, task parameters corresponding to at least one of the server devices within a target time period; Determining target power parameters of the plurality of power supply units within the target time period according to the task parameters; Determining target output currents corresponding to each of the plurality of power supply units according to the target power parameter; When the sum of the target output currents corresponding to the plurality of power supply units satisfies the peak power compensation condition, performing a charging operation on the plurality of peak power supply units; The voltages of the plurality of peak power supply units are adjusted to a reference voltage, wherein the reference voltage is lower than a target voltage corresponding to the power supply bus.
16. A complete cabinet, characterized in that: The power supply system comprises the power supply system according to any one of claims 1 to 8, wherein the power supply system supplies power to at least one server device through a power supply bus.
Citation Information
Patent Citations
Power supply circuit, server and time sequence control method
CN119472960A