Method and system for dynamic regulation of power supply of machine room based on load safety threshold

Abstract

The application provides a machine room power supply dynamic regulation method and system based on a load safety threshold, relates to the technical field of machine room power supply regulation, and comprises the following steps: collecting phase current, terminal temperature, inlet air temperature, equivalent series resistance of electrolytic capacitor, phase line joint impedance and voltage distortion rate in multiple time scales; calculating comprehensive load safety threshold and adjustable margin according to five types of reduction factors and breaker cooperation margin; dividing normal area, early warning area, intervention area and hard limit area by comparing current with the threshold; respectively executing light kneading, phase shift harmonic injection or emergency duty cycle clamping and other electric signal modulation in each area; starting hierarchical migration when entering the intervention area or the hard limit area, and recalculating the temporary threshold by combining with joint impedance reduction of the receiving side; and grading discharging the super capacitor and the lithium battery when the current impact slope is out of limit, and the migration is linked. The application makes the threshold dynamically slide with aging, temperature rise, impedance drift and harmonic, captures the branch out-of-limit trend in advance, and avoids redundant failure.

Description

Technical Field

[0001] This invention relates to the field of power control technology for computer rooms, and in particular to a method and system for dynamic power control of computer rooms based on load safety thresholds. Background Technology

[0002] With the continuous expansion of cloud computing and big data services, the power density of a single rack in data center server rooms has increased from 2-3kW in the early days to 10kW or even higher. To ensure power continuity, most medium and large-sized server rooms adopt a dual-redundant power supply architecture, that is, each rack is equipped with two independent power distribution links, which are connected to different upstream transformers or UPS output buses. Under normal circumstances, each link carries about 50% of the load. When one link is interrupted due to maintenance or failure, the other link should be able to take over the entire load without triggering the protection device. This design theoretically reserves sufficient fault tolerance margin. However, in actual operation, abnormal tripping accidents frequently occur: the total load rate meter of a certain rack showed that it was only about 60% of the rated capacity, and all power quality indicators were within the qualified range. However, during the nighttime batch data backup period, one of the sub-distribution power units of this rack suddenly tripped, causing dozens of virtual machines running on it to go offline instantly. Even more seriously, when the redundant link automatically takes over the load of the tripped link, since the aging of components and the cumulative effect of temperature rise are not taken into account, the redundant link also approaches the tripping threshold within a few seconds, and the power supply of the entire line faces the risk of complete interruption.

[0003] Subsequent disassembly and analysis revealed the root cause of the failure: half of the server power modules in the rack had been running continuously for over four years, and the equivalent series resistance of their internal electrolytic capacitors had increased from 0.015Ω at the factory to over 0.025Ω. This resulted in a significant increase in conduction and switching losses within the modules under the same load current, and a faster rate of temperature rise in the junction of the power devices. Simultaneously, the terminal blocks experienced micron-level oxidation and relaxation due to long-term thermal cycling, increasing the contact resistance from the initial 0.008Ω to 0.022Ω, causing an additional 8 to 10°C increase in terminal temperature rise under the same branch current. Furthermore, the intake air temperature at the top of the rack reached 35°C due to heat dissipation from adjacent high-density racks, far exceeding the design baseline of 24°C. The combined effect of these factors caused the actual safe current-carrying capacity of this sub-distribution power unit to quietly decrease by nearly 20%. However, the existing power management system still used the fixed 80% load rate threshold set when the equipment was put into operation. It neither detected the thermal degradation caused by capacitor aging, nor incorporated terminal temperature rise and ambient temperature into the dynamic calculation of the safety boundary, nor considered the erosion of current carrying capacity by the increase in connector impedance. When the backup task in the early morning caused a brief current surge, the deteriorated thermal boundary caused the circuit breaker to trip unexpectedly below the nameplate rating. The redundant circuit also lacked dynamic threshold detection and quickly approached its already reduced safety boundary after taking on the load.

[0004] To address the aforementioned challenges, various technologies have been explored in this field. Early solutions followed the approach of redundant power module management, where a baseboard management controller periodically polls the device load rate. When the reading falls to a preset threshold, some power modules are put into sleep mode; conversely, all are woken up to maintain power margin. The advantage of this approach is its simplicity and transparency to upper-layer services. However, its threshold value is a constant pre-fixed in the firmware, making it unable to detect the load-carrying capacity degradation caused by device aging, nor can it respond to the erosion of safety margins by upstream voltage distortion or ambient temperature drift. After three to five years of operation, the 75% or 80% threshold set at the time may have deviated from the actual failure threshold, while the system is still scheduling according to outdated parameters. This is the inherent implicit failure mode of the static threshold strategy. Another approach focuses on the power distribution unit itself, using remote communication and a centralized management platform to compensate for the sluggishness of local control. This approach revolves around the remote detection and control of the power distribution unit, involving technologies such as long-connection-based device status monitoring, heartbeat keep-alive, caching, and unified distribution of scheduled tasks. While this approach solves the problems of limited distance for local manual operation and fragmented protocols, its control actions are still limited to port on / off and planned power-on / off, without addressing the dynamic correction of the load safety threshold itself, and without incorporating the electrothermal response mechanism inside the power module into the decision-making chain. Therefore, when facing short-term impacts at the branch level, the remote platform can only obtain alarms of exceeded limits, rather than trends of impending exceedances. A newer approach introduces a layered architecture and energy storage buffer, such as layered data center hybrid energy storage systems and energy consumption control methods, which extend control granularity to the power distribution unit level, rack level, and server level, with a central controller coordinating energy allocation and dynamic scheduling. Another approach combines DC bus and battery transient intervention to fill the gap between load spikes and power reconfiguration. However, the layers still transmit macroscopic variables such as power demand and energy margin. The microscopic physical quantities that truly determine the safety threshold, such as the junction temperature rise rate of rectifier devices, the aging drift of capacitor equivalent series resistance, and the slight changes in phase impedance due to thermal expansion and contraction, are not fed back into the threshold calculation. Therefore, the layered architecture and threshold calculation are essentially disconnected in the control loop. Some more advanced explorations have introduced reinforcement learning to make sequential decisions on load allocation for distribution units. Topics include reinforcement learning-based load allocation systems and multidimensional execution cost models. These schemes incorporate virtual machine migration, server start-up and shutdown, and battery discharge depth into the reward function, attempting to find the optimal balance between energy consumption and performance. However, in actual deployment, the scheduler still needs to compare the load rate with a set of preset thresholds to determine the action space; the threshold itself remains an offline calibrated constant. The optimization capabilities of intelligent algorithms cannot compensate for the outdated data at the source of thresholds, and reinforcement learning requires a lot of trial and error to adapt to instantaneous impacts on the power supply side and load pattern drift. In scenarios with zero fault tolerance, such as financial-grade data centers, the cost of allowing intelligent agents to explore security boundaries through real business is too high.Therefore, there is an urgent need in the market for a new method and system for dynamic control of data center power supply based on load safety threshold. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for dynamic power control of a data center based on a load safety threshold. This addresses the technical bottlenecks in existing technologies, such as the use of a fixed constant load safety threshold, the inability to dynamically adjust the threshold based on device aging, temperature rise, connector impedance drift, and harmonic pollution, leading to a shrinking actual safety boundary that the system cannot detect. It also addresses the problems in existing solutions, such as static thresholds, remote monitoring, hierarchical scheduling, or reinforcement learning methods failing to integrate device-level electrothermal models, electrical signal modulation, and hierarchical control logic into a closed loop, resulting in a disconnect between thresholds and control, and an inability to suppress branch exceedance trends in advance. The specific technical solution is as follows: This invention provides a method for dynamic power control of a data center based on a load safety threshold, comprising: S10 collects electrical parameters, temperature parameters, capacitor aging parameters, connector impedance parameters, and voltage distortion parameters for each phase of the power distribution unit. S20, based on the collected data, calculate the comprehensive load safety threshold and adjustability margin for each phase; S30, compare the current phase current with the comprehensive load safety threshold and divide it into normal zone, early warning zone, intervention zone and hard limit zone; S40 performs different granularity of electrical signal-level dynamic modulation in the warning zone, intervention zone and hard limit zone, including phase shift fine adjustment, third harmonic injection and pulse width modulation duty cycle kneading, to reduce the current phase current or delay the current rise rate. S50: When any phase of this rack enters the intervention area or hard limit area, a hierarchical logic migration is initiated to migrate the non-real-time service virtual machine of this rack to a rack with ample space in the same row or a rack with ample space across rows. The migration rate is limited by the available margin on the receiving side and the impedance reduction of the connector. S60: When the current rise slope of any phase of the rack exceeds the preset slope threshold, the column top transient buffer unit is triggered to discharge in stages. If the discharge fails to effectively suppress the slope, cross-column migration is initiated.

[0006] Furthermore, in S10, the effective values ​​of phase current, phase voltage, and neutral current of each phase distribution unit are collected at a sampling cycle of 100 milliseconds. The temperature of the copper busbar at the terminal of the distribution unit is collected at a cycle of 1 second. The air intake temperature of the rack is collected at a cycle of 10 seconds. The equivalent series resistance of the electrolytic capacitor of each server power module is measured by a high-frequency detection pulse every 10 minutes. The impedance of each phase line connector is calculated by reverse calculation every 15 minutes. The column head voltage distortion rate is calculated by sampling 128 points per cycle and the maximum value of the 10-minute sliding window is taken.

[0007] Furthermore, in S20, the comprehensive load safety threshold is equal to the rated current carrying capacity multiplied by the terminal temperature reduction factor, the inlet air temperature reduction factor, the capacitor aging reduction factor, the connector impedance reduction factor, the voltage distortion reduction factor, and the circuit breaker matching margin; the adjustable margin is equal to the comprehensive load safety threshold minus the equivalent current reduction value and then minus the current.

[0008] Furthermore, in S30, the current current is less than or equal to 0.85 times the comprehensive load safety threshold, which is the normal zone; the current current is greater than 0.85 times but less than or equal to 0.93 times the comprehensive load safety threshold, which is the warning zone; the current current is greater than 0.93 times but less than or equal to 0.97 times the comprehensive load safety threshold, which is the intervention zone; and the current current is greater than 0.97 times the comprehensive load safety threshold, which is the hard limit zone.

[0009] Furthermore, the high-rate pulse command in S40 is a pulse command with an amplitude greater than 0.8 times the rated power and a duration greater than 10 seconds. Three strategies are adopted according to different ranges of the absolute value of the polarization residual and the recovery time interval: if the absolute value of the polarization residual is less than 0.05 volts and the recovery time interval is greater than or equal to 5 times the polarization time constant, it is issued in full; if the absolute value of the polarization residual is greater than or equal to 0.05 volts and less than 0.15 volts and the recovery time interval is greater than or equal to 5 times the polarization time constant, it is issued after being reduced by a reduction factor; if the absolute value of the polarization residual is greater than or equal to 0.15 volts or the recovery time interval is less than 5 times the polarization time constant, it is prohibited or further reduced.

[0010] Furthermore, in S50, non-real-time business virtual machines in the local rack will be migrated to the same row of spacious racks or across rows, with each virtual machine valued at 200W. The preset slope threshold in S60 is that the rising slope exceeds 7% of the rated current of the phase within every 400 milliseconds. The graded discharge includes first discharging by the supercapacitor at a rate of 1.2 times the impulse current for a maximum of 10 seconds. If the slope still does not drop below 2% of the rated current within every 400 milliseconds after 10 seconds, it will switch to lithium battery discharge and start cross-column migration at the same time.

[0011] Furthermore, the S30 also includes: S301 compares the current phase current with the comprehensive load safety threshold. If the current current is less than or equal to 0.85 times the comprehensive load safety threshold, it is determined to be in the normal zone. No modulation or migration action is performed, and only routine monitoring is maintained. S302, if the current current is greater than 0.85 times and less than or equal to 0.93 times the comprehensive load safety threshold, it is determined to be in the warning zone and enters the electrical signal mild modulation stage of S40; S303, if the current current is greater than 0.93 times and less than or equal to 0.97 times the comprehensive load safety threshold, it is determined to be an intervention zone, and enters the heavy modulation stage of electrical signal in S40, while triggering the hierarchical migration preparation in S50. S304 If the current current is greater than 0.97 times the comprehensive load safety threshold, it is determined to be a hard limit zone, and the emergency modulation of S40 and the forced migration of S50 are immediately started, and the buffer unit of S60 is activated.

[0012] Furthermore, the S40 also includes: S401, when in the warning zone, performs light modulation: the outputs connected to the current phase are re-proportionally redistributed according to the current load rate of each output. Specifically, the load ratio of the two power modules of each server is equal to the current of the first output divided by the current of the second output. For servers with a load ratio greater than 1.2, a pulse width modulation reference fine-tuning command is issued, causing the duty cycle of the first output to decrease by 0.5% per step and the duty cycle of the second output to increase accordingly. For servers with a load ratio less than 0.8, a symmetrical reverse squeezing is performed. The squeezing amplitude of a single squeeze does not exceed 3% of the rated current of the output, and the interval between two squeezings is not less than 400 milliseconds. S402, when in the intervention zone, adds phase shift fine-tuning and harmonic injection to S401: Through the bidirectional thyristor at the output of the distribution unit, the reference phase of each phase voltage relative to the grid phase is adjusted in steps of 0.3 cm within a ±4.5 degree range. The adjustment goal is to minimize the variance of the imaginary part of the input impedance of each power source under that phase. Simultaneously, when the neutral current exceeds 15% of the phase current and the third harmonic content exceeds 12% of the fundamental frequency, a signal generator within the distribution unit superimposes an amplitude of 2.8 times the phase voltage. The third harmonic component with a phase relative to the fundamental wave of 180 degrees is observed after 300 milliseconds. If the decrease is greater than 18%, the injection parameters are retained as steady-state compensation. If the decrease is less than 8%, the injection is stopped and the step size is changed to 0.7 minutes for retesting. The total adjustment of phase shift and harmonic injection is converted into the equivalent current reduction value. The equivalent current reduction value is equal to the sum of the phase shift adjustment value and the harmonic injection adjustment value multiplied by 0.6, and the equivalent current reduction value in S20 is updated. S403, when in a hard limit zone, performs emergency duty cycle clamping based on S402: temporarily reduces the upper limit of the pulse width modulation duty cycle of all servers under this phase from the rated value to 70% of the rated value, and at the same time switches the output relay of the distribution unit of this phase to the backup circuit. The switching sequence is to first reduce the duty cycle to 30%, then disconnect the main circuit relay, and after a delay of 20 milliseconds, close the backup circuit relay to restore the duty cycle with a ramp of 0.3% per millisecond.

[0013] Furthermore, the S50 also includes: S501, when any phase of this rack enters the intervention zone, the rack controller sends a migration request to the co-row controller. The request includes the adjustability margin of that phase in this rack, as well as the power estimate of each virtual machine in this rack. S502, the same-row controller collects all 72 adjustable margin values ​​of the 12 racks in its row at a 1-second interval. These adjustable margin values ​​are sorted from smallest to largest. The set of racks with negative adjustable margins is defined as the scarce side, and the set of racks with adjustable margins greater than 28% of the rack's rated current is defined as the abundant side. The abundant side is then queued for migration in descending order of adjustable margin value. Migration decisions follow the principle of prioritizing racks within the same row and providing a fallback across rows: when the sum of the absolute values ​​of negative adjustable margin values ​​of a rack on the scarce side is less than 70% of the total available adjustable margin in the same row on the abundant side, migration is completed only within that row; if this ratio is exceeded, the row controller sends a cross-row request to the room-level central controller. S503, the room-level central controller performs secondary convergence across all 8 columns in the room at a 3-second interval. During secondary convergence, the impedance of the upstream phase line connectors of each column is additionally considered. For the column to be transferred, the temporary connector impedance reduction factor of all distribution unit branches in the column is first calculated based on the impedance of each phase connector. Then, the temporary connector impedance reduction factor is used to replace the original connector impedance reduction factor to recalculate the temporary comprehensive load safety threshold and temporary adjustable margin of each phase in the column. The column is only allowed to transfer loads if the temporary adjustable margin is greater than 1.3 times the current to be transferred. The migration process employs a micro-step sequence of first decreasing, then switching, and then increasing: For the server to be migrated, the first power limit is reduced by 40W every 200 milliseconds. When the power limit drops to 30% of the rated power, the first output relay is disconnected. After a 20-millisecond delay, the server's second power supply performs a power factor correction soft start, with the soft start ramp increasing by 0.3% duty cycle every millisecond, for a total duration not exceeding 800 milliseconds. During the soft start, the central control room temporarily increases the second power supply's overall load safety threshold by 15% as a switching tolerance window. After the switching is complete, the tolerance window is removed 3 seconds later. The S60 also includes: S601: When the current in any phase of the rack rises at a rate greater than 7% of the rated current of that phase within 400 milliseconds, it is determined to be a short-term impact. The central controller at the room level will not schedule the migration, but will immediately command the supercapacitor at the top of the column to discharge at a rate of 1.2 times the current impact current, with a maximum discharge time of 10 seconds. During the discharge, the comprehensive load safety threshold of that phase is temporarily amplified by 8% as a buffer tolerance to avoid the buffer action and the threshold warning conflicting in the cycle time. S602, if the inrush current slope is still greater than the rated 4% every 400 milliseconds after 10 seconds, the lithium battery will continue to discharge, and the lithium battery discharge current will be 80% of the current when the supercapacitor terminates. At the same time, cross-row migration will be initiated. If the inrush slope drops to below the rated 2% every 400 milliseconds within 10 seconds, the supercapacitor will stop discharging, the buffer tolerance will be cancelled, and the reinforcement learning module in the room-level center will temporarily reduce the capacitor aging reduction factor of this phase of the rack by 0.03, marking that the rack may have business mode drift in the near future. S603: If the lithium battery discharge duration exceeds 5 minutes or the lithium battery state of charge drops below 20%, all non-critical loads in that column will be forcibly disconnected, and an operation and maintenance alarm will be issued. The internal resistance of the buffer unit is transmitted back to the central control room every 30 seconds. When the internal resistance increases by more than 35% compared to a new product, the maximum discharge rate of the buffer unit in that column will be reduced from 3.5 times to 2.2 times. At the same time, the central control room will add a coefficient of 0.96 to the impedance reduction factor of the connectors of all power distribution unit branches in that column.

[0014] This invention also provides a dynamic power control system for computer rooms based on a load safety threshold, used to implement the method described above. The system includes: The sensor array and data acquisition module are used to place current transformers, voltage transformers, negative temperature coefficient temperature sensors and ambient temperature probes at the output terminals, wiring terminals and air inlets on the top of each rack in the computer room of the dual-path power distribution unit. A parallel acquisition link including a high-frequency detection pulse injection and response acquisition unit is built after the DC-DC converter of each server power module to synchronously acquire the effective value of each phase current, effective value of phase voltage, effective value of neutral line current, temperature of copper busbar at wiring terminal, air inlet temperature of rack, equivalent series resistance of electrolytic capacitor, impedance of phase line connector and voltage distortion rate of column head. The signal conditioning and online calibration module is used to dynamically modulate, decouple, and compensate the acquired raw mixed electrical signals. Specifically, it uses a digital quadrature phase-locked loop demodulation algorithm to extract the response characteristics of high-frequency detection pulses in real time to quantify transmission link distortion, and applies a wavelet transform-based noise reduction algorithm for secondary filtering, thereby generating high-fidelity physical quantities such as terminal temperature, air inlet temperature, equivalent series resistance, connector impedance, and voltage distortion rate. The real-time load safety threshold calculation module is used to calculate the real-time threshold of each phase according to the high-fidelity physical quantity, the comprehensive load safety threshold formula and its sub-formulas, and to calculate the adjustable margin. The electrical signal level dynamic modulation module is used to perform mild kneading modulation, phase shift and harmonic injection combined modulation, or emergency duty cycle clamping and redundancy switching according to the warning zone, intervention zone, or hard limit zone where the current phase current is located, and output the equivalent current reduction value to the threshold calculation module. The hierarchical logic migration module is used to recalculate the temporary threshold according to the principle of priority in the same column and fall back across columns when any phase of the rack enters the interference zone or hard limit zone, and to perform virtual machine migration, and accelerate the migration within the switching tolerance window. The transient buffer and threshold linkage module is used to control the supercapacitor and lithium battery at the top of the column to discharge in stages when the current of any phase of the rack rises at a rate exceeding 7% of the rated current within 400 milliseconds. During the discharge, the overall load safety threshold is temporarily amplified, and the connector impedance reduction coefficient is adjusted according to the aging of the internal resistance of the buffer unit. The reinforcement learning threshold fine-tuning module is used to fine-tune five reduction coefficients within a strictly defined action range using short-term impact events as samples, and then superimposed into the threshold calculation after being centrally verified at the room level. The full lifecycle calibration and management module is used to compare the drift of each reduction factor with the factory value in a 30-day window, trigger maintenance work orders, and perform degradation and threshold compensation when the internal resistance of the buffer unit rises by more than 35%.

[0015] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, performs the steps of the method described herein.

[0016] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method.

[0017] The beneficial effects of this invention are as follows: Firstly, by establishing a comprehensive load safety threshold calculation model that includes terminal temperature reduction coefficients, inlet air temperature reduction coefficients, capacitor aging reduction coefficients, connector impedance reduction coefficients, voltage distortion reduction coefficients, and circuit breaker matching margins, this invention transforms the load safety threshold from an offline, fixed constant into a dynamic variable that slides in real-time with device aging, temperature rise, impedance drift, and harmonic contamination. This fundamentally eliminates the risk of branch over-limit and redundant failure caused by the static threshold blind zone. Secondly, by directly feeding the reduction factor output in real-time from the device-level electrothermal model into the electrical signal-level modulation stage, a positive feedback closed loop of "modulation changes current, current affects threshold, and threshold guides modulation" is formed. This achieves gradient intervention from mild squeezing to phase-shift harmonic injection to emergency duty cycle clamping, enabling the system to intervene early when the load is slowly increasing and act decisively during rapid impacts, avoiding oscillations or hysteresis caused by a single threshold. Thirdly, by... In the hierarchical migration process, impedance reduction of the receiving-side connector is introduced to dynamically correct temporary thresholds. Combined with micro-step switching sequences and a switching tolerance window mechanism, this effectively prevents the receiving-side from exceeding its limits due to migration loads and current surges caused by traditional hard switching. In addition, by linking the transient buffer unit with the threshold, the threshold is temporarily amplified to accommodate buffer discharge when an impact occurs. At the same time, the aging information of the buffer device's internal resistance is fed back to the threshold calculation, forming a closed loop of the entire life cycle of "buffer discharge, threshold tolerance, device aging, and threshold correction". Finally, through the reinforcement learning module to make minor corrections to the reduction coefficient within a strictly limited range of actions, and with full life cycle calibration management with a 30-day window, the system can adapt to business mode drift and trigger maintenance processes in a timely manner. Thus, under the conditions of severe load fluctuations and long-term aging in the data center, it achieves a control effect far exceeding the linear superposition of individual solutions, improving the safety and reliability of the data center power supply system.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram illustrating the steps of a dynamic power control method for a computer room based on a load safety threshold according to the present invention. Figure 2 This is a schematic diagram of the structure of a dynamic power control system for a computer room based on a load safety threshold, according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0022] It should also be noted that when a component is described as "fixed to" or "set on" another component, it can be directly on the other component or there may be an intervening component present. When a component is described as "connected to" another component, it can be directly connected to the other component or there may be an intervening component present.

[0023] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0024] In an embodiment of the present invention, a method for dynamic control of data center power supply based on a load safety threshold is provided. Please refer to [link to relevant documentation]. Figure 1The method includes the following steps: S10 collects the effective values ​​of phase current, phase voltage, and neutral current of each phase distribution unit at a sampling cycle of 100 milliseconds, collects the copper busbar temperature of the distribution unit terminal at a cycle of 1 second, collects the rack air intake temperature at a cycle of 10 seconds, measures the equivalent series resistance of the electrolytic capacitor of each server power module with a high-frequency detection pulse every 10 minutes, calculates the impedance of each phase line connector by reverse calculation every 15 minutes, and calculates the column head voltage distortion rate by sampling 128 points per cycle and taking the maximum value of the 10-minute sliding window. S20, based on the data collected in S10, calculates the comprehensive load safety threshold and adjustable margin for each phase; the comprehensive load safety threshold in S20 is equal to the rated current carrying capacity multiplied by the terminal temperature reduction factor multiplied by the inlet air temperature reduction factor multiplied by the capacitor aging reduction factor multiplied by the connector impedance reduction factor multiplied by the voltage distortion reduction factor multiplied by the circuit breaker matching margin; the adjustable margin is equal to the comprehensive load safety threshold minus the equivalent current reduction value minus the current.

[0025] S30 compares the current phase current with the comprehensive load safety threshold and divides it into four working zones: the normal zone when the current current is less than or equal to 0.85 times the comprehensive load safety threshold, the warning zone when the current current is greater than 0.85 times but less than or equal to 0.93 times the comprehensive load safety threshold, the intervention zone when the current current is greater than 0.93 times but less than or equal to 0.97 times the comprehensive load safety threshold, and the hard limit zone when the current current is greater than 0.97 times the comprehensive load safety threshold. S40 performs different granularity of electrical signal-level dynamic modulation in the warning zone, intervention zone and hard limit zone, including phase shift fine adjustment, third harmonic injection and pulse width modulation duty cycle kneading, to reduce the current phase current or delay the current rise rate. S50, when any phase of the rack enters the intervention area or hard limit area, a hierarchical logic migration is initiated, and the non-real-time service virtual machines of the rack are migrated to the same row of spacious racks or cross-row spacious racks at a rate of 200W per virtual machine. The migration rate is limited by the available margin on the receiving side and the impedance reduction of the connector. S60: When the current of any phase in the rack rises at a rate exceeding 7% of the rated current of that phase within 400 milliseconds, the transient buffer unit at the top of the column is triggered to discharge. First, the supercapacitor discharges at a rate of 1.2 times the surge current for a maximum of 10 seconds. If the rate of rise does not drop below 2% of the rated current every 400 milliseconds after 10 seconds, the lithium battery will continue to discharge and cross-column migration will be initiated at the same time.

[0026] The principle and effect of the above technical solution are as follows: This invention provides a dynamic power control method for computer room based on load safety threshold. By integrating the reduction factor calculated in real time by the device-level electrothermal model with the electrical signal-level modulation and hierarchical logic into a single closed loop, the load safety threshold is no longer an offline fixed constant, but a dynamic variable that slides with aging, temperature rise, impedance drift, and harmonic pollution. Thus, under computer room conditions of severe load fluctuation and long-term aging, the trend of branch-level exceeding the limit can be captured in advance and actively controlled, avoiding the failure of redundant design due to the blind zone of static threshold, and achieving a control effect far exceeding the linear superposition of individual solutions.

[0027] In one embodiment, S10 includes: S101, install current transformers and voltage transformers at each phase output terminal of each power distribution unit, and collect the effective values ​​of phase current, phase voltage, and neutral current at a period of 100 milliseconds, with a sampling accuracy of not less than 0.5 class; S102, a negative temperature coefficient temperature sensor is installed at each phase terminal of each power distribution unit to collect the copper busbar temperature in a 1-second cycle, with a temperature range of -20℃ to 85℃ and an accuracy of ±0.5℃. S103: An ambient temperature probe is installed at the air inlet at the top of each rack to collect the air inlet temperature every 10 seconds with an accuracy of ±0.3℃. S104, after the DC-DC converter of each server power module, injects a high-frequency detection pulse every 10 minutes with a pulse width of 100 nanoseconds and an amplitude of 5% of the rated output voltage through firmware. The equivalent series resistance of the current electrolytic capacitor is calculated using the ratio of pulse voltage drop to pulse current, in ohms. S105, install a voltage transformer at the incoming line of each column head, calculate the voltage distortion rate by sampling 128 points per cycle, and maintain a 10-minute sliding window. The maximum value within the window is recorded as the maximum voltage distortion rate, in percentage. S106, at each column head inlet, by back-calculating the total current and phase voltage drop every 15 minutes, calculate the phase line connector impedance of that column and phase, which is equal to the phase voltage drop divided by the total current, in ohms, and record the historical minimum connector impedance of that phase since its first power-on.

[0028] The principle and effect of the above technical solution are as follows: by collecting data in layers with multiple time scales and multiple physical quantities, real-time input from the reaction principle layer and the environment layer is provided for subsequent threshold deduction, avoiding the one-sidedness of traditional solutions that rely solely on total power or a single temperature.

[0029] The terminal temperature reduction factor in S20 is calculated using a piecewise linear formula based on the terminal temperature; the inlet air temperature reduction factor is calculated using a piecewise linear formula based on the inlet air temperature; the capacitor aging reduction factor is calculated using a cutoff linear formula based on the aging factor; the connector impedance reduction factor is calculated using a cutoff linear formula based on the impedance increment; the voltage distortion reduction factor is calculated using a piecewise linear formula based on the maximum voltage distortion rate; and the circuit breaker matching margin is always taken as 0.92. The initial equivalent current reduction value is 0.

[0030] In one embodiment, S20 includes: S201, for each phase, read the rated current carrying capacity, terminal temperature, air inlet temperature, current equivalent series resistance, new module nominal equivalent series resistance, current connector impedance, historical minimum connector impedance, and maximum voltage distortion rate obtained in S10, and calculate the terminal temperature reduction factor, air inlet temperature reduction factor, capacitor aging reduction factor, connector impedance reduction factor, and voltage distortion reduction factor in sequence, where the circuit breaker matching margin is always taken as 0.92; S202, multiply the six reduction factors together and then multiply by the rated current carrying capacity to obtain the comprehensive load safety threshold of the phase, in amperes; S203 reads the current phase current and the equivalent current reduction value generated by the current electrical signal modulation, calculates the adjustable margin, which is equal to the comprehensive load safety threshold minus the equivalent current reduction value minus the current, in amperes. A positive adjustable margin indicates that there is still room for adjustment, while a negative adjustable margin indicates that the warning zone has been entered.

[0031] The formula for calculating the comprehensive load safety threshold is I. 综合 =I max ·K t ·K a ·K g ·K z ·K v ·K sel , among which, I 综合 This is the comprehensive load safety threshold for this phase, in A and I. max The rated current carrying capacity of this phase is expressed in amperes (A), and is given on the nameplate of the distribution unit. (K) t K is the terminal temperature reduction factor. a K is the inlet air temperature reduction factor. g K is the capacitor aging reduction factor. z K is the impedance reduction factor for the connector. v K is the voltage distortion reduction factor. sel The circuit breaker matching margin is always taken as 0.92.

[0032] The terminal temperature reduction factor reflects the attenuation of current-carrying capacity due to the increase in temperature of the copper busbar at the terminal, and is described by a three-segment piecewise function: , Where T u The temperature of the copper busbar at the terminal block of the power distribution unit is obtained from S102; 0.012 is obtained by regression from the measured data in the aging room; 0.58 is the lower limit of the reduction value corresponding to the 65℃ temperature node.

[0033] The inlet air temperature reduction factor characterizes the impact of the rack inlet ambient temperature on the equipment's current-carrying capacity. The calculation formula is: , Where T a 0.007 represents the rack inlet air temperature, collected by S103; 0.007 is obtained by ASHRAE TC 9.9 inlet air range extrapolation fitting; 0.70 is the lower limit of reduction corresponding to the 42℃ temperature node.

[0034] The capacitor aging reduction factor characterizes the effect of capacitor performance aging on the safety threshold through the relative change of the equivalent series resistance (ESR) of the electrolytic capacitor. The calculation formula is as follows: , ; Among them, ESR 实 The measured equivalent series resistance (ESR) of the current electrolytic capacitor was obtained from S104. 新 α is the nominal equivalent series resistance of the new module; α is the dimensionless aging factor; 0.18 is the sensitivity coefficient of the aluminum electrolytic capacitor calculated from the manufacturer's life curve at typical junction temperature in the computer room; 0.72 is the lower limit of reduction corresponding to aging factor α=2.2.

[0035] The connector impedance reduction factor reflects the impact of the increase in phase connector contact impedance on the deviation of the safety threshold. The calculation formula is: , ; in The current phase connector impedance is derived from S106; 1. The historical minimum connector impedance for this phase, both in Ω; 0.018Ω is the common contact impedance rise reference value after 500 thermal cycles for a 100A-class terminal block; 0.35 is the regression sensitivity of connector temperature rise and safety threshold deviation; 0.70 is the reduction lower limit value corresponding to ΔZc=0.018Ω.

[0036] The voltage distortion reduction factor characterizes the nonlinear attenuation effect of voltage harmonic distortion on current-carrying capacity. First, the maximum voltage distortion rate within a 10-minute time window is statistically analyzed, and then the reduction factor is calculated using a piecewise function. , , Among them THD maxThe maximum voltage distortion rate within the 10-minute window obtained from S105 is 0.09, which is obtained statistically from the nonlinear junction temperature jump law of rectifier loads when THD>5%; 0.64 is the 8% cutoff lower limit.

[0037] To achieve protection level coordination and ensure that the operation of the distribution unit takes precedence over that of the upstream feeder, a current margin is reserved between the warning threshold and the thermal tripping range. The circuit breaker coordination margin is taken as a constant value: K sel = 0.92, meaning that in I 综合 Triggered warning zone (0.85I) 综合 An 8% current margin is reserved between the circuit breaker and the upstream molded case circuit breaker thermal tripping zone to form a protection level difference.

[0038] The principle and effect of the above technical solution are as follows: the five types of reduction factors are integrated in the form of multiplication, so that the threshold simultaneously reflects five independent degradation mechanisms: terminal temperature rise, ambient temperature, capacitor aging, connector micro-impedance, and voltage distortion. Moreover, each factor is a dimensionless pure number, which ensures homogeneity of dimensions. The adjustable margin serves as a unified criterion for subsequent modulation and migration decisions, realizing the connection of the three closed loops at the numerical level.

[0039] In one embodiment, S30 includes: S301 compares the current phase current with the comprehensive load safety threshold. If the current current is less than or equal to 0.85 times the comprehensive load safety threshold, it is determined to be in the normal zone. No modulation or migration action is performed, and only routine monitoring is maintained. S302, if the current current is greater than 0.85 times and less than or equal to 0.93 times the comprehensive load safety threshold, it is determined to be in the warning zone and enters the electrical signal mild modulation stage of S40; S303, if the current current is greater than 0.93 times and less than or equal to 0.97 times the comprehensive load safety threshold, it is determined to be an intervention zone, and enters the heavy modulation stage of electrical signal in S40, while triggering the hierarchical migration preparation in S50. S304 If the current current is greater than 0.97 times the comprehensive load safety threshold, it is determined to be a hard limit zone, and the emergency modulation of S40 and the forced migration of S50 are immediately started, and the buffer unit of S60 is activated.

[0040] The principle and effect of the above technical solution are as follows: the threshold is divided into four continuous intervals, each interval corresponding to a different intensity of control measures, so that the system can intervene in advance when the load is slowly rising and act decisively when there is a rapid impact, thus avoiding the oscillation or lag caused by a single threshold.

[0041] In one embodiment, S40 includes: S401, when in the warning zone, performs light modulation: the outputs connected to the current phase are re-proportionally redistributed according to the current load rate of each output. Specifically, the load ratio of the two power modules of each server is equal to the current of the first output divided by the current of the second output. For servers with a load ratio greater than 1.2, a pulse width modulation reference fine-tuning command is issued, causing the duty cycle of the first output to decrease by 0.5% per step and the duty cycle of the second output to increase accordingly. For servers with a load ratio less than 0.8, a symmetrical reverse squeezing is performed. The squeezing amplitude of a single squeeze does not exceed 3% of the rated current of the output, and the interval between two squeezings is not less than 400 milliseconds. S402, when in the intervention zone, adds phase shift fine-tuning and harmonic injection to S401: Through the bidirectional thyristor at the output of the distribution unit, the reference phase of each phase voltage relative to the grid phase is adjusted in steps of 0.3 cm within a ±4.5 degree range. The adjustment goal is to minimize the variance of the imaginary part of the input impedance of each power source under that phase. Simultaneously, when the neutral current exceeds 15% of the phase current and the third harmonic content exceeds 12% of the fundamental frequency, a signal generator within the distribution unit superimposes an amplitude of 2.8 times the phase voltage. The third harmonic component with a phase relative to the fundamental wave of 180 degrees is observed after 300 milliseconds. If the decrease is greater than 18%, the injection parameters are retained as steady-state compensation. If the decrease is less than 8%, the injection is stopped and the step size is changed to 0.7 minutes for retesting. The total adjustment of phase shift and harmonic injection is converted into the equivalent current reduction value. The equivalent current reduction value is equal to the sum of the phase shift adjustment value and the harmonic injection adjustment value multiplied by 0.6, and the equivalent current reduction value in S20 is updated. S403, when in a hard limit zone, performs emergency duty cycle clamping based on S402: temporarily reduces the upper limit of the pulse width modulation duty cycle of all servers under this phase from the rated value to 70% of the rated value, and at the same time switches the output relay of the power distribution unit of this phase to the backup circuit. The switching sequence is to first reduce the duty cycle to 30%, then disconnect the main circuit relay, delay for 20 milliseconds, close the backup circuit relay, and then restore the duty cycle with a ramp of 0.3% per millisecond. The whole process does not exceed 800 milliseconds.

[0042] The principle and effect of the above technical solution are as follows: the electrical signal level modulation forms an intervention gradient from shallow to deep by going from mild kneading to phase shift harmonics and then to emergency clamping. Each modulation generates an equivalent current reduction value that is fed back into the threshold calculation, realizing a positive feedback closed loop of "modulation changes the current, the current affects the threshold, and the threshold guides the modulation again". This is a dynamic self-balancing capability that traditional static threshold solutions do not have.

[0043] In one embodiment, S50 includes: S501, when any phase of this rack enters the intervention zone, the rack controller sends a migration request to the co-row controller. The request includes the adjustability margin of that phase in this rack, as well as the power estimate of each virtual machine in this rack. S502, the same-row controller collects all 72 adjustable margin values ​​of the 12 racks in its row at a 1-second interval. These adjustable margin values ​​are sorted from smallest to largest. The set of racks with negative adjustable margins is defined as the scarce side, and the set of racks with adjustable margins greater than 28% of the rack's rated current is defined as the abundant side. The abundant side is then queued for migration in descending order of adjustable margin value. Migration decisions follow the principle of prioritizing racks within the same row and providing a fallback across rows: when the sum of the absolute values ​​of negative adjustable margin values ​​of a rack on the scarce side is less than 70% of the total available adjustable margin in the same row on the abundant side, migration is completed only within that row; if this ratio is exceeded, the row controller sends a cross-row request to the room-level central controller. S503, the room-level central controller performs secondary convergence across all 8 columns in the room at a 3-second interval. During secondary convergence, the impedance of the upstream phase line connectors of each column is additionally considered. For the column to be transferred, the temporary connector impedance reduction factor of all distribution unit branches in the column is first calculated based on the impedance of each phase connector. Then, the temporary connector impedance reduction factor is used to replace the original connector impedance reduction factor to recalculate the temporary comprehensive load safety threshold and temporary adjustable margin of each phase in the column. The column is only allowed to transfer loads if the temporary adjustable margin is greater than 1.3 times the current to be transferred. The migration process employs a micro-step sequence of first decreasing, then switching, and then increasing: For the server to be migrated, the first power limiter is issued, reducing the power by 40W every 200 milliseconds. When the power limit drops to 30% of the rated power, the first output relay is disconnected. After a 20-millisecond delay, the server's second power supply performs a power factor correction soft start, increasing the duty cycle by 0.3% every millisecond, with the entire process not exceeding 800 milliseconds. During the soft start, the central control room temporarily increases the second power supply's overall load safety threshold by 15% as a switching tolerance window. After the switching is completed, the tolerance window is removed 3 seconds later.

[0044] The principle and effect of the above technical solution are as follows: hierarchical migration not only considers the load margin, but also introduces the dynamic correction of the receiving side threshold by the connector impedance reduction, preventing the receiving side from exceeding its limits due to the migration load; the micro-step switching sequence avoids the current surge and instantaneous overload of redundant paths caused by traditional hard switching, and the tolerance window mechanism further ensures the safety of switching.

[0045] In one embodiment, S60 includes: S601: When the current in any phase of the rack rises at a rate greater than 7% of the rated current of that phase within 400 milliseconds, it is determined to be a short-term impact. The central controller at the room level will not schedule the migration, but will immediately command the supercapacitor at the top of the column to discharge at a rate of 1.2 times the current impact current, with a maximum discharge time of 10 seconds. During the discharge, the comprehensive load safety threshold of that phase is temporarily amplified by 8% as a buffer tolerance to avoid the buffer action and the threshold warning conflicting in the cycle time. S602, if the inrush current slope is still greater than the rated 4% every 400 milliseconds after 10 seconds, the lithium battery will continue to discharge, and the lithium battery discharge current will be 80% of the current when the supercapacitor terminates. At the same time, cross-row migration will be initiated. If the inrush slope drops to below the rated 2% every 400 milliseconds within 10 seconds, the supercapacitor will stop discharging, the buffer tolerance will be cancelled, and the reinforcement learning module in the room-level center will temporarily reduce the capacitor aging reduction factor of this phase of the rack by 0.03, marking that the rack may have business mode drift in the near future. S603: If the lithium battery discharge duration exceeds 5 minutes or the lithium battery state of charge drops below 20%, all non-critical loads in that column will be forcibly disconnected, and an operation and maintenance alarm will be issued. The internal resistance of the buffer unit is transmitted back to the central control room every 30 seconds. When the internal resistance increases by more than 35% compared to a new product, the maximum discharge rate of the buffer unit in that column will be reduced from 3.5 times to 2.2 times. At the same time, the central control room will add a coefficient of 0.96 to the impedance reduction factor of the connectors of all power distribution unit branches in that column.

[0046] The principle and effect of the above technical solution are as follows: the buffer unit is linked with the threshold, and the comprehensive load safety threshold is temporarily amplified to avoid mutual interference between buffer discharge and threshold warning. At the same time, the aging information of the buffer device is fed back into the threshold calculation, so that the impact of aging on the safety boundary is captured in time, forming a closed loop of the entire life cycle of "buffer discharge, threshold tolerance, device aging, and threshold correction".

[0047] In one specific embodiment, in S40, if the current phase is in the warning zone, a light kneading modulation is adopted, and the specific steps are as follows: SA41: Read the current current of each of the 12 outputs connected to this phase and the total current of this phase, and calculate the load rate of each output by dividing the current current of the output by the rated current of the output. SA42: For each dual-power server, calculate its load ratio by dividing the current of the first power supply by the current of the second power supply. If the load ratio is greater than 1.2, send a pulse width modulation reference fine-tuning command to the server: decrease the duty cycle of the first power supply by 0.5% per step, and increase the duty cycle of the second power supply by the same amount, with each step interval being 400 milliseconds; if the load ratio is less than 0.8, perform the reverse operation: increase the duty cycle of the first power supply by 0.5% per step, and decrease the duty cycle of the second power supply; if the load ratio is between 0.8 and 1.2, no adjustment is made. SA43: After each adjustment, wait 400 milliseconds, re-collect the current of each circuit and update the load ratio. Repeat SA42 until the total current of the phase drops below 0.82 times the comprehensive load safety threshold or the total current decrease is less than 1% after 5 consecutive adjustments. SA44: Record the equivalent current reduction value generated by this kneading as equal to the total current before kneading minus the total current after kneading, and add it to the equivalent current reduction value in S20.

[0048] The mild kneading modulation reduces the total single-phase current without interrupting services by finely balancing the load distribution of the dual power supplies, and the adjustment range is controlled so as not to cause power supply loop oscillation.

[0049] In S40, if the current phase is in the interference region, then phase shift and harmonic injection are used for joint modulation. The specific steps are as follows: SB41: Read the phase difference of the current and the neutral current of the 12 outputs under this phase, and calculate the phase variance, which is equal to 1 / 12 multiplied by the sum of the squares of the phase difference of each channel and the average phase difference; SB42: The reference phase of the phase voltage is gradually adjusted in steps of 0.3 cm using a bidirectional thyristor within the power distribution unit. After each adjustment, the phase variance is recalculated after 200 milliseconds. If the phase variance decreases, the adjustment continues in the same direction; if it increases, the adjustment reverses, until the phase variance is less than the preset threshold of 0.02 radians squared or the number of adjustments reaches 15. The adjustment range is limited to ±4.5 degrees. SB43: After adjusting the phase, check if the neutral current is greater than 0.15 times the phase current and if the third harmonic content is greater than 0.12 times the phase current. If both conditions are met, inject a third harmonic voltage with an amplitude of 0.028 times the phase voltage and a phase 180 degrees relative to the fundamental frequency into the phase through the signal generator in the distribution unit, and continue for 300 milliseconds. After 300 milliseconds, check the neutral current again. If the decrease in neutral current is greater than 18%, retain the injected parameters as steady-state compensation. If the decrease is less than 8%, stop the injection and change the step size to 0.7 minutes to retest. If the decrease is still less than 8% after the second test, abandon the harmonic injection. SB44: Convert the phase shift adjustment amount into an equivalent current reduction value equal to the absolute value of the phase shift angle divided by 4.5 degrees, multiplied by 0.03, and multiplied by the phase current. Convert the harmonic injection adjustment amount into an equivalent current reduction value equal to the neutral line current before harmonic injection minus the neutral line current after harmonic injection multiplied by 0.5. The total equivalent current reduction value equals the phase shift equivalent current reduction value plus the harmonic equivalent current reduction value, and update the equivalent current reduction value in S20.

[0050] The combined phase shift and harmonic injection modulation reduces line loss and neutral current from two dimensions: impedance matching and harmonic suppression. This allows the total current to decrease naturally without changing the load power, and does not trigger the power factor correction protection of the server power supply.

[0051] In S40, if the current phase is in the hard limit region, emergency duty cycle clamping and redundancy switching are adopted. The specific steps are as follows: SC41: Immediately send a pulse width modulation duty cycle upper limit clamping command to all servers under this phase, temporarily reducing the duty cycle upper limit from 100% to 70%, with an execution time not exceeding 50 milliseconds; SC42: Simultaneously detect the current current of the other circuit of the distribution unit to which this phase belongs and the real-time comprehensive load safety threshold of the backup circuit. If the real-time comprehensive load safety threshold of the backup circuit minus the current current of the backup circuit is greater than 1.1 times the current current of the main circuit, then perform the handover; otherwise, skip the handover, only maintain the duty cycle clamp and trigger the forced migration of S50. SC43: During the switchover execution, first, a current limiting point is sent to the main server to reduce the current by 40W every 200 milliseconds. When the current drops to 30% of the rated current, the first output relay is disconnected. After a 20-millisecond delay after disconnection, the backup relay is closed. At the same time, a power factor correction soft-start command is sent to the server's second power supply. The soft-start ramp increases the duty cycle by 0.3% every millisecond until the target value is reached. The entire process takes no more than 800 milliseconds. SC44: After the handover is completed, the backup path temporarily obtains a handover tolerance window of 3 seconds, that is, its comprehensive load safety threshold is increased by 15% within 3 seconds after the handover is completed, and then returns to the original value after 3 seconds; at the same time, this handover event is recorded in the room-level central database, and the threshold of all phases of the rack is tightened for 1 minute, with the tightening amount being all reduction coefficients multiplied by 0.98.

[0052] The emergency duty cycle clamping and redundancy switching reduce current and transfer load as quickly as possible in the hard limit area, while preventing secondary limit overruns caused by switching through tolerance windows and post-tightening mechanisms.

[0053] In one specific embodiment, in S50, if the sum of the absolute values ​​of the negative values ​​of the adjustable margin of a rack on the shortage side is less than 70% of the total adjustable margin available in the same column on the surplus side, then intra-column migration is adopted, and the specific steps are as follows: SA51: The column controller sorts the list of racks on the scarce side from largest to smallest by the absolute value of the negative value of the adjustable margin, and sorts the list of racks on the abundant side from largest to smallest by the adjustable margin. SA52: Starting from the first rack on the scarce side, select the phase with the largest absolute value of its negative adjustable margin. From the first rack on the abundant side, select the phase with the largest transferable adjustable margin. Calculate the transferable current equal to the smaller of the absolute value of the adjustable margin on the scarce side and the adjustable margin on the abundant side multiplied by 0.8. SA53: Convert the transferable current to the number of virtual machines and round down (transferable current divided by 0.87A). Select the corresponding number of non-real-time virtual machines from the scarce racks. Based on the power estimate of 200W per virtual machine, send migration commands to the abundant racks one by one. The migration rate is limited to no more than 4 virtual machines per second. SA54: After migrating each virtual machine, recalculate the adjustable margins of the corresponding phases on the scarce side and the abundant side. If the adjustable margin on the scarce side has returned to a positive value or the adjustable margin on the abundant side has dropped below 15% of the rated value, then stop the migration and record the total amount of this migration.

[0054] The intra-column migration quickly digests local overload within the same column, avoids cross-column communication delays, and the migration rate is controlled to prevent instantaneous overload on the receiving side.

[0055] In S50, if the sum of the absolute values ​​of the negative values ​​of the adjustable margin of a rack on the shortage side is greater than or equal to 70% of the total adjustable margin that can be transferred in this column on the surplus side, then cross-column migration is adopted, and the specific steps are as follows: SB51: The column controller sends a cross-column request to the center of the room, along with the rack identifier on the scarce side, the phase, the total current to be transferred, and the current connector impedance value of each phase in that column; SB52: The central section of the room traverses the remaining 7 columns of the room. For each column and each phase, the temporary joint impedance reduction factor is calculated using the joint impedance of that column. This factor is equal to 1 minus 0.35 multiplied by the difference between the joint impedance minus 0.010 and the cutoff value of 0.018. Then, the temporary joint impedance reduction factor is used to replace the original joint impedance reduction factor to recalculate the temporary comprehensive load safety threshold and temporary adjustable margin of each phase in that column. Columns with a temporary adjustable margin greater than 0.2 times the rated current are selected as candidate receiving columns and sorted from largest to smallest according to the temporary adjustable margin. SB53: Starting from the first candidate in the receiving column, calculate the amount of current that can be accepted, which is equal to the smaller value between the temporary adjustable margin multiplied by 0.7 and the total current to be migrated. Then, calculate the number of virtual machines in the same way as SA53 and initiate the migration. The migration rate is limited to no more than 2 virtual machines per second. SB54: After each batch of virtual machines is migrated, the temporary adjustable margin of the receiving column is recalculated. If the temporary adjustable margin drops below 0.15 times the rated current, the receiving of that column is suspended and the process moves to the next candidate column. This continues until the sum of the absolute values ​​of the negative values ​​of the adjustable margins of all racks on the scarce side drops to 0 or all candidate columns are used up.

[0056] The cross-column migration dynamically corrects the receiving-side threshold by introducing connector impedance reduction, ensuring that the migration load does not cause the receiving column itself to exceed the limit due to connector aging, while the migration rate is lower to avoid cross-column communication congestion.

[0057] In S50, if a backup path switching tolerance window is encountered during the migration process, migration within the tolerance window is used to accelerate the process. The specific steps are as follows: SC51: When a rack is performing a redundancy switch in S40 and is within a 3-second tolerance window, the central control room marks the rack as being in a tolerance state and temporarily increases the combined load safety threshold of all its phases by 15%. SC52: In the forgiving state, this rack can serve as the margin side to take on the migration load of other racks, but the amount taken on must not exceed 50% of its temporary adjustable margin (calculated based on the amplified comprehensive load safety threshold), and the virtual machines taken on must be low-priority non-real-time services. SC53: After the tolerance period ends, the central room automatically cancels the tolerance side qualification of the rack and performs a 30-second threshold tightening on all its phases. The tightening amount is all reduction factors multiplied by 0.97 to prevent the additional load accumulated during the tolerance period from causing subsequent over-limits.

[0058] The migration acceleration within the tolerance window fully utilizes the temporary margin provided by redundant switching, improving overall resource utilization, while preventing abuse through post-event tightening.

[0059] In one specific embodiment, in S60, if the impact slope is greater than 7% of the rated value every 400 milliseconds and lasts for more than 10 seconds, then the lithium battery discharge and cross-column migration are linked, and the specific steps are as follows: SA61: When the supercapacitor discharges for 10 seconds, if the inrush current slope is still greater than the rated 4% every 400 milliseconds, the central control unit immediately commands the lithium battery to start discharging at 80% of the current at the time the supercapacitor terminates. The upper limit of the discharge current is the rated maximum continuous current of the lithium battery. SA62: At the same time, the central room will list the racks affected by the impact as the highest priority shortage side, and start the migration according to the cross-column migration process of S50. The migration target will be selected first, and the column far away from the impact source will be selected. The migration rate will be increased to no more than 6 virtual machines per second. SA63: During lithium battery discharge, the surge current slope is detected every 5 seconds. If the slope drops below 2% of the rated value every 400 milliseconds, the lithium battery discharge is stopped and the rack is marked as a surge recovery state. In the following 30 seconds, the capacitor aging reduction factor of all phases of the rack is temporarily reduced by 0.03 to reflect possible business mode drift. SA64: If the lithium battery discharge lasts for more than 5 minutes or the state of charge drops below 20%, all non-critical loads in the column will be forcibly disconnected and an operation and maintenance alarm will be issued. The disconnection order is to disconnect the test server first, then the cold standby node, and finally the low-priority virtual machine. After each disconnection, wait 1 second to observe whether the current drops to a safe level.

[0060] The lithium battery discharge and cross-column migration linkage provides continuous power support after the supercapacitor has had a brief buffering period. At the same time, the migration completely eliminates the root cause of overload, forming a complete emergency chain of "buffering, migration, and recovery".

[0061] In S60, if the impact slope drops below 2% of the rated value every 400 milliseconds within 10 seconds, then a supercapacitor is used for separate processing and threshold fine-tuning. The specific steps are as follows: SB61: After the supercapacitor stops discharging, the central rack in the room will be marked as impacted and in the impact recovery state, and a threshold fine-tuning will be performed on all its phases: the capacitor aging reduction factor will be temporarily reduced by 0.03, and will automatically recover after 30 minutes; SB62: At the same time, the reinforcement learning module of the central room level stores the event as a short-term impact sample in the experience pool. The state characteristics include the current current, terminal temperature, air inlet temperature, current equivalent series resistance, connector impedance, and maximum voltage distortion rate before the impact. The action characteristics are the supercapacitor discharge duration and discharge current. The reward is whether a hard limit zone appears after the impact. SB63: The reinforcement learning module performs a 5-step small update every 15 minutes using the experience pool data from the most recent 4 hours. The updated incremental adjustments to the terminal temperature reduction coefficient, air inlet temperature reduction coefficient, capacitor aging reduction coefficient, connector impedance reduction coefficient, and voltage distortion reduction coefficient are then superimposed on the corresponding reduction coefficients of S20 after being verified by the central verification at the room level.

[0062] The supercapacitor-based individual processing and threshold fine-tuning, after the impact disappears, use reinforcement learning to make minor corrections to the threshold, allowing the system to gradually adapt to the service characteristics of the rack and avoid repeated occurrences of similar impacts. At the same time, the update range of reinforcement learning is strictly limited and does not directly participate in modulation or migration decisions, ensuring security.

[0063] In S60, if the internal resistance of the buffer unit increases by more than 35% compared to a new product, buffer degradation and threshold compensation are adopted. The specific steps are as follows: SC61: When the internal resistance of a supercapacitor or lithium battery in a certain column is detected to exceed 1.35 times the nominal value of a new product in the central part of the room, the maximum discharge rate of the buffer unit in that column will be reduced from 3.5 times to 2.2 times, and the degradation time will be recorded. SC62: At the same time, the central control room will add a compensation factor of 0.96 to the impedance reduction factor of all the branch circuits of the distribution unit in this column. That is, the new impedance reduction factor is equal to the original impedance reduction factor multiplied by 0.96, so that the overall load safety threshold of all phases in this column will decrease by about 4% to compensate for the decrease in emergency capability after the buffer unit ages. SC63: The central alarm for room-level operation and maintenance is generated, notifying maintenance personnel to replace the buffer unit in this column within 48 hours. After replacement, the new buffer unit needs to undergo a 24-hour full-load aging test before going online. After the test is passed, the compensation coefficient is restored to 1, and the connector impedance reduction coefficient is restored to normal calculation.

[0064] The buffer degradation and threshold compensation directly map the buffer device aging information to the load safety threshold, so that the system automatically tightens the power supply margin when the buffer capacity deteriorates, avoiding over-limit accidents caused by buffer failure, and triggering the operation and maintenance process to form a closed loop of full life cycle management of "detection, degradation, alarm, replacement and recovery".

[0065] Accordingly, such as Figure 2As shown, based on a method for dynamic control of data center power supply based on a load safety threshold, this embodiment of the invention also provides a system for dynamic control of data center power supply based on a load safety threshold, which implements the method for dynamic control of data center power supply based on a load safety threshold according to this embodiment of the invention. The system includes: The sensor array and data acquisition module are used to place current transformers, voltage transformers, negative temperature coefficient temperature sensors and ambient temperature probes at the output terminals, wiring terminals and air inlets on the top of each rack in the computer room of the dual-path power distribution unit. A parallel acquisition link including a high-frequency detection pulse injection and response acquisition unit is built after the DC-DC converter of each server power module to synchronously acquire the effective value of each phase current, effective value of phase voltage, effective value of neutral line current, temperature of copper busbar at wiring terminal, air inlet temperature of rack, equivalent series resistance of electrolytic capacitor, impedance of phase line connector and voltage distortion rate of column head. The signal conditioning and online calibration module is used to dynamically modulate, decouple, and compensate the acquired raw mixed electrical signals. Specifically, it uses a digital quadrature phase-locked loop demodulation algorithm to extract the response characteristics of high-frequency detection pulses in real time to quantify transmission link distortion, and applies a wavelet transform-based noise reduction algorithm for secondary filtering, thereby generating high-fidelity physical quantities such as terminal temperature, air inlet temperature, equivalent series resistance, connector impedance, and voltage distortion rate. The real-time load safety threshold calculation module is used to calculate the real-time threshold of each phase according to the high-fidelity physical quantity, the comprehensive load safety threshold formula and its sub-formulas, and to calculate the adjustable margin.

[0066] The electrical signal level dynamic modulation module is used to perform mild kneading modulation, phase shift and harmonic injection combined modulation, or emergency duty cycle clamping and redundancy switching according to the warning zone, intervention zone, or hard limit zone where the current phase current is located, and output the equivalent current reduction value to the threshold calculation module. The hierarchical logic migration module is used to recalculate the temporary threshold according to the principle of priority in the same column and fall back across columns when any phase of the rack enters the interference zone or hard limit zone, and to perform virtual machine migration, and accelerate the migration within the switching tolerance window. The transient buffer and threshold linkage module is used to control the supercapacitor and lithium battery at the top of the column to discharge in stages when the current of any phase of the rack rises at a rate exceeding 7% of the rated current within 400 milliseconds. During the discharge, the overall load safety threshold is temporarily amplified, and the connector impedance reduction coefficient is adjusted according to the aging of the internal resistance of the buffer unit. The reinforcement learning threshold fine-tuning module is used to fine-tune five reduction coefficients within a strictly defined action range using short-term impact events as samples, and then superimposed into the threshold calculation after being centrally verified at the room level. The full lifecycle calibration and management module is used to compare the drift of each reduction factor with the factory value in a 30-day window, trigger maintenance work orders, and perform degradation and threshold compensation when the internal resistance of the buffer unit rises by more than 35%.

[0067] It should be noted that the data center power dynamic control system based on load safety threshold provided in this embodiment of the invention is used to execute all the process steps of the data center power dynamic control method based on load safety threshold in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0068] This invention also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a dynamic power control program for a data center based on a load safety threshold. When the processor executes the computer program, it implements the steps in each of the above embodiments of a dynamic power control method for a data center based on a load safety threshold, such as steps S10 to S60 shown in document A. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments.

[0069] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0070] The electronic device may be a data acquisition and processing industrial control computer, edge server, or cloud server deployed in a data center. The electronic device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include data acquisition cards, network communication devices, buses, etc., for connecting various sensors.

[0071] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the entire electronic device via various interfaces and lines.

[0072] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to system usage, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart memory card, secure digital card, flash memory card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0073] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0074] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for dynamic control of data center power supply based on load safety threshold, characterized in that, include: S10 collects electrical parameters, temperature parameters, capacitor aging parameters, connector impedance parameters, and voltage distortion parameters for each phase of the power distribution unit. S20, based on the collected data, calculate the comprehensive load safety threshold and adjustability margin for each phase; S30, compare the current phase current with the comprehensive load safety threshold and divide it into normal zone, early warning zone, intervention zone and hard limit zone; S40 performs different granularity of electrical signal-level dynamic modulation in the warning zone, intervention zone and hard limit zone, including phase shift fine adjustment, third harmonic injection and pulse width modulation duty cycle kneading, to reduce the current phase current or delay the current rise rate. S50: When any phase of this rack enters the intervention area or hard limit area, a hierarchical logic migration is initiated to migrate the non-real-time service virtual machine of this rack to a rack with ample space in the same row or a rack with ample space across rows. The migration rate is limited by the available margin on the receiving side and the impedance reduction of the connector. S60: When the current rise slope of any phase of the rack exceeds the preset slope threshold, the column top transient buffer unit is triggered to discharge in stages. If the discharge fails to effectively suppress the slope, cross-column migration is initiated.

2. The method as described in claim 1, characterized in that, In S10, the effective values ​​of phase current, phase voltage, and neutral current of each phase distribution unit are collected at a sampling interval of 100 milliseconds. The temperature of the copper busbar at the terminal of the distribution unit is collected at a 1-second interval. The air intake temperature of the rack is collected at a 10-second interval. The equivalent series resistance of the electrolytic capacitor of each server power module is measured by a high-frequency detection pulse every 10 minutes. The impedance of each phase line connector is calculated by reverse calculation every 15 minutes. The column head voltage distortion rate is calculated by sampling 128 points per cycle and the maximum value of the 10-minute sliding window is taken.

3. The method as described in claim 1, characterized in that, In S20, the comprehensive load safety threshold is equal to the rated current carrying capacity multiplied by the terminal temperature reduction factor, the inlet air temperature reduction factor, the capacitor aging reduction factor, the connector impedance reduction factor, the voltage distortion reduction factor, and the circuit breaker matching margin; the adjustable margin is equal to the comprehensive load safety threshold minus the equivalent current reduction value and then minus the current.

4. The method as described in claim 1, characterized in that, In S30, the current current is in the normal zone when it is less than or equal to 0.85 times the comprehensive load safety threshold; the current current is in the warning zone when it is greater than 0.85 times but less than or equal to 0.93 times the comprehensive load safety threshold; the current current is in the intervention zone when it is greater than 0.93 times but less than or equal to 0.97 times the comprehensive load safety threshold; and the current current is in the hard limit zone when it is greater than 0.97 times the comprehensive load safety threshold.

5. The method as described in claim 1, characterized in that, The high-rate pulse command in S40 is a pulse command with an amplitude greater than 0.8 times the rated power and a duration greater than 10 seconds. Three strategies are adopted according to different ranges of polarization residual absolute value and recovery time interval: if the polarization residual absolute value is less than 0.05 volts and the recovery time interval is greater than or equal to 5 times the polarization time constant, it is issued in full; if the polarization residual absolute value is greater than or equal to 0.05 volts and less than 0.15 volts and the recovery time interval is greater than or equal to 5 times the polarization time constant, it is issued after being reduced by a reduction factor; if the polarization residual absolute value is greater than or equal to 0.15 volts or the recovery time interval is less than 5 times the polarization time constant, it is prohibited or further reduced.

6. The method as described in claim 1, characterized in that, In S50, non-real-time business virtual machines in the local rack will be migrated to the same row of spacious racks or across rows of spacious racks at a rate of 2 million per virtual machine. The preset slope threshold in S60 is that the rising slope exceeds 7% of the rated current of the phase within every 400 milliseconds. The graded discharge includes first discharging by the supercapacitor at a rate of 1.2 times the impulse current for a maximum of 10 seconds. If the slope still does not drop below 2% of the rated current within every 400 milliseconds after 10 seconds, it will switch to lithium battery discharge and start cross-column migration at the same time.

7. The method as described in claim 1, characterized in that, The S30 also includes: S301 compares the current phase current with the comprehensive load safety threshold. If the current current is less than or equal to 0.85 times the comprehensive load safety threshold, it is determined to be in the normal zone. No modulation or migration action is performed, and only routine monitoring is maintained. S302, if the current current is greater than 0.85 times and less than or equal to 0.93 times the comprehensive load safety threshold, it is determined to be in the warning zone and enters the electrical signal mild modulation stage of S40; S303, if the current current is greater than 0.93 times and less than or equal to 0.97 times the comprehensive load safety threshold, it is determined to be an intervention zone, and enters the heavy modulation stage of electrical signal in S40, while triggering the hierarchical migration preparation in S50. S304 If the current current is greater than 0.97 times the comprehensive load safety threshold, it is determined to be a hard limit zone, and the emergency modulation of S40 and the forced migration of S50 are immediately started, and the buffer unit of S60 is activated.

8. The method as described in claim 1, characterized in that, The S40 also includes: S401, when in the warning zone, performs light modulation: the outputs connected to the current phase are re-proportionally redistributed according to the current load rate of each output. Specifically, the load ratio of the two power modules of each server is equal to the current of the first output divided by the current of the second output. For servers with a load ratio greater than 1.2, a pulse width modulation reference fine-tuning command is issued, causing the duty cycle of the first output to decrease by 0.5% per step and the duty cycle of the second output to increase accordingly. For servers with a load ratio less than 0.8, a symmetrical reverse squeezing is performed. The squeezing amplitude of a single squeeze does not exceed 3% of the rated current of the output, and the interval between two squeezings is not less than 400 milliseconds. S402, when in the intervention zone, adds phase shift fine-tuning and harmonic injection to S401: Through the bidirectional thyristor at the output of the distribution unit, the reference phase of each phase voltage relative to the grid phase is adjusted in steps of 0.3 cm within a ±4.5 degree range. The adjustment goal is to minimize the variance of the imaginary part of the input impedance of each power source under that phase. Simultaneously, when the neutral current exceeds 15% of the phase current and the third harmonic content exceeds 12% of the fundamental frequency, a signal generator within the distribution unit superimposes an amplitude of 2.8 times the phase voltage. The third harmonic component with a phase relative to the fundamental wave of 180 degrees is observed after 300 milliseconds. If the decrease is greater than 18%, the injection parameters are retained as steady-state compensation. If the decrease is less than 8%, the injection is stopped and the step size is changed to 0.7 minutes for retesting. The total adjustment of phase shift and harmonic injection is converted into the equivalent current reduction value. The equivalent current reduction value is equal to the sum of the phase shift adjustment value and the harmonic injection adjustment value multiplied by 0.6, and the equivalent current reduction value in S20 is updated. S403, when in a hard limit zone, performs emergency duty cycle clamping based on S402: temporarily reduces the upper limit of the pulse width modulation duty cycle of all servers under this phase from the rated value to 70% of the rated value, and at the same time switches the output relay of the distribution unit of this phase to the backup circuit. The switching sequence is to first reduce the duty cycle to 30%, then disconnect the main circuit relay, and after a delay of 20 milliseconds, close the backup circuit relay to restore the duty cycle with a ramp of 0.3% per millisecond.

9. The method as described in claim 1, characterized in that, The S50 also includes: S501, when any phase of this rack enters the intervention zone, the rack controller sends a migration request to the co-row controller. The request includes the adjustability margin of that phase in this rack, as well as the power estimate of each virtual machine in this rack. S502, the same-row controller collects all 72 adjustable margin values ​​of the 12 racks in its row at a 1-second interval. These adjustable margin values ​​are sorted from smallest to largest. The set of racks with negative adjustable margins is defined as the scarce side, and the set of racks with adjustable margins greater than 28% of the rack's rated current is defined as the abundant side. The abundant side is then queued for migration in descending order of adjustable margin value. Migration decisions follow the principle of prioritizing racks within the same row and providing a fallback across rows: when the sum of the absolute values ​​of negative adjustable margin values ​​of a rack on the scarce side is less than 70% of the total available adjustable margin in the same row on the abundant side, migration is completed only within that row; if this ratio is exceeded, the row controller sends a cross-row request to the room-level central controller. S503, the room-level central controller performs secondary convergence across all 8 columns in the room at a 3-second interval. During secondary convergence, the impedance of the upstream phase line connectors of each column is additionally considered. For the column to be transferred, the temporary connector impedance reduction factor of all distribution unit branches in the column is first calculated based on the impedance of each phase connector. Then, the temporary connector impedance reduction factor is used to replace the original connector impedance reduction factor to recalculate the temporary comprehensive load safety threshold and temporary adjustable margin of each phase in the column. The column is only allowed to transfer loads if the temporary adjustable margin is greater than 1.3 times the current to be transferred. The migration process employs a micro-step sequence of first decreasing, then switching, and then increasing: For the server to be migrated, the first power limit is reduced by 40W every 200 milliseconds. When the power limit drops to 30% of the rated power, the first output relay is disconnected. After a 20-millisecond delay, the server's second power supply performs a power factor correction soft start, with the soft start ramp increasing by 0.3% duty cycle every millisecond, for a total duration not exceeding 800 milliseconds. During the soft start, the central control room temporarily increases the second power supply's overall load safety threshold by 15% as a switching tolerance window. After the switching is complete, the tolerance window is removed 3 seconds later. The S60 also includes: S601: When the current in any phase of the rack rises at a rate greater than 7% of the rated current of that phase within 400 milliseconds, it is determined to be a short-term impact. The central controller at the room level will not schedule the migration, but will immediately command the supercapacitor at the top of the column to discharge at a rate of 1.2 times the current impact current, with a maximum discharge time of 10 seconds. During the discharge, the comprehensive load safety threshold of that phase is temporarily amplified by 8% as a buffer tolerance to avoid the buffer action and the threshold warning conflicting in the cycle time. S602, if the inrush current slope is still greater than the rated 4% every 400 milliseconds after 10 seconds, the lithium battery will continue to discharge, and the lithium battery discharge current will be 80% of the current when the supercapacitor terminates. At the same time, cross-row migration will be initiated. If the inrush slope drops to below the rated 2% every 400 milliseconds within 10 seconds, the supercapacitor will stop discharging, the buffer tolerance will be cancelled, and the reinforcement learning module in the room-level center will temporarily reduce the capacitor aging reduction factor of this phase of the rack by 0.03, marking that the rack may have business mode drift in the near future. S603: If the lithium battery discharge duration exceeds 5 minutes or the lithium battery state of charge drops below 20%, all non-critical loads in that column will be forcibly disconnected, and an operation and maintenance alarm will be issued. The internal resistance of the buffer unit is transmitted back to the central control room every 30 seconds. When the internal resistance increases by more than 35% compared to a new product, the maximum discharge rate of the buffer unit in that column will be reduced from 3.5 times to 2.2 times. At the same time, the central control room will add a coefficient of 0.96 to the impedance reduction factor of the connectors of all power distribution unit branches in that column.

10. A dynamic power control system for a computer room based on a load safety threshold, used to implement the method described in any one of claims 1-9, characterized in that, The system includes: The sensor array and data acquisition module are used to place current transformers, voltage transformers, negative temperature coefficient temperature sensors and ambient temperature probes at the output terminals, wiring terminals and air inlets on the top of each rack in the computer room of the dual-path power distribution unit. A parallel acquisition link including a high-frequency detection pulse injection and response acquisition unit is built after the DC-DC converter of each server power module to synchronously acquire the effective value of each phase current, effective value of phase voltage, effective value of neutral line current, temperature of copper busbar at wiring terminal, air inlet temperature of rack, equivalent series resistance of electrolytic capacitor, impedance of phase line connector and voltage distortion rate of column head. The signal conditioning and online calibration module is used to dynamically modulate, decouple, and compensate the acquired raw mixed electrical signals. Specifically, it uses a digital quadrature phase-locked loop demodulation algorithm to extract the response characteristics of high-frequency detection pulses in real time to quantify transmission link distortion, and applies a wavelet transform-based noise reduction algorithm for secondary filtering, thereby generating high-fidelity physical quantities such as terminal temperature, air inlet temperature, equivalent series resistance, connector impedance, and voltage distortion rate. The real-time load safety threshold calculation module is used to calculate the real-time threshold of each phase according to the high-fidelity physical quantity, the comprehensive load safety threshold formula and its sub-formulas, and to calculate the adjustable margin. The electrical signal level dynamic modulation module is used to perform mild kneading modulation, phase shift and harmonic injection combined modulation, or emergency duty cycle clamping and redundancy switching according to the warning zone, intervention zone, or hard limit zone where the current phase current is located, and output the equivalent current reduction value to the threshold calculation module. The hierarchical logic migration module is used to recalculate the temporary threshold according to the principle of priority in the same column and fall back across columns when any phase of the rack enters the interference zone or hard limit zone, and to perform virtual machine migration, and accelerate the migration within the switching tolerance window. The transient buffer and threshold linkage module is used to control the supercapacitor and lithium battery at the top of the column to discharge in stages when the current of any phase of the rack rises at a rate exceeding 7% of the rated current within 400 milliseconds. During the discharge, the overall load safety threshold is temporarily amplified, and the connector impedance reduction coefficient is adjusted according to the aging of the internal resistance of the buffer unit. The reinforcement learning threshold fine-tuning module is used to fine-tune five reduction coefficients within a strictly defined action range using short-term impact events as samples, and then superimposed into the threshold calculation after being centrally verified at the room level. The full lifecycle calibration and management module is used to compare the drift of each reduction factor with the factory value in a 30-day window, trigger maintenance work orders, and perform degradation and threshold compensation when the internal resistance of the buffer unit rises by more than 35%.

Images