A data center dual-power supply security management system and method

Abstract

This invention relates to the field of data center power supply security management technology, specifically to a data center dual-power supply security management system and method, comprising: a phase-splitting multi-point parameter acquisition module, connected to the incoming, outgoing, and busbar connection terminals of the dual-power supply main and backup circuits respectively, for synchronously acquiring phase and line impedance. This invention synchronously acquires real-time monitoring data of the electrical parameters of the dual-power supply main and backup circuits through the phase-splitting multi-point parameter acquisition module. Combined with load trend prediction information extracted from the computing power scheduling platform, a dual-parameter dynamic coupling analysis module establishes a correlation mapping between parameters and load and accurately determines the imbalance level. Based on this, the intelligent central control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module to dynamically fine-tune the phase difference and line impedance imbalance caused by dynamic load fluctuations, and synchronously suppresses busbar harmonics induced by parameter fine-tuning at the root.

Description

Technical Field

[0001] This invention relates to the field of data center power supply security management technology, specifically to a data center dual-power supply security management system and method. Background Technology

[0002] Dual power supply is a core configuration for ensuring the continuous and stable operation of high-density computing equipment in data centers. In this configuration, the two power supply circuits do not operate independently. Their core parameters, such as power supply phase and line impedance, need to be compatible with the power supply characteristics of the data center busbar. At the same time, the coordination between the two parameters will change with the dynamic changes of the end computing load. The core of power supply management is not only to realize abnormal switching of the circuit, but also to ensure the parameter coordination of the dual power supply at all times, and avoid power quality problems caused by parameter imbalance. This type of management requirement is particularly prominent in high-density computing data centers.

[0003] The invention patent with publication number CN115664798A discloses an intelligent switching device and method for dual-path power supply in a data center. This solution collects voltage, current and temperature parameters of the power supply circuit from multiple dimensions, realizes anomaly judgment and switching timing control based on parameter thresholds, and adds a circuit fault self-diagnosis unit, which can quickly locate the obvious fault point of the circuit. At the same time, it can adjust the switching power threshold according to the total load at the end, realizing basic power supply safety management in high-density computing data centers.

[0004] In practical applications, existing dual-power supply management technologies do not address the dynamic fluctuations of high-density computing loads by performing dynamic collaborative analysis and pre-calibration of the core power supply parameters of the dual-power supply circuits. The phase difference and line impedance of the two power supplies will dynamically become unbalanced as the end load changes in real time. This imbalance will not directly trigger an alarm for exceeding electrical parameters, but it will generate low-amplitude continuous harmonics on the power supply busbar during the operation or switching of dual power supply. These harmonics will directly affect the power supply ripple stability of high-density computing chips. Long-term ripple distortion will gradually reduce the computing accuracy of the computing chips. Moreover, existing technologies lack pre-identification methods for such dynamic imbalances and have no corresponding real-time calibration mechanism, making it impossible to avoid such problems at their root. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a dual-power supply safety management system and method for data centers. It solves the technical problems of existing technologies being unable to adapt to the dynamic fluctuation characteristics of high-density computing loads, lacking a pre-identification and real-time calibration mechanism for dynamic imbalance of dual-power supply parameters, and easily causing low-amplitude continuous harmonics on the busbar and reducing the computing accuracy of computing chips.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data center dual-power supply security management system, comprising: The phase-separated multi-point parameter acquisition module is connected to the incoming, outgoing, and busbar connection terminals of the dual-power supply main and backup circuits, respectively, to synchronously acquire real-time monitoring data of phase, line impedance, load current, and voltage ripple. The high-density load trend prediction module is connected to the computing power scheduling platform to extract historical load data and real-time scheduling instructions, and to predict the load change trend within a preset period of time. The dual-path parameter dynamic coupling analysis module is connected to the phase-separated multi-point parameter acquisition module and the high-density load trend prediction module, respectively, and is used to extract the coupling correlation characteristics between power supply parameters and load data, and determine the imbalance level. The phase impedance adaptive calibration module is used to dynamically adjust the phase difference and line impedance of the dual power supply circuits according to the imbalance level and coupling correlation characteristics. The busbar harmonic graded suppression module is connected to the power supply busbar and is used to synchronously start the harmonic suppression program of the corresponding level during the phase impedance adjustment process. The real-time calibration effect verification module is used to compare the parameter values ​​after calibration with the target values ​​and to detect the busbar harmonic elimination status. The self-learning and iterative module for control strategies is used to store control process data and analyze calibration feedback under different imbalance scenarios to generate updated control strategies. The intelligent central control module establishes bidirectional communication connections with each of the above modules to coordinate the runtime sequence and data interaction of each module.

[0007] Furthermore, the phase-separated multi-point parameter acquisition module includes an input monitoring unit located at the incoming end of the dual-power supply main and backup circuits, an output monitoring unit located at the outgoing end, and an access monitoring unit located at the busbar connection end. The input monitoring unit, output monitoring unit, and access monitoring unit all have phase-separated sampling functions, which can realize the synchronous extraction of the voltage phase angle and circuit equivalent impedance of phases A, B, and C, and convert the monitoring data into a synchronous data sequence with timestamps and send it to the intelligent master control module.

[0008] Furthermore, the high-density load trend prediction module includes a computing power data interface unit and a load projection model unit. The computing power data interface unit obtains the total amount of computing tasks allocated, the number of computing nodes started, and the expected power change gradient within a future preset period from the computing power scheduling platform. The load projection model unit calculates the load increase / decrease rate and load phase offset estimate within a future preset period based on the total amount of computing tasks allocated and the power change gradient, and generates a load prediction data package.

[0009] Furthermore, the dual-path parameter dynamic coupling analysis module includes a coupling feature extraction unit, an imbalance trend inference unit, and an imbalance level determination unit. The coupling feature extraction unit is used to construct a multi-dimensional feature space, associate and map the phase difference fluctuation and impedance deviation values ​​in the real-time monitoring data with the load prediction data, and extract the coupling feature quantities that reflect the fluctuation of power supply parameters with load changes. The imbalance trend inference unit calculates the phase difference and the offset trajectory of the line impedance on the future time axis based on the coupling characteristic quantity. The imbalance level determination unit determines the current imbalance level by matching the slope and influence range of the offset trajectory with a preset imbalance threshold range.

[0010] Furthermore, the phase impedance adaptive calibration module includes a tuning drive unit and a multi-level calibration execution unit. The multi-level calibration execution unit is connected in series or in parallel in the dual power supply circuit and includes multiple sets of adjustable reactance components and phase shifting components. The tuning drive unit receives the calibration command issued by the intelligent master control module, calculates the required impedance compensation amount and phase compensation angle, and controls the multi-level calibration execution unit to switch to the corresponding physical compensation level, so as to realize the dynamic fine adjustment of the line impedance and phase difference in the dual power supply circuit, so that the two power supply parameters tend to be balanced.

[0011] Furthermore, the busbar harmonic hierarchical suppression module includes a harmonic feature identification unit and a hierarchical suppression execution unit. The harmonic feature identification unit monitors the voltage and current waveforms of the power supply busbar in real time and extracts low-amplitude continuous harmonic components with frequencies within a preset range. The graded suppression execution unit presets the suppression power level in advance based on the imbalance level determination result, and synchronously adjusts the filter center frequency of the active filter or notch filter when the phase impedance adaptive calibration module is activated, so as to eliminate induced harmonics and residual harmonics caused by parameter fine-tuning.

[0012] Furthermore, the real-time calibration effect verification module includes a parameter verification unit and a harmonic detection unit. After the phase impedance adjustment is completed, the parameter verification unit reacquires the real-time phase difference data and impedance data of the dual power supply circuit and calculates the deviation value between them and the preset balance target. The harmonic detection unit is used to detect whether the busbar ripple coefficient is within a safe range after adjustment is completed; If the deviation value exceeds the range or the ripple coefficient exceeds the standard, the parameter verification unit generates a secondary calibration trigger signal and sends it to the intelligent master control module.

[0013] Furthermore, the self-learning and iterative module of the control strategy includes a database storage unit and a strategy optimization unit. The database storage unit stores the initial parameters, load prediction data, calibration level, suppression level and verification results of each control process in a time series. The strategy optimization unit uses deep learning algorithms to perform correlation analysis on the stored data, calculates the calibration response speed and harmonic suppression rate under different load fluctuation characteristics, corrects the matching weight between the calibration level and the load characteristics, generates an updated set of control parameters and stores them in the strategy library for the intelligent master control module to call.

[0014] Furthermore, the intelligent master control module adopts a master-slave communication architecture, establishing a real-time data bus and a control command bus. The intelligent master control module sends synchronous working pulses to the phase impedance adaptive calibration module and the bus harmonic hierarchical suppression module through the control command bus, ensuring that the physical calibration action and the harmonic suppression action are triggered synchronously in the time domain, and the synchronization error is controlled within a preset millisecond threshold.

[0015] This invention also provides a method for safe management and control of dual-power supply in a data center, comprising the following steps: Step S1: Real-time monitoring data of the incoming, outgoing and busbar connection terminals of the dual power supply main and backup circuits are acquired synchronously through the phase-separated multi-point parameter acquisition module. At the same time, load change instructions are extracted from the computing power scheduling platform using the high-density load trend prediction module. Step S2: The intelligent central control module transmits the real-time monitoring data and the load change command to the dual-path parameter dynamic coupling analysis module, extracts the coupling correlation characteristics between parameters and load, and combines the load prediction data to infer the imbalance trend and classify the imbalance level; Step S3: The intelligent master control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module according to the imbalance level, and performs phase impedance fine-tuning and busbar harmonic elimination actions according to the set timing sequence. Step S4: Use the real-time calibration effect verification module to detect the adjusted loop parameters and busbar harmonic status. If the preset balance target is not achieved, return to step S3 to perform a second calibration. Step S5: The self-learning and iteration module of the control strategy collects data from the entire control process, optimizes the control strategy model by analyzing the calibration effect, and provides the updated strategy model to the intelligent central control module to achieve closed-loop control iteration.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires real-time monitoring data of electrical parameters of the dual-power supply main and backup circuits simultaneously through a phase-splitting multi-point parameter acquisition module. Combined with load trend prediction information extracted from the computing power scheduling platform, a dual-parameter dynamic coupling analysis module establishes a correlation mapping between parameters and load and accurately determines the imbalance level. Based on this, the intelligent central control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module to dynamically fine-tune the phase difference and line impedance imbalance caused by dynamic load fluctuations. It also simultaneously suppresses busbar harmonics induced by parameter fine-tuning at the source. The calibration effect real-time verification module ensures that the adjusted parameters reach the preset balance target. Finally, the self-learning iteration module of the control strategy continuously optimizes the adaptive capability to load fluctuation patterns, realizing closed-loop control of the continuous collaborative state of the dual power supply circuits in high-density computing power data centers during long-term operation. This effectively avoids the core problem that parameter imbalance may cause electrical harmonics and affect the operating accuracy of computing power chips, ensuring the stability and security of power supply quality. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the overall architecture of the dual-power supply safety management system of the present invention. Figure 2 This is a flowchart illustrating the closed-loop execution process of parameter calibration and harmonic suppression in this invention. Figure 3 This is an overall flowchart of the dual-power supply safety management method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 This embodiment provides a detailed description of the composition and operation mode of the dual-power supply safety management system for data centers. This system is adapted to the dual-main and backup power supply circuits of high-density computing data centers. It can realize real-time acquisition of power supply parameters, accurate prediction of load trends, dynamic coupling analysis and adaptive calibration of dual-path parameters, and complete the self-learning iteration of harmonic suppression and control strategies. It fundamentally solves the parameter imbalance problem caused by dynamic load fluctuations in dual-path power supply and avoids the impact of busbar harmonics on the operating accuracy of computing chips.

[0020] Please see Figures 1-2 This invention provides a data center dual-power supply security management system, comprising: The phase-separated multi-point parameter acquisition module is connected to the incoming, outgoing, and busbar connection terminals of the dual-power supply main and backup circuits, respectively, to synchronously acquire real-time monitoring data of phase, line impedance, load current, and voltage ripple. The high-density load trend prediction module is connected to the computing power scheduling platform to extract historical load data and real-time scheduling instructions, and to predict the load change trend within a preset period of time. The dual-path parameter dynamic coupling analysis module is connected to the phase-separated multi-point parameter acquisition module and the high-density load trend prediction module, respectively, and is used to extract the coupling correlation characteristics between power supply parameters and load data, and determine the imbalance level. The phase impedance adaptive calibration module is used to dynamically adjust the phase difference and line impedance of the dual power supply circuits according to the imbalance level and coupling correlation characteristics. The busbar harmonic graded suppression module is connected to the power supply busbar and is used to synchronously start the harmonic suppression program of the corresponding level during the phase impedance adjustment process. The real-time calibration effect verification module is used to compare the parameter values ​​after calibration with the target values ​​and to detect the busbar harmonic elimination status. The self-learning and iterative module for control strategies is used to store control process data and analyze calibration feedback under different imbalance scenarios to generate updated control strategies. The intelligent central control module establishes bidirectional communication connections with each of the above modules to coordinate the runtime sequence and data interaction of each module.

[0021] Specifically, all of the above modules are built on industrial-grade embedded hardware, equipped with high-speed data processing chips and real-time communication modules. The modules interact with each other through a shielded industrial bus. The intelligent central control module, as the core control unit, is deployed in the power supply monitoring room of the data center. It realizes human-machine interaction through an industrial touch terminal, which can display the operating status of each module and the parameter data of the power supply circuit in real time, and supports the issuance of manual intervention calibration commands. The hardware of the phase-separated multi-point parameter acquisition module adopts a high-precision three-phase AC sampling chip with sampling accuracy reaching the microvolt and microampere levels, enabling distortion-free acquisition of power supply parameters. The high-density load trend prediction module is equipped with an edge computing chip, which can perform local extrapolation of load data and reduce communication latency with the computing power scheduling platform. The dual-path parameter dynamic coupling analysis module and the phase impedance adaptive calibration module share a high-speed digital signal processing chip, enabling millisecond-level issuance of parameter analysis and calibration commands. The busbar harmonic graded suppression module is equipped with modular active filters and notch filters, which can be switched according to actual needs. The real-time calibration effect verification module adopts a dual-channel parameter comparison acquisition unit to achieve synchronous acquisition and comparison of parameters before and after calibration. The control strategy self-learning iteration module is equipped with a deep learning inference framework and a large-capacity solid-state drive for storing full-process control data. All modules are powered by data center UPS uninterrupted power supply to ensure the continuity and stability of module operation.

[0022] In one embodiment, the phase-separated multi-point parameter acquisition module includes an input monitoring unit located at the incoming end of the dual-power supply main and backup circuit, an output monitoring unit located at the outgoing end, and an access monitoring unit located at the busbar connection end. The input monitoring unit, output monitoring unit, and access monitoring unit all have phase-separated sampling functions, which can realize the synchronous extraction of the voltage phase angle and circuit equivalent impedance of phase A, phase B, and phase C, and convert the monitoring data into a synchronous data sequence with timestamps and send it to the intelligent master control module.

[0023] Specifically, each monitoring unit of the phase-separated multi-point parameter acquisition module is an independent hardware module. They are physically connected to the terminals of the dual-power supply main and backup circuits via aviation plugs. The input monitoring unit is installed in the dual-power supply incoming cabinet and connected to the outgoing side of the high-voltage circuit breaker in the main and backup circuits to collect power supply parameters from the incoming end. The output monitoring unit is installed in the dual-power supply outgoing cabinet and connected to the incoming side of the distribution cabinet to collect power supply parameters after circuit distribution. The access monitoring unit is directly installed on the copper busbar of the power supply busbar, using snap-on current and voltage sensors, allowing installation without disconnecting the busbar, and is used to collect real-time parameters from the busbar access end. The phase-separated sampling function of each monitoring unit is achieved through three independent sampling channels. Each sampling channel corresponds to one of phases A, B, and C. Opto-isolation technology is used between channels to avoid inter-phase interference and ensure the accuracy of the sampling data. The voltage phase angle is extracted using the phase detection function of the sampling chip. The acquired voltage sine wave is compared with a standard reference sine wave to calculate the phase difference. The equivalent impedance of the loop is calculated using Ohm's law; that is, the voltage and current values ​​are acquired at the same moment, and the single-phase equivalent impedance is calculated using Z=U / I. U is the equivalent impedance of the loop, U is the effective value of the phase voltage, and I is the effective value of the phase current. The raw data collected by each monitoring unit is converted into digital signals after analog-to-digital conversion. Then, a timestamp is added by the internal clock module. The timestamp accuracy is at the millisecond level. The clock modules of all monitoring units are synchronized with the master clock of the intelligent control module to ensure the synchronization of the timestamps. The synchronization data sequence is encapsulated in JSON data format and sent to the intelligent control module through RS485 industrial bus. The baud rate of data transmission is set to 115200bps to ensure the real-time data transmission.

[0024] In one embodiment, the high-density load trend prediction module includes a computing power data interface unit and a load projection model unit. The computing power data interface unit obtains the total amount of computing tasks allocated, the number of computing nodes started, and the expected power change gradient within a future preset period from the computing power scheduling platform. The load projection model unit calculates the load increase / decrease rate and load phase offset estimate within a future preset period based on the total amount of computing tasks allocated and the power change gradient, and generates a load prediction data package.

[0025] Specifically, the computing power data interface unit of the high-density load trend prediction module adopts standardized industrial communication interfaces, including Ethernet and OPC UA interfaces, enabling seamless integration with mainstream computing power scheduling platforms without requiring secondary development. The interface unit has a built-in data parsing module that can parse scheduling instructions in JSON, XML, and other formats issued by the computing power scheduling platform and extract valid data. The preset period can be flexibly set according to the data center's computing power scheduling needs, for example, to 5 minutes, 10 minutes, or 30 minutes. The value is based on the data center's computing power task allocation cycle and is usually consistent with the task scheduling cycle of the computing power scheduling platform. The total amount of computing tasks allocated is measured based on the number of cores and runtime of the computing power tasks. The number of computing nodes started is the number of servers, switches, and other computing power devices to be started in the data center. The expected power change gradient is the power growth rate after the computing power devices are started, given by the computing power scheduling platform based on historical operating data. The core of the load projection model unit is a time-series-based load projection model. This model takes the total amount of computational tasks Q, the number of computational nodes N, and the power change gradient k as input parameters to calculate the load increase / decrease rate v and the estimated load phase shift Δφ within a future preset time period t. The formula for calculating the load increase / decrease rate is as follows: , In the formula, This is the average rated power of a single computing node, and the value is based on the nameplate parameters of the computing equipment in the data center, for example, 2kW / unit. The total number of tasks is calculated based on the total number of tasks at full load in the data center; α is the load correction coefficient, with a value ranging from 0.8 to 1.2, and is determined by the ratio of the actual power consumption to the theoretical power consumption of historical computing tasks through linear regression analysis of historical data.

[0026] The formula for calculating the estimated load phase offset is as follows: , In the formula, The reference load phase angle is determined based on the rated phase angle of the dual power supply circuit. The total rated load of the dual power supply circuits is determined based on the power supply design parameters of the data center; β is the phase offset correction coefficient, ranging from 0.9 to 1.1, and is determined based on the correlation data between historical load changes and phase offsets. The load projection model unit encapsulates the calculated load increase / decrease rate and load phase offset estimate into a load prediction data packet. The data packet also contains the time nodes of the preset time period and the corresponding load parameters, and is sent to the dual-parameter dynamic coupling analysis module and the intelligent central control module via the industrial bus.

[0027] In one embodiment, the dual-path parameter dynamic coupling analysis module includes a coupling feature extraction unit, an imbalance trend deduction unit, and an imbalance level determination unit. The coupling feature extraction unit is used to construct a multi-dimensional feature space, correlate and map the phase difference fluctuation and impedance deviation values ​​in real-time monitoring data with load prediction data, and extract coupling feature quantities that reflect the fluctuation of power supply parameters with load changes. The imbalance trend deduction unit calculates the offset trajectory of phase difference and line impedance on the future time axis based on the coupling feature quantities. The imbalance level determination unit matches a preset imbalance threshold range with the slope and influence range of the offset trajectory to determine the current imbalance level.

[0028] Specifically, the coupling feature extraction unit of the dual-path parameter dynamic coupling analysis module constructs a three-dimensional feature space, with phase difference fluctuation, impedance deviation, and load change as the three-dimensional coordinate axes. Phase difference fluctuation represents the real-time change in the phase difference between phases A, B, and C of the dual-path power supply main and backup circuits; impedance deviation represents the equivalent impedance difference between the dual-path power supply main and backup circuits; and load change is the product of the load increase / decrease rate in the load prediction data and the preset time period. The coupling feature extraction unit normalizes the real-time monitoring data from the phase-separated multi-point parameter acquisition module and the load prediction data from the high-density load trend prediction module to eliminate dimensional differences. The normalization process uses the maximum-minimum method to map the data to the range of 0 to 1. The normalized data is then substituted into the three-dimensional feature space, and the coupling feature quantity F is extracted using an association mapping algorithm. The calculation formula is as follows: , In the formula, Δθ is the normalized phase difference fluctuation value, ΔZ is the normalized impedance deviation value, and ΔP is the normalized load change. , , The weighting coefficients are determined based on the degree of influence of each parameter on the imbalance of dual-power supply parameters. They are obtained through multiple linear regression analysis of historical power supply failure data of the data center. For example, taking... , , And satisfy .

[0029] The imbalance trend projection unit uses the coupling characteristic quantity F as input and employs a polynomial fitting algorithm to calculate the offset trajectory of phase difference and line impedance on the future time axis. The fitting formula is as follows: , In the formula, y is the phase difference or line impedance offset, and t is the time node of the future time axis. , , … Here, is the fitting coefficient, and n is the fitting order, determined based on the changing trend of the coupled feature quantity, typically 3rd or 4th order, to achieve accurate fitting of the offset trajectory. The imbalance level determination unit presets an imbalance threshold range based on the data center's power supply standards and the tolerance of the computing equipment, dividing the imbalance level into three levels: mild, moderate, and severe. Each level corresponds to a different offset trajectory slope range and influence range. The slope of the offset trajectory reflects the rate of change of the parameter imbalance; a larger slope indicates faster imbalance development. The influence range reflects the number of phases and circuits involved in the parameter imbalance; for example, a single-phase deviation is a small range, while a three-phase average deviation is a large range. When both the slope and influence range of the offset trajectory are in the low range, it is determined to be mild imbalance; when the slope or influence range is in the middle range, it is determined to be moderate imbalance; when both the slope and influence range are in the high range, it is determined to be severe imbalance. The imbalance level determination result is sent to the intelligent central control module via a digital signal as the basis for issuing calibration commands.

[0030] In one embodiment, the phase impedance adaptive calibration module includes a tuning drive unit and a multi-level calibration execution unit. The multi-level calibration execution unit is connected in series or in parallel in a dual power supply circuit and includes multiple sets of adjustable reactance components and phase shifting components. The tuning drive unit receives the calibration command issued by the intelligent master control module, calculates the required impedance compensation amount and phase compensation angle, and controls the multi-level calibration execution unit to switch to the corresponding physical compensation level, so as to realize the dynamic fine adjustment of the line impedance and phase difference in the dual power supply circuit, so that the two power supply parameters tend to be balanced.

[0031] Specifically, the tuning drive unit of the phase impedance adaptive calibration module is the core control hardware, with a built-in microprocessor and digital-to-analog converter module. It can convert the digital calibration commands issued by the intelligent master control module into analog control signals. The calibration commands received by the tuning drive unit include information such as the imbalance level, the number of imbalance phases, the phase difference deviation value, and the impedance deviation value. Based on this information, the tuning drive unit calculates the impedance compensation amount and the phase compensation angle. The formula for calculating the impedance compensation amount is: , In the formula, This is the impedance compensation amount. This is the reference equivalent impedance for the dual power supply circuit, and its value is based on the power supply design parameters of the data center. This refers to the actual equivalent impedance acquired by the phase-splitting multi-point parameter acquisition module. The formula for calculating the phase compensation angle is: , In the formula, For phase compensation angle, The reference phase difference for the dual power supply circuits is set to 0°, meaning the main and backup circuits are in phase. This refers to the actual phase difference acquired by the phase-separated multi-point position parameter acquisition module.

[0032] The multi-level calibration execution unit is a hardware actuator. Depending on the voltage level and load capacity of the dual power supply circuits, it connects to the circuits in series or parallel. Series connection is used for high-voltage circuits, and parallel connection for low-voltage circuits. The reactor component uses a dry-type adjustable reactor, and impedance is adjusted by changing the number of coil turns. The phase-shifting component uses an electronic phase shifter, and phase adjustment is achieved by adjusting the conduction angle of the thyristor. Both the reactor and phase-shifting components are modularly designed, with each component corresponding to one phase, enabling phase-by-phase calibration. The multi-level calibration execution unit has multiple preset physical compensation levels, each corresponding to a fixed impedance compensation amount and phase compensation angle. The number of levels is set according to the load fluctuation range of the data center. For example, when the load fluctuation range is 0~100%, 10 levels are set, each corresponding to a 10% fluctuation range. The level values ​​are based on the calculated impedance compensation amount and phase compensation angle. The tuning drive unit compares the calculated results with the compensation values ​​of each level and controls the relay switch to switch to the closest level, achieving physical compensation. For mild imbalance, the tuning drive unit controls the multi-level calibration execution unit to switch to the low level for small-amplitude fine-tuning; for moderate imbalance, it switches to the medium level for medium-amplitude calibration; for severe imbalance, it switches to the high level for large-amplitude compensation. During the calibration process, the tuning drive unit receives feedback data from the phase-splitting multi-point parameter acquisition module in real time and dynamically adjusts the impedance compensation amount and phase compensation angle according to the feedback data to ensure the accuracy of calibration and achieve gradual balancing of the dual power supply parameters.

[0033] In one embodiment, the busbar harmonic hierarchical suppression module includes a harmonic feature identification unit and a hierarchical suppression execution unit. The harmonic feature identification unit monitors the voltage and current waveforms of the power supply busbar in real time and extracts low-amplitude continuous harmonic components with frequencies within a preset range. The graded suppression execution unit presets the suppression power level in advance based on the imbalance level determination result, and synchronously adjusts the filter center frequency of the active filter or notch filter when the phase impedance adaptive calibration module is activated, so as to eliminate induced harmonics and residual harmonics caused by parameter fine-tuning.

[0034] Specifically, the harmonic feature identification unit of the busbar harmonic hierarchical suppression module uses a high-speed waveform acquisition chip and harmonic analysis algorithm to acquire the voltage and current waveforms of the power supply busbar in real time. The sampling frequency is 2048Hz, which can achieve accurate acquisition of high-order harmonics. The preset harmonic frequency range is 3rd to 50th, which is set according to the harmonic distribution characteristics of the data center power supply system. The amplitude threshold of low-amplitude continuous harmonics is set to 0.5% of the rated voltage, and the duration threshold is set to 1 second. That is, harmonics with an amplitude lower than 0.5% of the rated voltage and a duration of more than 1 second are low-amplitude continuous harmonics. The harmonic feature identification unit converts the time-domain waveform of voltage and current into a frequency-domain spectrum through Fourier transform, and extracts the frequency, amplitude and phase of harmonic components that meet the preset range as the basis for harmonic suppression.

[0035] The graded suppression execution unit pre-sets suppression power levels based on the imbalance level determination results. Each suppression power level corresponds one-to-one with an imbalance level: low suppression power for mild imbalance, medium suppression power for moderate imbalance, and high suppression power for severe imbalance. The suppression power value is based on the total power of the harmonic components; for example, low suppression power is 10kVA, medium suppression power is 50kVA, and high suppression power is 100kVA, ensuring that the suppression power covers the total harmonic power. Simultaneously with the phase impedance adaptive calibration module, the graded suppression execution unit synchronously adjusts the center frequency of the active filter or notch filter to the harmonic frequency extracted by the harmonic feature identification unit. The active filter suppresses harmonics over a wide frequency range, while the notch filter suppresses harmonics at specific frequencies. Working together, they achieve comprehensive elimination of induced and residual harmonics. Induced harmonics are new harmonics caused by the operation of reactance components and phase shifting components during phase impedance adjustment. Residual harmonics are harmonics that were not eliminated and were originally present in the power supply system. The graded suppression execution unit ensures that the harmonic components of the busbar are always within a safe range throughout the entire phase impedance calibration process by adjusting the filtering parameters in real time, thus avoiding the impact of harmonics on the computing chip.

[0036] In one embodiment, the real-time calibration effect verification module includes a parameter verification unit and a harmonic detection unit. After the phase impedance adjustment is completed, the parameter verification unit reacquires the real-time phase difference data and impedance data of the dual power supply circuit and calculates the deviation value between them and the preset balance target. The harmonic detection unit is used to detect whether the busbar ripple coefficient is within a safe range after adjustment is completed; If the deviation value exceeds the range or the ripple coefficient exceeds the standard, the parameter verification unit generates a secondary calibration trigger signal and sends it to the intelligent master control module.

[0037] Specifically, the parameter verification unit of the real-time calibration effect verification module uses the same sampling hardware as the phase-separated multi-point parameter acquisition module. After the phase impedance adjustment is completed, the intelligent central control module issues a sampling command, and the parameter verification unit immediately re-acquires the phase difference and equivalent impedance of phases A, B, and C of the dual power supply circuit. The acquisition is performed 10 times consecutively, and the average value is taken as the real-time parameter data. The preset balance target is that the phase difference of the dual power supply circuit is ≤1° and the equivalent impedance deviation is ≤0.1Ω. This target value is set according to the power supply requirements of high-density computing equipment in the data center. The deviation value is calculated using the absolute deviation method. The phase difference deviation value is the absolute value of the actual phase difference acquired and the preset balance target phase difference, and the impedance deviation value is the absolute value of the actual equivalent impedance deviation acquired and the preset balance target impedance deviation. The harmonic detection unit uses a ripple coefficient detection chip to directly acquire the voltage ripple coefficient of the power supply bus. The safe range of the ripple coefficient is set to ≤0.1%. This range is set according to the power supply ripple requirements of the computing chip. If the ripple coefficient exceeds this range, it is determined that the harmonic suppression has not met the standard.

[0038] The detection results from the parameter verification unit and the harmonic detection unit are simultaneously sent to the intelligent control module. If the phase difference deviation and impedance deviation are both within the preset range and the ripple coefficient is within the safe range, the calibration effect is deemed satisfactory, and the calibration process ends. If any deviation value exceeds the range or the ripple coefficient exceeds the standard, the parameter verification unit immediately generates a secondary calibration trigger signal. This signal is a high-level digital signal with a duration of 1 second. After receiving the secondary calibration trigger signal, the intelligent control module immediately issues a secondary calibration command. The dual-path parameter dynamic coupling analysis module re-couples and analyzes the current power supply parameters and load data. The phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module perform secondary calibration based on the new analysis results until all parameters reach the preset standard, ensuring the balance of dual-path power supply parameters and the effective elimination of busbar harmonics.

[0039] In one embodiment, the control strategy self-learning iteration module includes a database storage unit and a strategy optimization unit. The database storage unit stores the initial parameters, load prediction data, calibration level, suppression level and verification results for each control process in a time series. The strategy optimization unit uses deep learning algorithms to perform correlation analysis on the stored data, calculates the calibration response speed and harmonic suppression rate under different load fluctuation characteristics, corrects the matching weight between the calibration level and the load characteristics, generates an updated set of control parameters and stores them in the strategy library for the intelligent master control module to call.

[0040] Specifically, the database storage unit of the self-learning and iterative module for control strategies adopts a combined relational and time-series database storage method. The relational database is used to store structured data such as calibration levels, suppression levels, and imbalance levels, while the time-series database is used to store massive amounts of data that change over time, such as initial parameters, load prediction data, and verification results. The database storage period can be set according to the data center's needs, for example, to one year. Data exceeding the storage period can be automatically archived or deleted to save storage space. The database storage unit sorts the control data according to the time sequence, and each piece of control data contains a unique timestamp and control process number, enabling quick querying and retrieval of data from any control process.

[0041] The strategy optimization unit is equipped with a gradient boosting tree algorithm based on deep learning. This algorithm can perform correlation analysis on the massive amount of stored management and control data. The core dimensions of the analysis are load fluctuation characteristics, calibration level, suppression level, calibration response speed and harmonic suppression rate. Load fluctuation characteristics include load increase / decrease rate, load phase shift, load fluctuation range, etc. The calibration response speed is the time required from the determination of the imbalance level to the parameter reaching the standard. The harmonic suppression rate is the proportion of harmonic components eliminated. The strategy optimization unit calculates the actual effect of each calibration level and suppression level under different load fluctuation characteristics using algorithms. For matching relationships with slow calibration response speed and low harmonic suppression rate, its weight is reduced; for matching relationships with fast calibration response speed and high harmonic suppression rate, its weight is increased. At the same time, based on new load fluctuation characteristics, corresponding matching relationships of calibration levels and suppression levels are added. The corrected matching weights and the newly added matching relationships together constitute the updated control parameter set. The control parameter set is stored in the strategy library, which is an independent data partition in the database storage unit. When the intelligent central control module issues calibration instructions, it prioritizes retrieving the corresponding control parameters from the strategy library to achieve accurate issuance of calibration instructions. As control data continues to accumulate, the strategy optimization unit will continuously update the control parameter set, making the adaptability of the control strategy increasingly higher and realizing the self-learning iteration of the control strategy.

[0042] In one embodiment, the intelligent control module adopts a master-slave communication architecture, establishing a real-time data bus and a control command bus. The intelligent control module sends synchronous working pulses to the phase impedance adaptive calibration module and the bus harmonic hierarchical suppression module through the control command bus, ensuring that the physical calibration action and the harmonic suppression action are triggered synchronously in the time domain, and the synchronization error is controlled within a preset millisecond threshold.

[0043] Specifically, the master-slave communication architecture of the intelligent central control module uses the intelligent central control module as the master and the other modules as slaves. The master is responsible for sending control commands to the slaves, and the slaves are responsible for uploading operating status and collected data to the master. The master and slaves communicate using a polling method, with the master sending communication commands to each slave in sequence. Upon receiving a command, the slave immediately replies with data, avoiding data conflicts caused by multiple slaves communicating simultaneously. The real-time data bus and the control command bus are two independent industrial buses. The real-time data bus uses an Ethernet bus to transmit massive amounts of data from the phase-splitting multi-point parameter acquisition module, the high-density load trend prediction module, the dual-channel parameter dynamic coupling analysis module, and the calibration effect real-time verification module, with a data transmission rate of 1000Mbps. The control command bus uses a CAN bus to transmit control commands and synchronous working pulses issued by the intelligent central control module. The data transmission has high real-time performance and reliability, enabling zero-delay command issuance.

[0044] The intelligent master control module sends a square wave signal with a frequency of 1kHz to the phase impedance adaptive calibration module and the bus harmonic hierarchical suppression module. The rising edge of the pulse is the action trigger signal. Upon receiving the rising edge of the synchronous working pulse, the two modules immediately execute the corresponding physical calibration and harmonic suppression actions to ensure synchronous triggering in the time domain. The preset millisecond threshold for synchronization error is set to 5ms. This value is based on the action response time of phase impedance calibration and harmonic suppression. By performing high-precision time synchronization between the master clock of the intelligent master control module and the slave clocks of each module, and optimizing the communication delay of the bus, the synchronization error is controlled within 5ms to avoid new parameter imbalances or harmonic generation caused by asynchronous actions. The intelligent central control module coordinates the operation sequence of each module and sets the working priority of each module. For example, the phase-separated multi-point parameter acquisition module has the highest priority and is always in operation; the high-density load trend prediction module has the second priority and performs load simulation according to a preset cycle; the dual-path parameter dynamic coupling analysis module starts after receiving the acquired data and prediction data; the phase impedance adaptive calibration module and the busbar harmonic graded suppression module start after receiving the imbalance level judgment result; the calibration effect real-time verification module starts after the calibration action is completed; and the control strategy self-learning iteration module starts after each control process ends. The operation sequence of each module is set through the internal program of the intelligent central control module and can be flexibly adjusted according to the actual needs of the data center.

[0045] The data center dual-power supply security management system in this embodiment achieves all-time, all-dimensional monitoring and control of dual-power supply parameters through the collaborative work of various modules. Compared with existing technologies, this system can achieve pre-identification and real-time calibration of parameter imbalances caused by dynamic load fluctuations, avoiding bus harmonics generated during grid connection or switching due to parameter imbalances. At the same time, through the self-learning iteration of the control strategy, the system continuously optimizes the calibration and harmonic suppression effects, ensuring the continuous coordination of dual-power supply parameters and providing a reliable power supply guarantee for the stable operation of high-density computing equipment.

[0046] Example 2 Please see Figure 3 The present invention also provides a method for safe management and control of dual power supply in a data center, comprising the following steps: Step S1: Real-time monitoring data of the incoming, outgoing and busbar connection terminals of the dual power supply main and backup circuits are acquired synchronously through the phase-separated multi-point parameter acquisition module. At the same time, load change instructions are extracted from the computing power scheduling platform using the high-density load trend prediction module. Step S2: The intelligent central control module transmits the real-time monitoring data and the load change command to the dual-path parameter dynamic coupling analysis module, extracts the coupling correlation characteristics between parameters and load, and combines the load prediction data to infer the imbalance trend and classify the imbalance level; Step S3: The intelligent master control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module according to the imbalance level, and performs phase impedance fine-tuning and busbar harmonic elimination actions according to the set timing sequence. Step S4: Use the real-time calibration effect verification module to detect the adjusted loop parameters and busbar harmonic status. If the preset balance target is not achieved, return to step S3 to perform a second calibration. Step S5: The self-learning and iteration module of the control strategy collects data from the entire control process, optimizes the control strategy model by analyzing the calibration effect, and provides the updated strategy model to the intelligent central control module to achieve closed-loop control iteration.

[0047] Specifically, during step S1, the input monitoring unit, output monitoring unit, and access monitoring unit of the phase-separated multi-point parameter acquisition module simultaneously activate the phase-separated sampling function to synchronously acquire the phase, line impedance, load current, and voltage ripple of phases A, B, and C of the dual-power supply main and backup circuits. The sampling frequency is set to 100Hz, which is based on the rate of change of the power supply parameters, enabling real-time tracking of the parameters. The acquired raw data is converted from analog to digital and then opto-isolated. A millisecond-level timestamp is added to convert it into a synchronous data sequence, which is then sent to the intelligent central control module via the real-time data bus. At the same time, it is backed up to the database storage unit of the self-learning and iterative module of the control strategy. The computing power data interface unit of the high-density load trend prediction module establishes a real-time communication connection with the computing power scheduling platform. It extracts load change instructions from the task scheduling module and power monitoring module of the computing power scheduling platform, including information such as the total amount of computing tasks allocated within a preset future period, the number of computing nodes started, and the expected power change gradient. The computing power data interface unit parses and deduplicates the extracted instructions, removing invalid and duplicate data, and then transmits the parsed valid data to the load projection model unit. The load projection model unit immediately starts the load projection program, calculates the load increase / decrease rate and load phase shift estimate within a preset future period based on the valid data, and generates a load prediction data packet. This load prediction data packet is simultaneously sent to the intelligent central control module and the dual-path parameter dynamic coupling analysis module, providing data support for subsequent coupling analysis. Step S1 is a continuous process; the phase-separated multi-point parameter acquisition module is always in sampling mode. The high-density load trend prediction module extracts load change instructions from the computing power scheduling platform according to a preset period, which is consistent with the task update cycle of the computing power scheduling platform, for example, 5 minutes, to ensure the timeliness of the load prediction data.

[0048] During step S2, the intelligent central control module, acting as the core data relay, unifies the data format of the real-time monitoring data and load change commands obtained in step S1, converting them all into JSON format. This data is then synchronously transmitted to the dual-path parameter dynamic coupling analysis module via the real-time data bus. A data verification mechanism is employed during transmission, using the CRC32 checksum algorithm to ensure data integrity and prevent data loss or errors. The coupling feature extraction unit of the dual-path parameter dynamic coupling analysis module immediately activates, constructing a three-dimensional feature space with phase difference fluctuations, impedance deviation values, and load changes as coordinate axes. The real-time monitoring data and load prediction data are normalized to eliminate dimensional differences before being substituted into the three-dimensional feature space for correlation mapping. The correlation mapping algorithm extracts coupling feature quantities, which reflect the intrinsic relationship between power supply parameters and load changes, serving as the core basis for predicting imbalance trends. The imbalance trend prediction unit uses the coupling feature quantities as input and employs a polynomial fitting algorithm to calculate the offset trajectories of phase difference and line impedance on the future time axis. The fitting order is determined based on the changing trend of the coupling feature quantities, ensuring that the offset trajectory accurately reflects the development trend of parameter imbalance. The imbalance level determination unit matches the preset imbalance threshold range with the slope and influence range of the offset trajectory, and divides the imbalance level into mild imbalance, moderate imbalance and severe imbalance. The imbalance level division result is sent to the intelligent central control module through the control command bus, and is simultaneously fed back to the local storage unit of the dual-path parameter dynamic coupling analysis module. The execution time of step S2 is controlled within 100ms to ensure rapid deduction and determination of the imbalance trend.

[0049] During step S3, after receiving the imbalance level determination result, the intelligent central control module immediately initiates the synchronization scheduling program. Based on the imbalance level, it determines the amplitude of phase impedance fine-tuning and the power level of harmonic suppression. Simultaneously, it generates a synchronization pulse, which is synchronously sent to the phase impedance adaptive calibration module and the busbar harmonic classification suppression module via the control command bus. The rising edge of the synchronization pulse serves as the action trigger signal, ensuring that the actions of the two modules are synchronized in the time domain, with the synchronization error controlled within a preset millisecond threshold. Upon receiving the calibration command, the tuning drive unit of the phase impedance adaptive calibration module calculates the required impedance compensation and phase compensation angle based on the imbalance level, the number of unbalanced phases, and the parameter deviation value. Then, based on the calculation results, it controls the multi-level calibration execution unit to switch to the corresponding physical compensation level, initiating dynamic fine-tuning of the phase impedance using the reactance component and the phase shifting component. The fine-tuning process is a gradual adjustment, receiving feedback data from the phase-separated multi-point parameter acquisition module after each adjustment. The compensation amplitude is adjusted based on the feedback data. For mild imbalances, a small-step, multi-time fine-tuning method is used; for severe imbalances, a large-step, fewer-time compensation method is used. Upon receiving the synchronization pulse, the harmonic characteristic identification unit of the busbar harmonic graded suppression module monitors the voltage and current waveforms of the power supply busbar in real time, extracting the frequency, amplitude, and phase of low-amplitude persistent harmonic components. The graded suppression execution unit, based on the pre-set suppression power level according to the imbalance level, synchronously adjusts the filter center frequencies of the active filter and notch filter, precisely matching the filter center frequency to the harmonic frequency. This achieves synchronous elimination of induced and residual harmonics. Throughout the phase impedance fine-tuning process, the harmonic suppression action remains operational, ensuring that busbar harmonics are always within a safe range. The execution time of step S3 is determined according to the imbalance level; approximately 12 seconds for mild imbalance and approximately 35 seconds for severe imbalance, ensuring the efficiency of calibration and harmonic suppression.

[0050] When executing step S4, after the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module have completed their actions, the intelligent central control module immediately issues a detection command to start the real-time calibration effect verification module. The parameter verification unit and the harmonic detection unit of the real-time calibration effect verification module are started simultaneously. The parameter verification unit continuously collects the real-time phase difference and impedance data of the dual power supply circuit 10 times, takes the average value as the actual parameter data, and then calculates the deviation value between the actual parameter data and the preset balance target. The preset balance target is that the phase difference of the dual power supply circuit is ≤1° and the equivalent impedance deviation is ≤0.1Ω. This target value is set according to the power supply requirements of the high-density computing equipment. The harmonic detection unit performs real-time detection of the ripple coefficient of the power supply busbar. The safe range of the ripple coefficient is set to ≤0.1%. If the phase difference deviation and impedance deviation values ​​calculated by the parameter verification unit are both within the preset range, and the ripple coefficient detected by the harmonic detection unit is within the safe range, then the calibration is determined to have achieved the preset balance target, step S4 is completed, and the calibration phase of this control process ends. If any deviation value exceeds the range or the ripple coefficient exceeds the standard, the parameter verification unit immediately generates a secondary calibration trigger signal and sends it to the intelligent control module. After receiving the trigger signal, the intelligent control module immediately returns to step S3 and issues a secondary calibration command. The phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module adjust the calibration and harmonic suppression strategies according to the current parameter deviation and perform secondary calibration. The compensation amplitude of the secondary calibration is determined according to the deviation value of the primary calibration, which is usually 1.2 to 1.5 times that of the primary calibration, until the results of parameter verification and harmonic detection both reach the preset standard, ensuring the balance of dual power supply parameters and the effective elimination of busbar harmonics.

[0051] When executing step S5, after the calibration phase of this control process is completed, the intelligent central control module packages and sends the entire process data of this control to the control strategy self-learning iteration module. The entire process data includes the real-time monitoring data and load prediction data of step S1, the coupling characteristic quantity and imbalance level judgment result of step S2, the calibration level, suppression level and action duration of step S3, and the verification result and number of secondary calibrations of step S4. The database storage unit of the control strategy self-learning iteration module stores these data into the corresponding database partitions according to the time sequence, providing data support for strategy optimization. The strategy optimization unit initiates a deep learning algorithm to perform correlation analysis on the current and historical control data stored in the database. The core of the analysis is the matching relationship between different load fluctuation characteristics and calibration levels and suppression levels, as well as the calibration response speed and harmonic suppression rate under this matching relationship. The algorithm calculates the weight of each matching relationship, reducing the weight of matching relationships with slow calibration response speed and low harmonic suppression rate, and increasing the weight of matching relationships with fast calibration response speed and high harmonic suppression rate. At the same time, according to new load fluctuation characteristics, new corresponding matching relationships are added. The revised weights and the new matching relationships together constitute the optimized control strategy model, which is stored in the strategy library in the form of a control parameter set. In subsequent control processes, the intelligent master control module prioritizes retrieving the corresponding control parameters from the strategy library and distributes them to each execution module, realizing the self-learning iteration of the control strategy. Step S5 is executed automatically in the background and does not affect the real-time operation of the front-end modules. After each control process is completed, a strategy optimization will be performed. As control data is continuously accumulated, the adaptability and accuracy of the control strategy model will continue to improve, ultimately achieving fully automatic dual-circuit power supply safety control without manual intervention, forming a closed-loop control iteration system.

[0052] Example 3 To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below in conjunction with a specific application scenario.

[0053] This embodiment selects a high-density cloud computing data center as a practical application scenario to verify the practical application of the dual-power supply security management system and method for the data center. The data center is equipped with thousands of high-performance computing servers. The rated voltage of the dual main and backup power supply circuits is 10kV, and the rated load is 2000kVA. The computing power scheduling platform adopts a distributed scheduling architecture, and the load fluctuation frequency is high and the fluctuation range is large. Traditional dual-power supply management technology cannot adapt to the characteristics of its dynamic load fluctuation, and low-amplitude continuous harmonics on the busbar often occur, affecting the operating accuracy of the computing power chip. The present invention is applied to the dual-power supply system of this data center to verify its practical application effect.

[0054] The data center has two main power supply circuits: mains circuit 1 and mains circuit 2. Both circuits are connected to the same power supply busbar to power the computing server cluster. When applying the management and control system of this invention, the hardware installation and parameter configuration of each module are first completed according to the power supply parameters and load characteristics of the data center. The input monitoring unit of the phase-separated multi-point parameter acquisition module is installed in the 10kV incoming line cabinet, the output monitoring unit is installed in the low-voltage distribution cabinet, and the access monitoring unit is snap-fitted onto the copper busbar of the power supply busbar. The sampling frequency of each monitoring unit is set to 100Hz, and the timestamp is synchronized with the Beidou clock of the data center. The computing power data interface unit of the high-density load trend prediction module is connected to the computing power scheduling platform of the data center via Ethernet. The preset cycle is set to 5 minutes, and the fitting order of the load projection model is set to order 3. The imbalance threshold range of the dual-path parameter dynamic coupling analysis module is set according to the tolerance of the computing power equipment of the data center, and the slopes of mild, moderate, and severe imbalance are defined. The ranges are set to 0.5° / s, 0.51° / s, and >1° / s, respectively. The multi-level calibration execution unit of the phase impedance adaptive calibration module is connected to the low-voltage circuit in parallel, with 10 physical compensation levels covering a load range of 0-2000kVA. The active filter and notch filter of the busbar harmonic graded suppression module have a filter center frequency range of 3rd and 50th, and suppression power levels of 10kVA, 50kVA, and 100kVA, respectively. The preset balance target of the calibration effect real-time verification module is set to phase difference ≤1°, equivalent impedance deviation ≤0.1Ω, and ripple coefficient safety range ≤0.1%. The intelligent master control module is deployed in the power supply monitoring room of the data center, equipped with an industrial touch terminal to realize the monitoring of the operating status of each module and manual intervention. The modules are connected through a shielded industrial bus, the real-time data bus uses Ethernet, the control command bus uses CAN bus, and the synchronization error of the synchronous working pulse is controlled within 5ms. After the system hardware is installed, software debugging and parameter calibration are performed to ensure that each module is running normally and that data interaction is without delay or error. Table 1 shows the basic parameters of the dual power supply and the configuration parameters of the management and control system for this data center.

[0055] Table 1

[0056] After the installation and configuration of the control system are completed, the control method of this invention is applied to the dual power supply system of the data center. The full-process control process is initiated. The phase-separated multi-point parameter acquisition module is always in sampling state, acquiring phase, line impedance, load current and voltage ripple data of the main and backup power supply circuits, incoming and outgoing terminals and busbar connection terminals in real time. The high-density load trend prediction module extracts a load change command from the computing power scheduling platform every 5 minutes and predicts the load change trend in the next 5 minutes. The dual-parameter dynamic coupling analysis module performs coupling analysis on the acquired data and prediction data, predicts the imbalance trend in real time and determines the imbalance level. The intelligent central control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module according to the imbalance level to perform phase impedance calibration and harmonic suppression actions. The calibration effect real-time verification module performs effect detection after the action is completed. If the standard is not met, a second calibration is performed. The control strategy self-learning iteration module collects full-process data and optimizes the control strategy model after each control process.

[0057] In practical applications, the data center's computing power scheduling platform issued multiple computing task scheduling instructions of varying scales, including the phased startup of computing nodes and the dynamic allocation of computing tasks. This resulted in varying degrees of load fluctuations in the dual power supply circuits, with the load increase / decrease rate varying between 0.5 kVA / s and 5 kVA / s. Before applying the management system and method of this invention, these load fluctuations would cause phase differences and impedance deviations in the dual power supply circuits, resulting in low-amplitude continuous harmonics on the busbars and ripple coefficients exceeding the safe range, affecting the operational accuracy of the computing chips. After applying the management system and method of this invention, the management system can quickly extract load fluctuation characteristics, identify parameter imbalance trends in advance, and perform precise phase impedance calibration and harmonic suppression actions based on the imbalance level. The real-time verification module for calibration effects can ensure that parameters reach the preset balance target, and the ripple coefficient remains within the safe range. Even under conditions of large load fluctuations, the phase difference of the dual power supply circuits can still be controlled within 1°, the equivalent impedance deviation within 0.1Ω, and no significant harmonics are generated on the busbars. Meanwhile, as the control process continues, the self-learning iteration module of the control strategy continuously optimizes the control strategy model, and the calibration response speed and harmonic suppression effect are continuously improved. In subsequent control processes, the number of secondary calibrations is significantly reduced, realizing rapid adaptation to dynamic load fluctuations. Table 2 shows the key monitoring indicators of the control system under different load fluctuation scenarios in this data center.

[0058] Table 2

[0059] During a month-long practical application verification, the control system and method of this invention operated stably throughout, with no module failures, no delays or errors in data interaction, and the parameters of the dual power supply circuits remained coordinated. No low-amplitude persistent harmonics caused by dynamic load fluctuations appeared on the power supply busbar, and the operating accuracy of the computing server was not affected by power quality. The safety and stability of the dual power supply in the data center were significantly improved. Simultaneously, the intelligent central control module of the control system can display the real-time operating status and power supply parameters of each module, supporting manual intervention and facilitating daily monitoring and management by data center maintenance personnel. The hardware of the control system adopts a modular design, facilitating future maintenance and upgrades, and the software strategy library can be continuously optimized through self-learning iteration to adapt to future computing power expansion and load change requirements of the data center.

[0060] The practical application verification results of this embodiment show that the data center dual-power supply safety management and control system and method of the present invention can effectively adapt to the characteristics of dynamic load fluctuations in high-density computing power data centers, realize the pre-identification and real-time calibration of dynamic imbalance of dual-power supply parameters, and simultaneously eliminate busbar harmonics. It solves the problems existing in the prior art from the root, improves the safety management and control level of dual-power supply in data centers, and has good practical application effects and promotion value.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data center dual-power supply security management system, characterized in that, include: The phase-separated multi-point parameter acquisition module is connected to the incoming, outgoing, and busbar connection terminals of the dual-power supply main and backup circuits, respectively, to synchronously acquire real-time monitoring data of phase, line impedance, load current, and voltage ripple. The high-density load trend prediction module is connected to the computing power scheduling platform to extract historical load data and real-time scheduling instructions, and to predict the load change trend within a preset period of time. The dual-path parameter dynamic coupling analysis module is connected to the phase-separated multi-point parameter acquisition module and the high-density load trend prediction module, respectively, and is used to extract the coupling correlation characteristics between power supply parameters and load data, and determine the imbalance level. The phase impedance adaptive calibration module is used to dynamically adjust the phase difference and line impedance of the dual power supply circuits according to the imbalance level and coupling correlation characteristics. The busbar harmonic graded suppression module is connected to the power supply busbar and is used to synchronously start the harmonic suppression program of the corresponding level during the phase impedance adjustment process. The real-time calibration effect verification module is used to compare the parameter values ​​after calibration with the target values ​​and to detect the busbar harmonic elimination status. The self-learning and iterative module for control strategies is used to store control process data and analyze calibration feedback under different imbalance scenarios to generate updated control strategies. The intelligent central control module establishes bidirectional communication connections with each of the above modules to coordinate the runtime sequence and data interaction of each module.

2. The data center dual-power supply security management system according to claim 1, characterized in that, The phase-separated multi-point parameter acquisition module includes an input monitoring unit located at the incoming end of the dual-power supply main and backup circuits, an output monitoring unit located at the outgoing end, and an access monitoring unit located at the busbar connection end. The input monitoring unit, output monitoring unit, and access monitoring unit all have phase-separated sampling functions, which can realize the synchronous extraction of the voltage phase angle and circuit equivalent impedance of phases A, B, and C, and convert the monitoring data into a synchronous data sequence with timestamps and send it to the intelligent master control module.

3. The data center dual-power supply security management system according to claim 1, characterized in that, The high-density load trend prediction module includes a computing power data interface unit and a load projection model unit. The computing power data interface unit obtains the total amount of computing tasks allocated, the number of computing nodes started, and the expected power change gradient within a future preset period from the computing power scheduling platform. The load projection model unit calculates the load increase / decrease rate and load phase offset estimate within a future preset period based on the total amount of computing tasks allocated and the power change gradient, and generates a load prediction data package.

4. The data center dual-power supply security management system according to claim 1, characterized in that, The dual-path parameter dynamic coupling analysis module includes a coupling feature extraction unit, an imbalance trend inference unit, and an imbalance level determination unit. The coupling feature extraction unit is used to construct a multi-dimensional feature space, associate and map the phase difference fluctuation and impedance deviation values ​​in the real-time monitoring data with the load prediction data, and extract the coupling feature quantities that reflect the fluctuation of power supply parameters with load changes. The imbalance trend inference unit calculates the phase difference and the offset trajectory of the line impedance on the future time axis based on the coupling characteristic quantity. The imbalance level determination unit determines the current imbalance level by matching the slope and influence range of the offset trajectory with a preset imbalance threshold range.

5. A data center dual-power supply security management system according to claim 1, characterized in that, The phase impedance adaptive calibration module includes a tuning drive unit and a multi-level calibration execution unit. The multi-level calibration execution unit is connected in series or in parallel in the dual power supply circuit and includes multiple sets of adjustable reactance components and phase shifting components. The tuning drive unit receives the calibration command issued by the intelligent master control module, calculates the required impedance compensation amount and phase compensation angle, and controls the multi-level calibration execution unit to switch to the corresponding physical compensation level, so as to realize the dynamic fine adjustment of the line impedance and phase difference in the dual power supply circuit, so that the two power supply parameters tend to be balanced.

6. A data center dual-power supply security management system according to claim 1, characterized in that, The busbar harmonic hierarchical suppression module includes a harmonic feature identification unit and a hierarchical suppression execution unit. The harmonic feature identification unit monitors the voltage and current waveforms of the power supply busbar in real time and extracts low-amplitude continuous harmonic components with frequencies within a preset range. The graded suppression execution unit presets the suppression power level in advance based on the imbalance level determination result, and synchronously adjusts the filter center frequency of the active filter or notch filter when the phase impedance adaptive calibration module is activated, so as to eliminate induced harmonics and residual harmonics caused by parameter fine-tuning.

7. A data center dual-power supply security management system according to claim 1, characterized in that, The real-time calibration effect verification module includes a parameter verification unit and a harmonic detection unit. After the phase impedance adjustment is completed, the parameter verification unit reacquires the real-time phase difference data and impedance data of the dual power supply circuit and calculates the deviation value between them and the preset balance target. The harmonic detection unit is used to detect whether the busbar ripple coefficient is within a safe range after adjustment is completed; If the deviation value exceeds the range or the ripple coefficient exceeds the standard, the parameter verification unit generates a secondary calibration trigger signal and sends it to the intelligent master control module.

8. A data center dual-power supply security management system according to claim 1, characterized in that, The self-learning and iterative module of the control strategy includes a database storage unit and a strategy optimization unit. The database storage unit stores the initial parameters, load prediction data, calibration level, suppression level and verification results of each control process in a time series. The strategy optimization unit uses deep learning algorithms to perform correlation analysis on the stored data, calculates the calibration response speed and harmonic suppression rate under different load fluctuation characteristics, corrects the matching weight between the calibration level and the load characteristics, generates an updated set of control parameters and stores them in the strategy library for the intelligent master control module to call.

9. A data center dual-power supply security management system according to claim 1, characterized in that, The intelligent control module adopts a master-slave communication architecture, establishing a real-time data bus and a control command bus. The intelligent control module sends synchronous working pulses to the phase impedance adaptive calibration module and the bus harmonic hierarchical suppression module through the control command bus, ensuring that the physical calibration action and the harmonic suppression action are triggered synchronously in the time domain, and the synchronization error is controlled within a preset millisecond threshold.

10. A data center dual-power supply security management method, applied to a data center dual-power supply security management system according to any one of claims 1-9, characterized in that, Includes the following steps: Step S1: Real-time monitoring data of the incoming, outgoing and busbar connection terminals of the dual power supply main and backup circuits are acquired synchronously through the phase-separated multi-point parameter acquisition module. At the same time, load change instructions are extracted from the computing power scheduling platform using the high-density load trend prediction module. Step S2: The intelligent central control module transmits the real-time monitoring data and the load change command to the dual-path parameter dynamic coupling analysis module, extracts the coupling correlation characteristics between parameters and load, and combines the load prediction data to infer the imbalance trend and classify the imbalance level; Step S3: The intelligent master control module synchronously schedules the phase impedance adaptive calibration module and the busbar harmonic hierarchical suppression module according to the imbalance level, and performs phase impedance fine-tuning and busbar harmonic elimination actions according to the set timing sequence. Step S4: Use the real-time calibration effect verification module to detect the adjusted loop parameters and busbar harmonic status. If the preset balance target is not achieved, return to step S3 to perform a second calibration. Step S5: The self-learning and iteration module of the control strategy collects data from the entire control process, optimizes the control strategy model by analyzing the calibration effect, and provides the updated strategy model to the intelligent central control module to achieve closed-loop control iteration.

Images