Flywheel energy storage system-based computing power load power demand data filtering method

Abstract

The invention relates to the technical field of flywheel energy storage, and discloses a calculation load power demand data filtering method based on a flywheel energy storage system, and the method comprises the steps: dividing a load interval according to the fluctuation characteristics of a power load, obtaining a historical cycle power demand record, analyzing the distribution of fluctuation coefficients, and dividing the load interval according to a preset increment; dynamically associating the load intervals with the fluctuation coefficients, calculating the mean value and the offset of the associated fluctuation values, and calibrating the reference fluctuation value of each load interval; configuring an operation environment of the energy storage system, and acquiring data in real time by using a load acquisition module; monitoring the power load in real time, generating a frequency modulation trigger signal when the drop amplitude exceeds a preset threshold value, and detecting the compensation amount to judge whether frequency modulation is completed; and calculating response time according to the reference fluctuation value when frequency modulation is triggered, and triggering an alarm if the frequency modulation is not completed on time. The load acquisition module comprises a high-frequency sampling and data preprocessing unit. The method improves the data processing precision and the energy storage regulation and control efficiency, and guarantees the stability of a power grid and the reliable operation of computing power equipment.

Description

Technical Field

[0001] This invention relates to the field of flywheel energy storage technology, specifically to a method for filtering computing load power demand data based on flywheel energy storage systems. Background Technology

[0002] Against the backdrop of the rapid development of the digital economy, the stability and efficiency of power demand for computing infrastructure, as the core carrier of data processing and computing, has become a key factor restricting the industry's development. During operation, computing equipment experiences frequent and complex power load fluctuations due to dynamic changes in task load. These fluctuations not only impact the stability of the power grid but may also lead to low regulation efficiency of energy storage systems and even affect the normal operation of computing equipment.

[0003] Traditional methods for processing electricity demand data struggle to accurately capture the dynamic characteristics of computing load and cannot perform refined management based on load fluctuations. For example, in load zone division, existing technologies often employ fixed thresholds or simple statistical methods, ignoring the differences and dynamism of load fluctuations under different computing scenarios. This leads to delayed or over-adjusted responses from energy storage systems, resulting in energy waste and equipment wear.

[0004] In terms of fluctuation coefficient analysis, traditional methods typically rely on a single indicator or a simple average of historical data, failing to fully consider the real-time and multi-dimensional characteristics of load fluctuations. This results in inaccurate determination of the baseline fluctuation value, making it difficult to adapt to complex and ever-changing computing load demands. Furthermore, existing technologies lack effective data filtering and dynamic correlation mechanisms when dealing with the coordinated regulation of load fluctuations and energy storage systems. This makes it impossible to achieve accurate prediction and real-time regulation of computing load power demand, leading to long frequency regulation response times and insufficient regulation accuracy in energy storage systems.

[0005] With the continuous development of flywheel energy storage technology, its application in power system frequency regulation and peak shaving is becoming increasingly widespread. Flywheel energy storage systems have advantages such as fast response speed, high energy density, and long cycle life. However, how to combine them with the power demand characteristics of computing loads to achieve accurate filtering and efficient control of load fluctuations remains a pressing technical challenge. Therefore, there is an urgent need for a method for filtering computing load power demand data based on flywheel energy storage systems to improve the processing accuracy of power demand data from computing equipment, optimize the control strategy of energy storage systems, and ensure the stability of the power grid and the reliable operation of computing equipment. Summary of the Invention

[0006] The purpose of this invention is to provide a method for filtering power demand data based on computing load using a flywheel energy storage system, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for filtering power demand data based on a flywheel energy storage system, the method comprising: Step 1: Divide the load into zones based on the characteristics of power load fluctuations. The specific method for dividing the load into zones is as follows: S1: Obtain several power demand records within the historical period of the corresponding computing power scenario. The power demand records include load fluctuation range and fluctuation coefficient. S2: Extract all fluctuation coefficients and perform distribution analysis on the fluctuation coefficients. If the concentration of the fluctuation coefficients reaches the preset threshold, the benchmark fluctuation value is directly determined; otherwise, the minimum value of all load fluctuation intervals is taken as the lower limit load and the maximum value is taken as the upper limit load. S3: Starting from the lower limit load, each increase The load is divided into a load range until the upper limit load is reached, thus generating several load ranges. This is the preset increment; S4: Dynamically correlate the load range with the fluctuation coefficient. If the overlap ratio between the load range and the fluctuation range exceeds... Then the fluctuation coefficient of the corresponding fluctuation range is assigned to the load range, generating the associated fluctuation value of each load range; This is a preset ratio; S5: Select a load range and obtain all its associated fluctuation values, denoted as... , ;calculate mean Seeking each and The sum of the absolute differences, divided by Get the offset ; when At that time, Use this as the benchmark fluctuation value for the load range; otherwise, eliminate them in descending order of the difference. Recalculate the offset until the condition is met. Determine the final benchmark fluctuation value; Preset tolerance; S6: Repeat step S5 for the remaining load intervals to complete the calibration of the reference fluctuation value for all load intervals.

[0008] Preferably, the following settings need to be completed before executing step one: Configure the operating environment of the energy storage system, including energy storage units and conversion units. The energy storage unit refers to the flywheel energy storage device, and the conversion unit refers to the power frequency regulation control module. The energy storage unit has a built-in load acquisition module for acquiring real-time power demand data from computing devices.

[0009] Preferably, in step S1: Historical cycles are defined as the period preceding the current moment. Time period Preset duration; The load fluctuation range refers to the range from the minimum to the maximum instantaneous load of computing equipment during power frequency regulation. The fluctuation coefficient is obtained by calculating the rate of change of electricity demand data within the frequency regulation cycle, specifically by dividing the load range by the frequency regulation response time.

[0010] Preferably, the specific method for determining the benchmark fluctuation value in step S2 is as follows: Let the volatility coefficient be denoted as , ,calculate mean ; Seek each and The sum of the absolute differences, divided by Get the offset ; when At that time, It is used directly as the benchmark fluctuation value.

[0011] Preferably, the dynamic association method in step S4 is as follows: Traverse all load intervals; if the ratio of the overlap between a load interval and any fluctuation interval to the total range of that fluctuation interval exceeds [a certain threshold], [the following condition is met]. Then the fluctuation coefficient of the fluctuation range is bound to the load range; the ratio of the overlapping part is calculated as the intersection length of the load range and the fluctuation range divided by the total length of the fluctuation range.

[0012] Preferably, after step one is completed, the following steps should be performed: Step 2: Monitor the power load of the computing devices within the energy storage unit in real time and record it as the real-time load. When the real-time load decreases compared to the initial value and the decrease exceeds At that time, a frequency modulation trigger signal is generated; The preset reduction threshold; The load compensation amount of the synchronous detection conversion unit is measured, and a frequency modulation completion signal is generated when the compensation amount equals the real-time load amount.

[0013] Preferably, after step two is completed, the following steps should be performed: Step 3: When the frequency modulation trigger signal is generated, obtain the reference fluctuation value of the current load range; The frequency modulation response time is calculated based on the rated load limit of the computing power equipment, specifically by dividing the difference between the rated load and the real-time load by the reference fluctuation value. If a frequency modulation completion signal is not received within the frequency modulation response time, an energy storage anomaly alarm will be triggered.

[0014] Preferably, the step S6 is to remove The specific method is as follows: Sort all by difference from largest to smallest. Remove the first N fluctuation values ​​in sequence, and recalculate the remaining values ​​after each removal. offset ; If the condition cannot be satisfied after removing N consecutive times... Then the current remaining The mean was marked as a temporary baseline, and the removed values ​​were... Recalculate based on the difference in ascending order, until... N is the preset number of rejections.

[0015] Preferably, in step S3: The value ranges from 5% to 15% of the rated load and is dynamically adjusted according to the type of computing equipment.

[0016] Preferably, the load acquisition module includes: A high-frequency sampling unit is used to acquire power waveform data at a frequency of not less than 1000Hz; The data preprocessing unit is used to denoise and normalize the sampled data.

[0017] Compared with the prior art, the beneficial effects of the present invention are: In terms of load interval division, by acquiring historical power demand records for different computing scenarios, extracting fluctuation coefficients, and performing distribution analysis, load intervals can be dynamically determined based on actual load fluctuation characteristics. This data-driven division method fully considers the differences in load fluctuations under different computing scenarios and avoids the limitations of traditional fixed threshold methods. For example, when the concentration of fluctuation coefficients reaches a preset threshold, a baseline fluctuation value is directly determined; otherwise, the interval is divided based on the extreme values ​​of the load fluctuation intervals, ensuring the scientific and rational nature of the load interval division. By dividing the load intervals by each preset incremental load, fine-grained segmentation of load intervals is achieved, providing a solid foundation for subsequent fluctuation coefficient correlation and baseline fluctuation value calibration.

[0018] In the process of correlation of fluctuation coefficients and calibration of benchmark fluctuation values, load intervals are dynamically correlated with fluctuation coefficients. When the overlap ratio between a load interval and a fluctuation interval exceeds a preset ratio, the fluctuation coefficient of the corresponding fluctuation interval is assigned to that load interval, achieving precise matching between the fluctuation coefficient and the load interval. Simultaneously, by calculating the mean and offset of the correlated fluctuation values ​​and combining them with preset tolerances, fluctuation values ​​are eliminated and adjusted, ensuring the accuracy and stability of the benchmark fluctuation values. This dynamic correlation and iterative optimization method effectively filters abnormal fluctuation data, improves the ability of benchmark fluctuation values ​​to characterize the actual load fluctuation characteristics, and provides a reliable basis for the precise control of energy storage systems.

[0019] In terms of energy storage system regulation, by monitoring the power load of the computing equipment in real time, a frequency regulation trigger signal is generated when the load drop exceeds a preset threshold, and the load compensation of the conversion unit is detected simultaneously, realizing real-time monitoring and precise control of the frequency regulation process of the energy storage system. Based on the baseline fluctuation value of the current load range and the rated load limit of the computing equipment, the frequency regulation response time is calculated, and an abnormal alarm is triggered if the frequency regulation is not completed on time, ensuring the timeliness and reliability of energy storage system regulation. This regulation strategy based on real-time data and baseline fluctuation values ​​can significantly shorten the frequency regulation response time of the energy storage system, improve regulation accuracy, and effectively reduce the impact of grid fluctuations on the computing equipment.

[0020] Furthermore, the load acquisition module employs a high-frequency sampling unit and a data preprocessing unit to acquire power waveform data at a frequency of no less than 1000Hz, and performs noise reduction and normalization processing to ensure the accuracy and reliability of the raw data, providing a high-quality data source for subsequent data processing and analysis. The dynamic adjustment mechanism of preset parameters, such as the preset increment dynamically adjusting according to the type of computing equipment, further improves the adaptability and versatility of the method, enabling it to meet the power demand data filtering and energy storage regulation needs of different types of computing equipment. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working principle of the computing load power demand data filtering method based on a flywheel energy storage system described in this invention. Figure 2 Flowchart for determining the benchmark fluctuation value; Figure 3 A flowchart illustrating the dynamic relationship between load range and fluctuation coefficient; Figure 4 This is a flowchart for frequency modulation triggering and detection completion. Detailed Implementation

[0022] 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.

[0023] Please see Figures 1-4 This invention provides a method for filtering computing load power demand data based on a flywheel energy storage system, specifically including the following steps: Step 1: Divide the load range according to the characteristics of power load fluctuation. S1: Retrieve several power demand records for the corresponding computing power scenario within a historical period. Each record includes the load fluctuation range and fluctuation coefficient. The historical period is defined as the preset duration preceding the current time. The time period, the load fluctuation range is the range from the minimum to the maximum instantaneous load of the computing equipment during the power frequency regulation process. The fluctuation coefficient is obtained by calculating the rate of change of power demand data within the frequency regulation cycle, specifically the load range divided by the frequency regulation response time.

[0024] S2: Extract all fluctuation coefficients and perform distribution analysis. If the concentration of fluctuation coefficients reaches a preset threshold, the baseline fluctuation value is directly determined; otherwise, the minimum value of all load fluctuation ranges is taken as the lower limit load, and the maximum value is taken as the upper limit load.

[0025] S3: Starting from the lower limit load, increment by a preset amount. (The value range is 5%-15% of the rated load, and is dynamically adjusted according to the type of computing equipment) Divide the load into a step size until the upper limit load is reached, and generate several load intervals.

[0026] S4: Traverse all load intervals. If the overlap between a load interval and any fluctuation interval exceeds a preset ratio in the total range of that fluctuation interval... (The overlap ratio is calculated by dividing the intersection length by the total length of the fluctuation interval), then the fluctuation coefficient of the corresponding fluctuation interval is assigned to the load interval to generate the associated fluctuation value of each load interval.

[0027] S5: Select a load range and obtain all its associated fluctuation values, denoted as... ( ),calculate mean Seeking each and The sum of the absolute differences, divided by Get the offset .when (When the preset tolerance is set, the following will be used:) Use this as the benchmark fluctuation value for the load range; otherwise, eliminate them in descending order of the difference. Recalculate the offset until the condition is met. To determine the final benchmark fluctuation value.

[0028] S6: Repeat step S5 for the remaining load intervals until the reference fluctuation values ​​for all load intervals have been calibrated.

[0029] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0030] Example 1: Before executing step one, the operating environment of the energy storage system must be configured, specifically including the setup of the energy storage unit and the conversion unit. The energy storage unit is a flywheel energy storage device, whose core structure includes a high-speed rotating flywheel body, a bearing system supporting the flywheel, and an electric motor / generator coaxially connected to the flywheel. The flywheel body is made of high-strength composite material to withstand the centrifugal force during high-speed rotation; the bearing system can use magnetic levitation bearings or air bearings to reduce friction loss and improve system efficiency; the electric motor / generator has a bidirectional energy conversion function, which can convert electrical energy into kinetic energy of the flywheel for storage when there is excess electricity, and can also convert the kinetic energy of the flywheel into electrical energy output when electricity demand increases.

[0031] The flywheel energy storage device incorporates a load acquisition module, which includes a high-frequency sampling unit and a data preprocessing unit. The high-frequency sampling unit acquires power waveform data at a frequency of at least 1000Hz, using high-precision current transformers and voltage sensors to obtain voltage and current signals from the computing equipment in real time, ensuring the capture of transient changes and high-frequency fluctuations in the power load. The data preprocessing unit performs denoising and normalization on the sampled data. The denoising process uses digital filtering algorithms (such as FIR filtering or wavelet filtering) to remove noise signals caused by electromagnetic interference, equipment start-up and shutdown, etc., retaining valid data that accurately reflects the load characteristics. The normalization process converts the sampled data into a uniform numerical range (such as 0-1), eliminating the influence of different units on subsequent data analysis and improving data consistency and comparability.

[0032] The conversion unit is a power frequency regulation control module, whose main function is to realize power conversion and frequency regulation control between the energy storage unit and the computing equipment. This module includes power electronic conversion devices (such as inverters and converters) and a control algorithm unit. The power electronic conversion devices are responsible for completing the AC / DC conversion and power regulation of electrical energy, ensuring that the power quality output by the energy storage unit meets the power supply requirements of the computing equipment. Based on real-time collected load data and energy storage unit status information, the control algorithm unit dynamically adjusts the charging and discharging power of the flywheel energy storage device through preset control strategies (such as PID control or adaptive control), realizing rapid response and effective smoothing of power load fluctuations of the computing equipment.

[0033] In practical applications, flywheel energy storage devices can achieve millisecond-level response times, enabling rapid tracking of sudden power load changes caused by task switching and computational load variations in computing power equipment. The high-frequency sampling characteristics of the load acquisition module ensure continuous and complete data sequences during rapid load fluctuations, providing an accurate data foundation for subsequent load interval division and baseline fluctuation value calculation. The denoising and normalization steps of the data preprocessing unit effectively improve the quality of the raw data, avoiding load characteristic analysis deviations caused by noise interference, thereby ensuring the accuracy of load interval division and the reliability of baseline fluctuation value calibration. The power frequency regulation control module dynamically balances the power demand of computing power equipment by monitoring load changes in real time and adjusting the operating status of the energy storage unit, ensuring the stability and reliability of the power supply system. The configuration of the entire energy storage system's operating environment, from hardware devices to data processing modules and control units, forms a complete technical chain, providing a solid foundation and guarantee for the implementation of computing load power demand data filtering methods based on flywheel energy storage systems.

[0034] Example 2: Step S1 specifies the definitions of historical period, load fluctuation range, and fluctuation coefficient. The historical period is set to a preset duration preceding the current time. The preset time period can be flexibly adjusted based on the operating characteristics of the computing power equipment, data statistics needs, and the real-time requirements of the application scenario. For example, for high-frequency computing scenarios that require real-time load management optimization (such as real-time AI inference and high-frequency trading calculations), the preset duration can be adjusted accordingly. Set the timeframe to minutes or hours to ensure that the acquired historical data reflects the current load fluctuation trend in a timely manner; for computing scenarios with obvious periodicity (such as batch data processing, nighttime offline training, etc.), the timeframe can be adjusted. Set to daily or weekly to cover the entire operating cycle and capture long-term load patterns.

[0035] The load fluctuation range is defined as the range from the minimum to the maximum instantaneous load of computing equipment during power frequency regulation. During power frequency regulation, the power load of computing equipment exhibits dynamic fluctuation characteristics due to factors such as task scheduling and changes in computing resource allocation. Instantaneous load refers to the real-time load value collected at a specific point in time. By continuously monitoring and recording the extreme values ​​of all instantaneous loads during frequency regulation, the upper and lower limits of the load fluctuation range can be determined. For example, when computing equipment switches from a low-load task to a high-load task, the instantaneous load may change from... Rapidly rise to At this time, the load fluctuation range is This range directly reflects the load change amplitude and dynamic range of the equipment during frequency regulation.

[0036] The fluctuation coefficient is calculated by dividing the load range by the frequency regulation response time, where the load range is the difference between the maximum and minimum load values ​​within the frequency regulation cycle (i.e., ...). The frequency modulation response time is the time required from the moment of frequency modulation triggering until the load stabilizes within the target range. The specific calculation process is as follows: First, a complete frequency modulation cycle is determined, starting from the generation of the frequency modulation trigger signal and ending when load compensation is completed and the fluctuation amplitude is less than a preset threshold (e.g., ±5% of the rated load); then, the load range within this cycle is calculated, and the frequency modulation response time is recorded; finally, the response time is calculated using the formula... The fluctuation coefficient is obtained. This coefficient quantitatively describes the rate of change of electricity demand data within the frequency regulation cycle. The larger the value, the more drastic the load fluctuation, and vice versa.

[0037] In practical applications, different types of computing power devices (such as CPU-intensive, GPU-intensive, and FPGA-accelerated devices) exhibit different load fluctuation patterns and frequency modulation response characteristics due to differences in their computing architecture and power characteristics. For example, GPU-intensive deep learning training tasks may generate significant load spikes during parameter update phases, resulting in large load ranges, while FPGA-accelerated real-time data processing tasks may exhibit smaller load fluctuations due to the stability of the pipeline architecture. Therefore, when calculating the fluctuation coefficient, it is necessary to reasonably divide the frequency modulation period and collect corresponding load data based on the specific device type and task characteristics to ensure that the fluctuation coefficient can accurately reflect the actual load change rate of the device. In addition, the accuracy of the fluctuation coefficient calculation also depends on the sampling frequency and time synchronization accuracy of the load acquisition module. High-frequency sampling units (such as sampling frequencies of not less than 1000Hz) can ensure that key time points of load changes are captured, avoiding the omission of extreme values ​​or distortion of fluctuation trends due to excessively large sampling intervals, thereby ensuring the accuracy and reliability of the fluctuation coefficient calculation. By clearly defining and accurately calculating historical cycles, load fluctuation ranges, and fluctuation coefficients, key data support and parameter basis are provided for subsequent load range division, fluctuation coefficient distribution analysis, and benchmark fluctuation value calibration.

[0038] Example 3: The specific method for determining the benchmark volatility value in step S2 is as follows: All volatility coefficients are denoted as... (in , (representing the total number of volatility coefficients), first calculate mean The calculation formula is:

[0039] In the formula, This is the mean of the volatility coefficients, reflecting the average level of all volatility coefficients; For the first Each fluctuation coefficient, whose physical meaning is the rate of change of electricity demand data within the corresponding frequency regulation cycle; This represents the number of fluctuation coefficients involved in the calculation.

[0040] Next, calculate each with the mean Sum of the absolute differences, then divide by Get the offset The specific calculation process is as follows:

[0041] in, This indicates the average deviation of the volatility coefficient from its mean, and is used to measure the concentration of the volatility coefficient. When the deviation... ( When the preset tolerance is used to determine whether the volatility coefficient is concentrated near the mean, it indicates that the distribution of the volatility coefficient is relatively concentrated. In this case, the mean is set... This value is directly used as a benchmark fluctuation value, which can represent the load fluctuation characteristics in most cases; if If the variance is high, it indicates that the volatility coefficient is highly dispersed and further analysis is needed through subsequent steps (such as load range division and correlation fluctuation value processing) to eliminate the interference of abnormal data and ensure the accuracy of the benchmark volatility value.

[0042] In practice, the fluctuation coefficient needs to be collected based on high-frequency sampled power load data, obtained through precise calculation of load range and frequency regulation response time. For example, for a certain computing device in continuous... Fluctuation coefficient within each frequency modulation cycle First, the mean and offset are calculated using the formulas described above. If the offset meets the preset tolerance requirements, the mean is directly used as the baseline fluctuation value for that scenario. If not, the load fluctuation range needs to be divided into different intervals using a load interval division method. Then, the fluctuation coefficient within each interval is analyzed separately to determine a baseline value that better reflects the actual fluctuation characteristics of each load interval. This logic of first statistically analyzing the overall fluctuation coefficient concentration and then determining the calculation method for the baseline fluctuation value can effectively address the differences in load fluctuation characteristics under different computing power scenarios, ensuring that the calibration of the baseline value conforms to the statistical laws of data and reflects the actual load change characteristics.

[0043] Example 4: The dynamic correlation between load intervals and fluctuation coefficients in step S4 is as follows: First, after completing the load interval division (i.e., generating several intervals with preset increments through step S3), After calculating the load interval (within a step size) and the fluctuation coefficient (obtained through step S1 to acquire the fluctuation coefficient for each fluctuation interval), the two need to be iterated and matched. For each load interval... (in This represents the initial load value for the interval. ), one by one with all fluctuation ranges (i.e., the load fluctuation range of each power frequency regulation process recorded in step S1) is subjected to overlap analysis.

[0044] The calculation method for the overlap ratio is as follows: First, determine the intersection of the load range and the fluctuation range, that is, the overlapping part of their numerical ranges. If and Then the initial value of the intersection is The end value is The intersection length is If the two have no intersection (i.e. or If the intersection length is 0, then the total length of the fluctuation interval is 0. Therefore, the ratio of the overlapping parts is .

[0045] When the ratio of the overlap between a certain load range and a certain fluctuation range exceeds a preset proportion When a fluctuation range is determined to have a strong correlation with a load range, the fluctuation coefficient is then bound to that load range as one of its associated fluctuation values. For example, if the load range is [50kW, 60kW] and a fluctuation range is [55kW, 70kW], the intersection is [55kW, 60kW], the intersection length is 5kW, the total length of the fluctuation range is 15kW, and the overlap ratio is [55kW, 60kW]. ,like If set to 30%, the volatility coefficient for that fluctuation range will be assigned to that load range.

[0046] In practice, a preset ratio is used. The value needs to be set according to the load fluctuation characteristics and data correlation requirements of the computing scenario. It is typically set to 50% or higher to ensure a high correlation between the bound fluctuation coefficient and the load range. For scenarios with frequent load fluctuations and high overlap in fluctuation ranges (such as multi-task concurrent operation in a cloud computing center), adjustments may be necessary. To avoid associating too many invalid fluctuation coefficients, which could affect the efficiency and accuracy of subsequent benchmark fluctuation value calculations, the dynamic association process needs to be automated through algorithms. For example, a computer program could traverse all load and fluctuation intervals, calculate the overlap ratio in real time, and perform conditional checks to ensure the efficiency and accuracy of the association operation. This dynamic association mechanism allows historical data with similar load fluctuation characteristics to be matched with the current load interval, ensuring that the associated fluctuation value for each load interval accurately reflects the potential rate of load change within that interval. This provides more targeted data samples for subsequent benchmark fluctuation value calculations, thereby improving the fitting accuracy and adaptability of the filtering method to actual load fluctuations.

[0047] Example 5: After step one is completed, steps two and three, namely real-time monitoring and frequency modulation control, need to be performed. The implementation method is described in detail below with reference to a specific scenario: Assume a computing center is equipped with a flywheel energy storage system, and the rated load of its computing equipment is... Preset reduction threshold (Right now The current initial load is .

[0048] Step Two: Real-time monitoring and frequency modulation trigger the load acquisition module of the energy storage unit to acquire the power load data of the computing equipment in real time at a frequency of 1000Hz. At a certain moment, due to the completion of some computing tasks, the real-time load changes from... It begins to decline when the load is detected to have dropped to a certain level. At that time, the decrease was Exceeding the preset reduction threshold The system immediately generates a frequency modulation trigger signal. Simultaneously, the conversion unit (power frequency modulation control module) activates the load compensation mechanism, supplementing the computing equipment with electrical energy by adjusting the discharge power of the flywheel energy storage device, and monitoring the compensation amount in real time. When the compensation amount reaches... When the load is equal to the real-time load, it indicates that the load fluctuation has been completely suppressed, and a frequency modulation completion signal is generated.

[0049] Step 3: Frequency Modulation Response Time Calculation and Anomaly Monitoring After the frequency modulation trigger signal is generated, the system first determines the current real-time load. The load range in which it is located. Assume that the load range was already set up in step one according to the preset increment. Divided into , … ,but belong The range. Based on the calibration results from step one, the benchmark fluctuation value for this range is... .

[0050] Next, we calculate the frequency modulation response time: According to the formula, the frequency modulation response time... The system starts timing. If a frequency modulation completion signal is not received within 8 seconds (e.g., due to a flywheel energy storage device malfunction causing the compensation amount to fail to reach the real-time load in time), the energy storage system is determined to be abnormal, triggering an alarm mechanism to prompt maintenance personnel to check the equipment.

[0051] Key parameters and logic descriptions Preset reduction threshold This is used to determine whether a load drop necessitates frequency regulation. Its value must be set comprehensively based on the stability requirements of the computing equipment and the responsiveness of the energy storage system. For example, if the equipment is sensitive to sudden load drops, it can... Set it to a smaller value (e.g., 10%) to trigger compensation earlier; if a certain degree of load fluctuation is allowed, the threshold can be appropriately increased.

[0052] Benchmark fluctuation value The value, derived from the load interval division and fluctuation coefficient correlation calculation in step one, reflects the average fluctuation rate within a specific load interval and is a core parameter for calculating the frequency regulation response time. Different load intervals... The differences may exist due to variations in the associated fluctuation coefficients. For example, high-load areas may exhibit more severe fluctuations because computing equipment approaches its rated power, leading to... Relatively large.

[0053] The physical meaning of frequency regulation response time: This time represents the duration required for the energy storage system to theoretically complete load compensation from triggering frequency regulation under the current benchmark fluctuation value. If the actual compensation time exceeds this value, it indicates abnormal system performance, requiring intervention through an alarm mechanism.

[0054] For scenarios with expanded scope and adaptability, if the computing power equipment is deployed in a distributed manner (such as multi-cluster servers in a cloud computing data center), real-time load can be obtained by aggregating the load data of each cluster. When the load of a certain cluster drops sharply, the system can call the corresponding benchmark fluctuation value according to the load range to which that cluster belongs, achieving fine-grained control by partition. In addition, for scenarios with obvious intermittent load characteristics (such as the alternation between parameter update and data preprocessing stages in deep learning model training), by dynamically tracking the range to which the real-time load belongs and calling the corresponding benchmark fluctuation value, it can be ensured that the frequency regulation response time calculation always matches the current load characteristics, improving the adaptability and control accuracy of the energy storage system.

[0055] The above implementation method achieves dynamic management of load fluctuations of computing power equipment through a complete process of real-time monitoring, threshold judgment, interval matching, parameter calculation and abnormal response. It ensures that the flywheel energy storage system can perform frequency regulation operation in a timely and accurate manner when the load changes suddenly, and ensures the reliability of system operation through logical closed loop.

[0056] 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.

[0057] 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 method for filtering power demand data based on a flywheel energy storage system, characterized in that, include: Step 1: Divide the load into zones based on the characteristics of power load fluctuations. The specific method for dividing the load into zones is as follows: S1: Obtain several power demand records within the historical period of the corresponding computing power scenario. The power demand records include load fluctuation range and fluctuation coefficient. S2: Extract all fluctuation coefficients and perform distribution analysis on the fluctuation coefficients. If the concentration of the fluctuation coefficients reaches the preset threshold, the benchmark fluctuation value is directly determined; otherwise, the minimum value of all load fluctuation intervals is taken as the lower limit load and the maximum value is taken as the upper limit load. S3: Starting from the lower limit load, each increase The load is divided into a load range until the upper limit load is reached, thus generating several load ranges. This is the preset increment; S4: Dynamically correlate the load range with the fluctuation coefficient. If the overlap ratio between the load range and the fluctuation range exceeds... Then the fluctuation coefficient of the corresponding fluctuation range is assigned to the load range, generating the associated fluctuation value of each load range; This is a preset ratio; S5: Select a load range and obtain all its associated fluctuation values, denoted as... , ;calculate mean Seeking each and The sum of the absolute differences, divided by Get the offset ; when At that time, Use this as the benchmark fluctuation value for the load range; otherwise, eliminate them in descending order of the difference. Recalculate the offset until the condition is met. Determine the final benchmark fluctuation value; Preset tolerance; S6: Repeat step S5 for the remaining load intervals to complete the calibration of the reference fluctuation value for all load intervals.

2. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, Before executing step one, the following settings must be completed: Configure the operating environment of the energy storage system, including energy storage units and conversion units. The energy storage unit refers to the flywheel energy storage device, and the conversion unit refers to the power frequency regulation control module. The energy storage unit has a built-in load acquisition module for acquiring real-time power demand data from computing devices.

3. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, In step S1: Historical cycles are defined as the period preceding the current moment. Time period Preset duration; The load fluctuation range refers to the range from the minimum to the maximum instantaneous load of computing equipment during power frequency regulation. The fluctuation coefficient is obtained by calculating the rate of change of electricity demand data within the frequency regulation cycle, specifically by dividing the load range by the frequency regulation response time.

4. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, The specific method for determining the benchmark fluctuation value in step S2 is as follows: Let the volatility coefficient be denoted as , ,calculate mean ; Seek each and The sum of the absolute differences, divided by Get the offset ; when At that time, It is used directly as the benchmark fluctuation value.

5. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, The dynamic association method in step S4 is as follows: Traverse all load intervals; if the ratio of the overlap between a load interval and any fluctuation interval to the total range of that fluctuation interval exceeds [a certain threshold], [the following condition is met]. Then the fluctuation coefficient of the fluctuation range is bound to the load range; the ratio of the overlapping part is calculated as the intersection length of the load range and the fluctuation range divided by the total length of the fluctuation range.

6. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, After completing step one, the following steps must be performed: Step 2: Monitor the power load of the computing devices within the energy storage unit in real time and record it as the real-time load. When the real-time load decreases compared to the initial value and the decrease exceeds At that time, a frequency modulation trigger signal is generated; The preset reduction threshold; The load compensation amount of the synchronous detection conversion unit is measured, and a frequency modulation completion signal is generated when the compensation amount equals the real-time load amount.

7. The method for filtering power demand data based on a flywheel energy storage system according to claim 6, characterized in that, After completing step two, the following steps must be performed: Step 3: When the frequency modulation trigger signal is generated, obtain the reference fluctuation value of the current load range; The frequency modulation response time is calculated based on the rated load limit of the computing power equipment, specifically by dividing the difference between the rated load and the real-time load by the reference fluctuation value. If a frequency modulation completion signal is not received within the frequency modulation response time, an energy storage anomaly alarm will be triggered.

8. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, Remove in step S6 The specific method is as follows: Sort all by difference from largest to smallest. Remove the first N fluctuation values ​​in sequence, and recalculate the remaining values ​​after each removal. offset ; If the condition cannot be satisfied after removing N consecutive times... Then the current remaining The mean was marked as a temporary baseline, and the removed values ​​were... Recalculate based on the difference in ascending order, until... N is the preset number of rejections.

9. The method for filtering power demand data based on a flywheel energy storage system according to claim 1, characterized in that, In step S3: The value ranges from 5% to 15% of the rated load and is dynamically adjusted according to the type of computing equipment.

10. A method for filtering power demand data based on a flywheel energy storage system according to claim 2, characterized in that, The load acquisition module includes: A high-frequency sampling unit is used to acquire power waveform data at a frequency of not less than 1000Hz; The data preprocessing unit is used to denoise and normalize the sampled data.

Images