A method and system for calculating carbon emissions using a three-dimensional time-series dynamic coupling method.
By dynamically coupling computing power, electricity, and carbon intensity data in three dimensions over time, the problem of large measurement errors in fixed carbon emission factors is solved, improving the accuracy and real-time performance of carbon emission measurement. The total carbon emissions are also broken down into traceable sub-items to meet the needs of refined carbon management.
Patent Information
- Application Number
- CN202611319370.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-10
AI Technical Summary
The fixed carbon emission factor in existing carbon emission measurement methods cannot reflect the proportion of green electricity in the regional power grid, the fluctuation of cabinet load rate, and the difference between peak and valley periods of the power grid. This results in a large deviation between the measurement results and the actual situation, which cannot meet the needs of refined carbon management. Furthermore, the carbon accounting output is a single total carbon emission amount, which cannot distinguish the specific causes and cannot provide supporting data for each item.
A three-dimensional time-series dynamic coupling carbon emission measurement method based on power grid carbon is adopted. By standardizing and aligning multi-source heterogeneous time-series data, and using UTC global clock correction, unit conversion and deduplication, a standard time axis with equal intervals is constructed for high-frequency downsampling and interpolation. Combined with green electricity correction, load efficiency correction and peak-valley correction terms, a real-time dynamic carbon emission conversion factor is obtained, and the total carbon emission is split into the baseline steady state, load fluctuation additional and grid carbon intensity deviation carbon emission.
It has improved the accuracy and real-time performance of carbon emission measurement, enabling precise differentiation of carbon emission causes. It provides sub-item data support for computing power scheduling optimization, green electricity procurement decisions, and peak-valley staggered production, and meets the traceability and decomposability requirements of carbon audit and dual-carbon verification.
Smart Images

Figure CN123070486A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of carbon accounting technology, and in particular to a three-dimensional time-series dynamic coupling carbon emission measurement method and system. Background Technology
[0002] The operation of a computing cluster involves three independent business platforms: a computing power scheduling system, a data center power monitoring system, and a regional power grid carbon ledger. Computing load and rack power consumption are typically collected at high frequencies, on a second- or minute-by-minute basis, while the regional power grid carbon intensity is updated hourly based on the monthly power grid ledger. Existing carbon accounting methods usually simply combine these three metrics, estimating the computing cluster's carbon emissions by multiplying a fixed carbon emission factor by the total power consumption.
[0003] However, fixed carbon emission factors cannot reflect real-time changes in the proportion of green electricity in the regional power grid, the impact of cabinet load rate fluctuations on the overall energy efficiency of power supply and cooling, and dynamic operating conditions such as the difference in output between thermal power and renewable energy during peak and valley periods of the power grid. This results in a large deviation between carbon emission measurement results and actual conditions, making it difficult to meet the needs of refined carbon management.
[0004] Furthermore, existing carbon accounting outputs only provide a single total carbon emission figure, which cannot distinguish the specific causes of carbon emissions. That is, it cannot clarify which carbon emissions are inherent to the operation of the server rack, which are caused by uneven load scheduling, and which are due to differences in the energy structure of the external power grid. The single total output cannot provide detailed data support for computing power scheduling optimization, green electricity procurement decisions, and peak-valley staggered production, nor does it meet the hard requirements of traceability and decomposability in carbon audits and dual carbon verification. Summary of the Invention
[0005] The purpose of this invention is to address the technical problems in existing carbon emission measurement methods, such as large measurement errors caused by fixed carbon emission factors and the inability to distinguish carbon emission causes by only outputting the total carbon emissions. The invention provides a three-dimensional time-series dynamic coupling carbon emission measurement method and system.
[0006] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:
[0007] A computational carbon three-dimensional time-series dynamic coupling carbon emission measurement method includes the following sub-steps:
[0008] The original computing load and original real-time power consumption of each rack are sampled, and the original regional power grid carbon intensity is sampled to obtain the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence. The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are then corrected by the UTC global clock, converted by unit, and deduplicated to obtain a three-dimensional time series dataset.
[0009] A standard time axis with equal intervals is constructed using a three-dimensional time series dataset. The computing load time series of the three-dimensional time series dataset is downsampled at high frequency using the standard time axis with equal intervals to obtain the average load time series. The real-time power consumption time series and the regional power grid carbon intensity time series are interpolated using the standard time axis with equal intervals to obtain the cabinet power consumption and the granular regional dynamic carbon intensity.
[0010] Green electricity correction terms are obtained based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient. Load efficiency correction terms are obtained using the average load time sequence. Peak and valley time correction terms are set according to the peak and valley time of the power grid. Real-time dynamic carbon emission conversion factor is obtained using the national standard basic carbon emission coefficient, green electricity correction terms, load efficiency correction terms and peak and valley time correction terms.
[0011] Based on rack power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item and peak and valley period correction item, the total carbon emissions, benchmark steady-state carbon emissions, load fluctuation additional carbon emissions and grid carbon intensity deviation carbon emissions are obtained.
[0012] Threshold boundary verification, timing smoothing verification, and coupling logic verification are performed, and the verified total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions are stored.
[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: By standardizing and preprocessing multi-source heterogeneous time-series data and aligning the time series, cross-platform fusion and unified granular coupling measurement of three-dimensional data of computing power, electricity, and carbon intensity are achieved; through four-level progressive dynamic correction, the impact of changes in the proportion of green electricity in the regional power grid, fluctuations in cabinet load rate, and differences in power output during peak and valley periods on carbon emissions is comprehensively reflected, overcoming the defect of large measurement error of fixed carbon emission factors and improving the accuracy and real-time performance of carbon emission measurement; by decomposing the total carbon emissions into three traceable components—baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and power grid carbon intensity deviation carbon emissions—the precise differentiation and responsibility definition of carbon emission causes are achieved, providing component data support for computing power scheduling optimization, green electricity procurement decisions, and peak-valley staggered production; through a three-layer progressive verification mechanism, abnormal data is filtered from three dimensions—numerical boundaries, time series statistical distribution, and physical coupling logic—ensuring the integrity and reliability of carbon ledger data and meeting the traceability and decomposability requirements of carbon auditing and dual-carbon verification.
[0014] Furthermore, a three-dimensional time-series dynamic coupling carbon emission measurement method for computing power includes sampling the original computing load and original real-time power consumption of each rack, sampling the original regional power grid carbon intensity, obtaining the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence, and then correcting the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence to UTC global clock, performing unit conversion and deduplication to obtain a three-dimensional time-series dataset, including the following sub-steps:
[0015] The original computing load of each rack in the computing power scheduling system and the original real-time power consumption of each rack in the computer room power monitoring system are sampled at high frequency to obtain the original computing load sequence and the original real-time power consumption sequence. The original regional carbon intensity of the regional power grid carbon ledger system is sampled at low frequency to obtain the original regional power grid carbon intensity sequence.
[0016] The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are subjected to UTC global clock correction and unified unit conversion to obtain standardized computing load timing, real-time power consumption timing, and regional power grid carbon intensity timing.
[0017] The computing load time series, real-time power consumption time series, and regional power grid carbon intensity time series are subjected to MD5 hash deduplication and field normalization to obtain a three-dimensional time series dataset.
[0018] In the above solution, the raw data from three independent business platforms—the computing power scheduling system, the data center power monitoring system, and the regional power grid carbon ledger system—are synchronously read and classified for sampling. This constructs computing load sequences, real-time power consumption sequences, and regional power grid carbon intensity sequences, respectively. This solves the problem of multi-source heterogeneous data being unable to be directly mixed for calculation due to inconsistencies in storage format, sampling period, local clock, and measurement units. UTC global clock correction eliminates random offset errors in the local clocks of each business server. The conversion between watts and kilowatts unifies the physical dimensions of power consumption and carbon intensity, avoiding amplification of measurement errors due to inconsistent unit dimensions. MD5 hashing removes duplicate records generated by network retries and repeated database writes in high-concurrency acquisition scenarios, preventing duplicate carbon emission calculations and resulting in a clean and usable three-dimensional time-series dataset.
[0019] Furthermore, a three-dimensional time-series dynamic coupling carbon emission measurement method for computing power involves performing UTC global clock correction and unified unit conversion on the original computing power load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence to obtain standardized computing power load time series, real-time power consumption time series, and regional power grid carbon intensity time series. The formula is as follows:
[0020] ;
[0021] in, This represents the original computing power load sequence at the k-th high-frequency sampling time, where k is the sequence number of the high-frequency sampling time point. This is the original real-time power consumption sequence at the k-th high-frequency sampling time. This represents the original regional power grid carbon intensity sequence at the m-th low-frequency sampling time, where m is the sequence number of the low-frequency carbon intensity sampling time point. This is a clock skew correction function. This is the original local timestamp corresponding to the k-th high-frequency sampling time. This represents the deviation between the local clock and the UTC global clock. Let be the computing load time series at time t, where t is a standardized time. The real-time power consumption timing is at time t. For the regional power grid carbon intensity time series at time t, This is the original local timestamp corresponding to the m-th low-frequency sampling time. For UTC global unified time, This is a time-series mapping function for low-frequency carbon intensity.
[0022] In the above scheme, the random offset error of the local clock of each service server is eliminated by the UTC global clock correction function, so that the three types of time series data of computing power, power consumption and carbon intensity are unified to the same standard time reference; the physical dimensions of power consumption and carbon intensity are unified by the unit conversion of watts and kilowatts, so as to avoid the amplification of measurement error by the inconsistency of unit dimension; and the hourly regional power grid carbon intensity sampling time is mapped to the unified UTC time axis by the low-frequency carbon intensity time series mapping function, so as to achieve the accurate correspondence of the three sets of time series in the time dimension, and provide an accurate time alignment basis for the standardized three-dimensional time series dataset.
[0023] Furthermore, a three-dimensional time-series dynamic coupling carbon emission measurement method is provided, wherein the three-dimensional time-series dataset is defined by the following formula:
[0024] ;
[0025] in, This is a three-dimensional time series dataset at time t. The computing load time series at time t. The real-time power consumption timing is at time t. For the regional power grid carbon intensity time series at time t, For UTC global unified time, This is a unique code for the server rack.
[0026] In the above scheme, the computing load time series, real-time power consumption time series and regional power grid carbon intensity time series of each rack are grouped and associated by the unique rack code, and bound to the global unified UTC time. This makes each record in the three-dimensional time series dataset traceable to the specific rack and standard time, providing a clear and traceable basic data for subsequent time series alignment and carbon emission calculation.
[0027] Furthermore, a method for dynamically coupled three-dimensional time-series carbon emission measurement using computing power includes the following sub-steps: constructing an equally spaced standard time axis using a three-dimensional time-series dataset; performing high-frequency downsampling on the computing power load time series of the three-dimensional time-series dataset using the equally spaced standard time axis to obtain the average load time series; and interpolating the real-time power consumption time series and the regional power grid carbon intensity time series using the equally spaced standard time axis to obtain the cabinet power consumption and granular regional dynamic carbon intensity.
[0028] A computing power window is configured for the computing power load time series in the three-dimensional time series dataset, an electricity window is configured for the real-time power consumption time series in the three-dimensional time series dataset, and a carbon intensity window is configured for the regional power grid carbon intensity time series in the three-dimensional time series dataset. The electricity window is used as a global alignment benchmark to construct an equally spaced standard time axis.
[0029] The computing load time series in the three-dimensional time series dataset is divided into windows using an equally spaced standard time axis. High-frequency downsampling is performed in each 5-minute window to obtain the average load time series.
[0030] By traversing any point on the equally spaced standard time axis, interpolation is performed on the real-time power consumption time series and regional power grid carbon intensity time series in the three-dimensional time series dataset to obtain the cabinet power consumption and granular regional dynamic carbon intensity.
[0031] In the above scheme, a high-frequency acquisition window is configured for computing load time series, a global alignment benchmark window is configured for real-time power consumption time series, and a low-frequency matching window is configured for regional power grid carbon intensity time series. The power window is used as a benchmark to construct an equally spaced standard time axis, so that the time series data of computing power, power consumption and carbon intensity with different sampling periods can be uniformly scheduled and processed on the same time axis. By using high-frequency downsampling, the computing load is compressed from second-level data to minute-level window representative values, reducing the data dimensionality while retaining the load characteristics within the window. Piecewise linear interpolation is used to complete real-time power consumption and regional power grid carbon intensity to each moment of the equally spaced standard time axis, eliminating the time series misalignment caused by the difference in sampling period, so that the three types of data are accurately aligned in time.
[0032] Furthermore, a three-dimensional time-series dynamic coupled carbon emission measurement method for electricity generation includes the following sub-steps: obtaining a green electricity correction term based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient; obtaining a load efficiency correction term using the average load time series; setting a peak-valley time correction term according to the peak-valley time period of the power grid; and obtaining a real-time dynamic carbon emission conversion factor using the national standard basic carbon emission coefficient, green electricity correction term, load efficiency correction term, and peak-valley time correction term.
[0033] Based on the national standard basic carbon emission coefficient and the dynamic carbon intensity of the particle size region, green electricity correction terms are obtained;
[0034] Set an efficient load benchmark and adjustment coefficient, and obtain the load efficiency correction term using the efficient load benchmark, adjustment coefficient, and average load timing.
[0035] Set peak-valley time correction items according to the peak-valley time of the power grid;
[0036] By utilizing the national standard basic carbon emission coefficient, green electricity correction item, load efficiency correction item, and peak-valley time correction item, a real-time dynamic carbon emission conversion factor is obtained.
[0037] In the above scheme, the green electricity ratio is linearly deducted from the national standard basic carbon emission coefficient by the dynamic carbon intensity of the granular region, so that the carbon emission measurement reflects the change of the green electricity ratio of the regional power grid in real time. By setting a high-efficiency load benchmark and adjustment coefficient, the carbon conversion factor is dynamically adjusted according to the degree of deviation of the average load from the benchmark, reflecting the impact of load fluctuations on the comprehensive energy efficiency of power supply and cooling. By segmenting the peak and valley periods of the power grid, the impact of the output difference between thermal power and renewable energy at different times on carbon emissions is reflected. The real-time dynamic carbon emission conversion factor is obtained by linearly superimposing the four corrections, which overcomes the defect that the fixed carbon emission factor cannot reflect multiple dynamic operating conditions and improves the accuracy and real-time performance of carbon emission measurement.
[0038] Furthermore, a three-dimensional time-series dynamic coupling carbon emission measurement method based on power grid carbon consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction term, green electricity correction term, and peak-valley time period correction term, to obtain total carbon emissions, baseline steady-state carbon emissions, load fluctuation additional carbon emissions, and grid carbon intensity deviation carbon emissions includes the following sub-steps:
[0039] The total carbon emissions are obtained based on the power consumption of the server rack and the real-time dynamic carbon emission conversion factor.
[0040] The total carbon emissions are obtained based on the power consumption of the server rack and the real-time dynamic carbon emission conversion factor.
[0041] Based on rack power consumption and load efficiency correction, the additional carbon emissions from load fluctuations are obtained.
[0042] Based on the power consumption of the server rack, the green electricity correction term, and the peak-valley time period correction term, the carbon emission of the grid carbon intensity deviation is obtained.
[0043] In the above scheme, the total carbon emissions are divided into three components: baseline steady-state carbon emissions, additional carbon emissions due to load fluctuations, and carbon emissions due to grid carbon intensity deviations. These correspond to the inherent basic carbon emissions of the server rack, the additional carbon emissions caused by load deviations from the high-efficiency baseline, and the carbon emission deviations caused by differences in the external grid energy structure, respectively. This allows for the quantification and traceability of the composition of carbon emissions item by item. The linear addition of the three components strictly equals the total carbon emissions. This approach achieves accurate differentiation of carbon emission causes while ensuring data self-consistency, providing directly corresponding sub-item carbon emission data support for balanced computing power scheduling, green electricity procurement, and peak-valley staggered production.
[0044] Furthermore, a three-dimensional time-series dynamic coupled carbon emission measurement method for power grid carbon includes the following sub-steps: threshold boundary verification, time-series smoothing verification, and coupled logic verification. The method stores the verified total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions.
[0045] Threshold boundary checks were performed on average load timing, rack power consumption, and real-time dynamic carbon emission conversion factor.
[0046] If the boundary is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate.
[0047] If the boundary is not exceeded, continue with the next step;
[0048] The total carbon emissions were averaged over 20 consecutive 5-minute windows. Time-series smoothing verification was then performed based on the average total carbon emissions and the total carbon emissions.
[0049] If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window are replaced by the mean interpolation of the adjacent windows, and the steps continue;
[0050] If the range is not exceeded, continue with the next step;
[0051] Based on the average load timing and the gradient of rack power consumption changes, perform coupled logic verification:
[0052] If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate.
[0053] If the range is not exceeded, continue with the next step;
[0054] The total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions, which have undergone threshold boundary verification, timing smoothing verification, and coupling logic verification, are grouped and stored according to cabinet number and metering period.
[0055] In the above scheme, threshold boundary verification is used to determine the physical range of average load, rack power consumption, and dynamic carbon emission conversion factor, eliminating abnormal windows that exceed the normalization boundary, overload redundancy boundary, and reasonable range of carbon factor to prevent source data distortion. Time-series smoothing verification captures extreme values caused by instantaneous network fluctuations or sensor failures, and uses the average of adjacent windows to synchronously replace the total carbon emission and its three sub-items in the current window, eliminating noise while maintaining the identity relationship between the sub-items. Coupled logic verification determines whether the load and power consumption change trends meet the physical positive correlation constraint, eliminating distorted data that violates the physical laws of hardware. The three-layer verification filters anomalies step by step from three dimensions: numerical boundary, statistical distribution, and physical laws, ensuring the integrity and reliability of the carbon emission data in the database.
[0056] Furthermore, a three-dimensional time-series dynamic coupling carbon emission measurement method is provided, wherein threshold boundary verification is performed on the average load time series, cabinet power consumption, and real-time dynamic carbon emission conversion factor, and the formula is:
[0057] ;
[0058] in, The 5-minute average load time series is at time t. The power consumption of the rack in 5 minutes at time t. The real-time dynamic carbon emission reduction factor at time t is denoted as . The rated power consumption of the server rack is 5 minutes.
[0059] In the above scheme, the average load, cabinet power consumption and dynamic carbon emission conversion factor are determined by threshold boundary verification: the load is normalized to the range of 0 to 1, the power consumption limit is set to 1.2 times the rated overload redundancy, and the dynamic carbon factor is limited to a reasonable physical range of 0.2 to 0.8. Any data exceeding any boundary is judged as abnormal dirty data and the subsequent calculation of the window is terminated. This eliminates obviously impossible values caused by sensor failure, abnormal data collection and other reasons from the source, and prevents extreme abnormal data from entering the carbon emission measurement and traceability process.
[0060] A three-dimensional time-series dynamic coupling carbon emission metering system based on carbon emission calculation includes a data acquisition module, a carbon emission calculation data unification module, a correction item data acquisition module, a carbon emission differentiation module, and a carbon emission verification and storage module.
[0061] The data acquisition module samples the original computing load and original real-time power consumption of each rack, and samples the original regional power grid carbon intensity to obtain the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence. The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are then corrected by the UTC global clock, converted by unit, and deduplicated to obtain a three-dimensional time series dataset.
[0062] The unified carbon data module for computing power uses a three-dimensional time series dataset to construct an equally spaced standard time axis. The computing power load time series of the three-dimensional time series dataset is downsampled at high frequency through the equally spaced standard time axis to obtain the average load time series. The real-time power consumption time series and the regional power grid carbon intensity time series are interpolated through the equally spaced standard time axis to obtain the cabinet power consumption and the granular regional dynamic carbon intensity.
[0063] The correction data acquisition module obtains green electricity correction items based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient, obtains load efficiency correction items using the average load time sequence, sets peak and valley time correction items according to the peak and valley time periods of the power grid, and obtains real-time dynamic carbon emission conversion factors using the national standard basic carbon emission coefficient, green electricity correction items, load efficiency correction items and peak and valley time correction items.
[0064] The carbon emission differentiation module obtains total carbon emissions, benchmark steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions based on cabinet power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item, and peak-valley time period correction item.
[0065] The carbon emission verification and storage module performs threshold boundary verification, timing smoothing verification, and coupling logic verification, and stores the verified total carbon emissions, benchmark steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions. Attached Figure Description
[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0067] Figure 1 This is a flowchart of a three-dimensional time-series dynamic coupling carbon emission measurement method based on electrical calculation. Detailed Implementation
[0068] 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0069] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance, or suggesting any such actual relationship or order between these entities or operations. Additionally, the terms "connected," "linked," etc., can refer to a direct connection between elements or an indirect connection via other elements.
[0070] This invention is achieved through the following technical solutions, such as... Figure 1 As shown, a three-dimensional time-series dynamic coupling carbon emission measurement method includes the following sub-steps:
[0071] S1: Sample the raw computing load and raw real-time power consumption of each rack, and sample the raw regional power grid carbon intensity to obtain the raw computing load sequence, raw real-time power consumption sequence and raw regional power grid carbon intensity sequence. Then, correct the raw computing load sequence, raw real-time power consumption sequence and raw regional power grid carbon intensity sequence through UTC global clock, unit conversion and deduplication to obtain a three-dimensional time series dataset.
[0072] Sampling, clock correction, and unified unit conversion are performed to obtain standardized computing load time series, real-time power consumption time series, and regional power grid carbon intensity time series. By deduplicating the computing load time series, real-time power consumption time series, and regional power grid carbon intensity time series using unified identifiers, a three-dimensional time series dataset is obtained.
[0073] It should be noted that the computing power scheduling system, the data center power monitoring system, and the regional power grid carbon ledger belong to different business platforms. Their data storage formats, sampling periods, local clocks, and units of measurement are completely inconsistent. Directly mixing them for calculation will cause problems such as timing misalignment, unit mismatch, and duplicate dirty data. Therefore, they are extracted separately.
[0074] S11: High-frequency sampling is performed on the original computing load of each rack in the computing power scheduling system and the original real-time power consumption of each rack in the data center power monitoring system to obtain the original computing load sequence and the original real-time power consumption sequence. Low-frequency sampling is performed on the original regional carbon intensity of the regional power grid carbon ledger system to obtain the original regional power grid carbon intensity sequence. The formula is:
[0075] ;
[0076] in, This represents the original computing load sequence at the k-th high-frequency sampling time, where k is the high-frequency sampling time point number (sampling interval is 5 seconds). Let be the original computing power load value at the 'a'-th high-frequency sampling time, where 'a' is the total number of high-frequency sampling points per day. This is the original real-time power consumption sequence at the k-th high-frequency sampling time. Let be the original real-time power consumption value at the a-th high-frequency sampling time. This is the original regional power grid carbon intensity sequence at the m-th low-frequency sampling time, where m is the sequence number of the low-frequency carbon intensity sampling time (sampling interval 1h). Let be the original regional carbon intensity value at the b-th low-frequency sampling time, where b is the total number of low-frequency sampling points for carbon intensity per day.
[0077] In the embodiments, , .
[0078] It should be noted that the local clocks of various business servers in the data center have random offsets. In industrial energy monitoring scenarios, there are generally minute-level synchronization errors. The original unit of power consumption is watts, while the common unit of carbon intensity measurement is tons of carbon dioxide equivalent / kilowatt-hour. Inconsistent unit dimensions will directly amplify the measurement error. Therefore, the UTC global clock is used as the global reference to correct the clock offset error and synchronously complete the conversion of physical quantity units.
[0079] S12: Perform UTC global clock correction and unified unit conversion on the original computing load sequence, original real-time power consumption sequence, and original regional power grid carbon intensity sequence to obtain standardized computing load timing, real-time power consumption timing, and regional power grid carbon intensity timing. The formula is as follows:
[0080] ;
[0081] in, This is a clock skew correction function. This is the original local timestamp corresponding to the k-th high-frequency sampling time. This represents the deviation between the local clock and the UTC global clock. Let be the computing load time series at time t, where t is a standardized time. The real-time power consumption timing is at time t. For the regional power grid carbon intensity time series at time t, This is the original local timestamp corresponding to the m-th low-frequency sampling time. For UTC global unified time, This is a time-series mapping function for low-frequency carbon intensity.
[0082] In the embodiment, a single record If so, delete the corresponding computing load timing and real-time power consumption timing for that record. If so, delete the corresponding regional power grid carbon intensity time series.
[0083] S13: Perform MD5 hash deduplication and field normalization on the computing load time series, real-time power consumption time series, and regional power grid carbon intensity time series to obtain a three-dimensional time series dataset. The formula is as follows:
[0084] ;
[0085] in, This is a three-dimensional time series dataset at time t. This is a unique code for the server rack.
[0086] It is important to note that by grouping records using the unique rack code and generating a unique identifier by combining time, rack, and indicator group using MD5 hash, duplicate records are eliminated.
[0087] In this embodiment, a 5-minute window for one server rack is selected and used to obtain S1. , , .
[0088] S2: Construct an equally spaced standard time axis using a 3D time series dataset. Perform high-frequency downsampling on the computing load time series of the 3D time series dataset using the equally spaced standard time axis to obtain the average load time series. Interpolate the real-time power consumption time series and the regional power grid carbon intensity time series using the equally spaced standard time axis to obtain the cabinet power consumption and granular regional dynamic carbon intensity.
[0089] S21: Configure computing power windows for computing power load time series in the 3D time series dataset, power windows for real-time power consumption time series in the 3D time series dataset, and carbon windows for regional power grid carbon intensity time series in the 3D time series dataset. Using the power windows as a global alignment reference, construct an equally spaced standard time axis, as shown in the formula:
[0090] ;
[0091] in, A standard timeline with 5-minute intervals. The UTC time at which the metering begins. The UTC time at which measurement ends.
[0092] In the embodiment, computing power window Electricity window Carbon strength window The computing power window is a high-frequency acquisition window to capture instantaneous fluctuations, the power window is a global alignment benchmark, and the carbon intensity window is a low-frequency matching window, consistent with the release cycle of the regional power grid carbon ledger system.
[0093] S22: Divide the computing load time series in the 3D time series dataset into windows using an equally spaced standard time axis, and perform high-frequency downsampling for each 5-minute window to obtain the average load time series. The formula is as follows:
[0094] ;
[0095] in, Let N be the 5-minute average load time series at time t, and N be the number of sampling points contained in a single 5-minute window. For each high-frequency sampling time within a 5-minute window, For a moment The computing load timing.
[0096] In the embodiments, 0 represents the rack being unloaded, and 1 represents the rack being fully loaded. .
[0097] S23: Traverse any point on the equally spaced standard time axis, interpolate the real-time power consumption time series and regional power grid carbon intensity time series in the three-dimensional time series dataset to obtain the cabinet power consumption and granular regional dynamic carbon intensity. The formula is:
[0098] ;
[0099] in, The power consumption of the rack in 5 minutes at time t. The dynamic carbon intensity of the particle size region at time t is 5 minutes. Standard time axis with 5-minute intervals Index at any time, For any time The interpolated power consumption value, For any time The interpolated carbon strength value.
[0100] Specifically, it detects any point on a standard time axis with equal intervals. If the real-time power consumption value at any given time is missing, then the interpolated power consumption value at any given time is obtained by interpolating the real-time power consumption values before and after that time. The formula is as follows:
[0101] ;
[0102] in, This represents the real-time power consumption value at the left end of the missing interval. This represents the real-time power consumption value at the right end of the missing interval. This represents the UTC time corresponding to the real-time power consumption value at the left end of the missing interval. This represents the UTC time corresponding to the real-time power consumption value at the right end of the missing interval.
[0103] Specifically, it detects any point on a standard time axis with equal intervals. If the regional carbon intensity value at any given time is missing, then the interpolated carbon intensity value at any given time is obtained by interpolating the regional carbon intensity values before and after that time. The formula is as follows:
[0104] ;
[0105] in, This represents the carbon intensity value of the region at the left end of the missing interval. This represents the carbon intensity value of the region at the right end of the missing interval. The UTC time corresponds to the carbon intensity of the region at the left end of the missing interval. The UTC time corresponds to the carbon intensity of the region at the right end of the missing interval.
[0106] In this embodiment, a 5-minute window for one server rack is selected and used to obtain S2. , , .
[0107] In existing technologies, traditional carbon metering uses a fixed carbon emission factor, which does not take into account regional green electricity, cabinet load efficiency, and differences in peak and valley operating conditions of the power grid, resulting in large metering errors. In contrast, this application sets up a four-level progressive correction logic to obtain a real-time dynamic carbon emission conversion factor, thereby reducing metering errors.
[0108] S3: Based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient, a green electricity correction term is obtained, a load efficiency correction term is obtained using the average load time series, a peak-valley time correction term is set according to the peak-valley time of the power grid, and a real-time dynamic carbon emission conversion factor is obtained using the national standard basic carbon emission coefficient, green electricity correction term, load efficiency correction term and peak-valley time correction term.
[0109] S31: Based on the national standard basic carbon emission coefficient and the dynamic carbon intensity of the particle size region, the green electricity correction term is obtained, and the formula is:
[0110] ;
[0111] in, For green electricity correction items, This is the national standard basic carbon emission coefficient.
[0112] It is important to note that the logic behind the green electricity adjustment is that the higher the proportion of green electricity in a regional power grid, the lower the carbon emissions per unit of electricity.
[0113] In the embodiments, .
[0114] S32: Set the high-efficiency load reference and adjustment coefficient. Using the high-efficiency load reference, adjustment coefficient, and average load timing, obtain the load efficiency correction term. The formula is:
[0115] ;
[0116] in, For load efficiency correction, For adjustment coefficients, For high-efficiency load benchmarks.
[0117] In the embodiments, The load-power consumption of the intelligent computing center racks was determined through multiple sets of load-power fitting experiments. The load on the computing rack directly affects the overall energy efficiency of power supply and cooling. Long-term real-world test data from the data center shows that the rack has the best overall energy consumption when the load range is 60%. Excessive or insufficient load will result in additional energy consumption and loss, and carbon emissions will increase accordingly.
[0118] It is important to note that when The server rack operates efficiently. It is a negative value; The server rack is underutilized. It is a positive value.
[0119] S33: Set the peak-valley time correction term according to the peak-valley time period of the power grid. The formula is:
[0120] ;
[0121] in, This is a correction term for peak and valley periods.
[0122] It is important to note that the logic behind the peak-valley correction is as follows: During peak periods, thermal power output accounts for a higher proportion of total output, resulting in a generally higher carbon intensity. During off-peak periods, wind, solar, and hydropower output is sufficient, leading to a lower carbon intensity. Statistics on the annual carbon factor difference across provincial power grids show that the average difference in carbon intensity during peak and off-peak periods is approximately... .
[0123] S34: Using the national standard basic carbon emission coefficient, green electricity correction term, load efficiency correction term, and peak-valley time period correction term, the real-time dynamic carbon emission conversion factor is obtained. The formula is as follows:
[0124] ;
[0125] in, is the real-time dynamic carbon emission reduction factor at time t.
[0126] In this example, a 5-minute window for one server rack is selected as the case. , , If the current time period is a normal period, then , .
[0127] In existing technologies, carbon accounting only outputs the total carbon emissions and cannot distinguish the causes of carbon emissions. This application splits carbon emissions into three categories with clear physical meanings. The sum of these components is strictly equal to the total carbon emissions, thus distinguishing which carbon emissions are caused by cabinets, dispatching, and power grids.
[0128] S4: Based on cabinet power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item and peak and valley period correction item, obtain total carbon emissions, benchmark steady-state carbon emissions, load fluctuation additional carbon emissions and grid carbon intensity deviation carbon emissions.
[0129] S41: The total carbon emissions are obtained based on the rack power consumption and real-time dynamic carbon emission conversion factor, using the following formula:
[0130] ;
[0131] in, The total carbon emissions for a 5-minute window at time t (in units of...) ).
[0132] S42: Based on the cabinet power consumption and the national standard basic carbon emission coefficient, the baseline steady-state carbon emission is obtained using the following formula:
[0133] ;
[0134] in, This represents the baseline steady-state carbon emissions for the current window.
[0135] It is important to note that the baseline steady-state carbon emissions represent the basic carbon emissions of the cabinet under ideal operating conditions of 60% high-efficiency load, no green electricity, and no peak-valley differences.
[0136] S43: Based on rack power consumption and load efficiency correction items, the additional carbon emissions from load fluctuations are obtained, using the following formula:
[0137] ;
[0138] in, Add carbon emissions to the load fluctuation at time t.
[0139] It is important to note that the additional carbon emissions from load fluctuations are generated when the rack load deviates from the 60% high-efficiency benchmark, and the additional carbon emission losses caused by no-load / overload can be quantified.
[0140] S44: Based on the power consumption of the server rack, the green electricity correction term, and the peak-valley time correction term, the carbon emissions of the grid carbon intensity deviation are obtained, using the following formula:
[0141] ;
[0142] in, This represents the carbon intensity deviation of the power grid and its carbon emissions.
[0143] It should be noted that the carbon intensity deviation of the power grid is caused by the energy structure of the external power grid and the difference in time-of-use output. It is used to guide enterprises to purchase green electricity and reduce carbon emissions by staggering peak dispatching.
[0144] It should be noted that, in fact Due to the limitations of floating-point precision, minor deviations are allowed in the engineering implementation for carbon audit data consistency verification.
[0145] In this example, a 5-minute window for one server rack is selected as the case. , , , , .
[0146] S5: Perform threshold boundary verification, timing smoothing verification, and coupling logic verification, and store the verified total carbon emissions, benchmark steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions.
[0147] S51: Perform threshold boundary checks on average load timing, rack power consumption, and real-time dynamic carbon emission conversion factor. The formula is:
[0148] ;
[0149] in, The cabinet is rated to consume 5 minutes of power.
[0150] If the boundary is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate.
[0151] If the boundary is not exceeded, continue with the next step.
[0152] It should be noted that, based on the annual carbon factor values released by the power grids of various provinces across the country, the reasonable range for the dynamic carbon coefficient is 0.2 to 0.8; the power distribution design of computer rooms generally reserves 20% overload redundancy, and the upper limit of power consumption is set at 1.2 times the rated power consumption; after normalization, the server load rate is naturally in the range of 0 to 1, and any value exceeding any range is directly judged as abnormal dirty data.
[0153] S52: The total carbon emissions are averaged over 20 consecutive 5-minute windows. A time-series smoothing verification is then performed based on the average total carbon emissions and the total carbon emissions. The formula is as follows:
[0154] ;
[0155] in, This represents the average total carbon emissions over 20 five-minute windows. The standard deviation of total carbon emissions over 20 five-minute windows.
[0156] If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window are replaced by the mean interpolation of the adjacent windows, and the steps continue;
[0157] If the range is not exceeded, continue with the steps.
[0158] It should be noted that transient network fluctuations and sensor malfunctions can cause extreme value jumps. Therefore, a time-series smoothing check is used. The criterion is that, under a normal distribution, 99.7% of the valid data fall within the mean ± 3. Within the range, data outside the range is considered abnormal, but the server rack is operating normally, so the average of the adjacent windows is used to complete the data.
[0159] S53: Based on the average load timing and the gradient of rack power consumption changes, perform coupled logic verification, the formula is:
[0160] ;
[0161] in, The gradient of the average load time series. This represents the gradient of power consumption changes in the server rack.
[0162] If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate.
[0163] If the range is not exceeded, continue with the steps.
[0164] It should be noted that, from the perspective of server hardware physical characteristics, when the computing load increases, the power consumption of CPU and GPU increases simultaneously. The two have a monotonically positive correlation. The product of the two is less than 0, which means that the load and power consumption change in opposite directions, which violates the physical law, so it is excluded.
[0165] S54: The total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions, which have undergone threshold boundary verification, timing smoothing verification, and coupling logic verification, are grouped and stored according to cabinet number and metering period.
[0166] It should be noted that extreme abnormal values may still exist in the entire process of data collection, interpolation, and calculation. Directly entering these values into the database will cause the monthly carbon ledger to be distorted. This invention sets up a three-layer progressive verification mechanism to filter dirty data from three layers: numerical boundaries, time-series statistical distribution, and electricity-carbon coupling logic. After verification, the data is archived and stored in a unified manner to form a complete metering business closed loop.
[0167] In this example, a 5-minute window for one server rack is selected as the case. Threshold boundary verification passed. , , Timing smoothing verification passed. The coupling logic verification passed, so the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions for the 5-minute window were grouped by cabinet number and metering period and written into the database for storage and archiving.
[0168] It should be noted that the specific methods of each module in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.
[0169] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A three-dimensional time-series dynamic coupling carbon emission measurement method, characterized in that, Includes the following sub-steps: The original computing load and original real-time power consumption of each rack are sampled, and the original regional power grid carbon intensity is sampled to obtain the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence. The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are then corrected by the UTC global clock, converted by unit, and deduplicated to obtain a three-dimensional time series dataset. A standard time axis with equal intervals is constructed using a three-dimensional time series dataset. The computing load time series of the three-dimensional time series dataset is downsampled at high frequency using the standard time axis with equal intervals to obtain the average load time series. The real-time power consumption time series and the regional power grid carbon intensity time series are interpolated using the standard time axis with equal intervals to obtain the cabinet power consumption and the granular regional dynamic carbon intensity. Green electricity correction terms are obtained based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient. Load efficiency correction terms are obtained using the average load time sequence. Peak and valley time correction terms are set according to the peak and valley time of the power grid. Real-time dynamic carbon emission conversion factor is obtained using the national standard basic carbon emission coefficient, green electricity correction terms, load efficiency correction terms and peak and valley time correction terms. Based on rack power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item and peak and valley period correction item, the total carbon emissions, benchmark steady-state carbon emissions, load fluctuation additional carbon emissions and grid carbon intensity deviation carbon emissions are obtained. Threshold boundary verification, timing smoothing verification, and coupling logic verification are performed, and the verified total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions are stored.
2. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 1, characterized in that, The process of sampling the raw computing load and raw real-time power consumption of each rack, and sampling the raw regional power grid carbon intensity to obtain the raw computing load sequence, raw real-time power consumption sequence, and raw regional power grid carbon intensity sequence, followed by UTC global clock correction, unit conversion, and deduplication to obtain a three-dimensional time series dataset, includes the following sub-steps: The original computing load of each rack in the computing power scheduling system and the original real-time power consumption of each rack in the computer room power monitoring system are sampled at high frequency to obtain the original computing load sequence and the original real-time power consumption sequence. The original regional carbon intensity of the regional power grid carbon ledger system is sampled at low frequency to obtain the original regional power grid carbon intensity sequence. The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are subjected to UTC global clock correction and unified unit conversion to obtain standardized computing load timing, real-time power consumption timing, and regional power grid carbon intensity timing. The computing load time series, real-time power consumption time series, and regional power grid carbon intensity time series are subjected to MD5 hash deduplication and field normalization to obtain a three-dimensional time series dataset.
3. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 2, characterized in that, The original computing load sequence, original real-time power consumption sequence, and original regional power grid carbon intensity sequence are subjected to UTC global clock correction and unified unit conversion to obtain standardized computing load timing, real-time power consumption timing, and regional power grid carbon intensity timing, as shown in the following formula: ; in, This represents the original computing load sequence at the k-th high-frequency sampling time, where k is the high-frequency sampling time point number. This is the original real-time power consumption sequence at the k-th high-frequency sampling time. This represents the original regional power grid carbon intensity sequence at the m-th low-frequency sampling time, where m is the sequence number of the low-frequency carbon intensity sampling time point. This is a clock skew correction function. This is the original local timestamp corresponding to the k-th high-frequency sampling time. This represents the deviation between the local clock and the UTC global clock. Let be the computing load time series at time t, where t is a standardized time. The real-time power consumption timing is at time t. For the regional power grid carbon intensity time series at time t, This is the original local timestamp corresponding to the m-th low-frequency sampling time. For UTC global unified time, This is a time-series mapping function for low-frequency carbon intensity.
4. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 2, characterized in that, The formula for the three-dimensional time series dataset is: ; in, This is a three-dimensional time series dataset at time t. The computing load time series at time t. The real-time power consumption timing is at time t. For the regional power grid carbon intensity time series at time t, For UTC global unified time, This is a unique code for the server rack.
5. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 1, characterized in that, The process of constructing an equally spaced standard time axis using a three-dimensional time series dataset, performing high-frequency downsampling on the computing load time series of the three-dimensional time series dataset using the equally spaced standard time axis to obtain the average load time series, and interpolating the real-time power consumption time series and the regional power grid carbon intensity time series using the equally spaced standard time axis to obtain the rack power consumption and granular regional dynamic carbon intensity includes the following sub-steps: A computing power window is configured for the computing power load time series in the three-dimensional time series dataset, an electricity window is configured for the real-time power consumption time series in the three-dimensional time series dataset, and a carbon intensity window is configured for the regional power grid carbon intensity time series in the three-dimensional time series dataset. The electricity window is used as a global alignment benchmark to construct an equally spaced standard time axis. The computing load time series in the three-dimensional time series dataset is divided into windows using an equally spaced standard time axis. High-frequency downsampling is performed in each 5-minute window to obtain the average load time series. By traversing any point on the equally spaced standard time axis, interpolation is performed on the real-time power consumption time series and regional power grid carbon intensity time series in the three-dimensional time series dataset to obtain the cabinet power consumption and granular regional dynamic carbon intensity.
6. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 1, characterized in that, The process of obtaining a green energy correction term based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient, obtaining a load efficiency correction term using the average load time series, setting a peak-valley time correction term according to the peak-valley time period of the power grid, and obtaining a real-time dynamic carbon emission conversion factor using the national standard basic carbon emission coefficient, green energy correction term, load efficiency correction term, and peak-valley time period correction term includes the following sub-steps: Based on the national standard basic carbon emission coefficient and the dynamic carbon intensity of the particle size region, green electricity correction terms are obtained; Set an efficient load benchmark and adjustment coefficient, and obtain the load efficiency correction term using the efficient load benchmark, adjustment coefficient, and average load timing. Set peak-valley time correction items according to the peak-valley time of the power grid; By utilizing the national standard basic carbon emission coefficient, green electricity correction item, load efficiency correction item, and peak-valley time correction item, a real-time dynamic carbon emission conversion factor is obtained.
7. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 1, characterized in that, The process of obtaining total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions based on cabinet power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item, and peak-valley time correction item includes the following sub-steps: The total carbon emissions are obtained based on the power consumption of the server rack and the real-time dynamic carbon emission conversion factor. The total carbon emissions are obtained based on the power consumption of the server rack and the real-time dynamic carbon emission conversion factor. Based on rack power consumption and load efficiency correction, the additional carbon emissions from load fluctuations are obtained. Based on the power consumption of the server rack, the green electricity correction term, and the peak-valley time period correction term, the carbon emission of the grid carbon intensity deviation is obtained.
8. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 1, characterized in that, The process of performing threshold boundary verification, timing smoothing verification, and coupling logic verification, and storing the verified total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions, includes the following sub-steps: Threshold boundary checks were performed on average load timing, rack power consumption, and real-time dynamic carbon emission conversion factor. If the boundary is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate. If the boundary is not exceeded, continue with the next step; The total carbon emissions were averaged over 20 consecutive 5-minute windows. Time-series smoothing verification was then performed based on the average total carbon emissions and the total carbon emissions. If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window are replaced by the mean interpolation of the adjacent windows, and the steps continue; If the range is not exceeded, continue with the next step; Based on the average load timing and the gradient of rack power consumption changes, perform coupled logic verification: If the range is exceeded, the total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions of the current window will be removed, and the process will terminate. If the range is not exceeded, continue with the next step; The total carbon emissions, baseline steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions, which have undergone threshold boundary verification, timing smoothing verification, and coupling logic verification, are grouped and stored according to cabinet number and metering period.
9. The three-dimensional time-series dynamic coupling carbon emission measurement method according to claim 7, characterized in that, The threshold boundary verification for average load timing, rack power consumption, and real-time dynamic carbon emission conversion factor is performed using the following formula: ; in, The 5-minute average load time series is at time t. The power consumption of the rack in 5 minutes at time t. The real-time dynamic carbon emission reduction factor at time t is denoted as . The rated power consumption of the server rack is 5 minutes.
10. A computational carbon three-dimensional time-series dynamic coupling carbon emission metering system, used to implement the computational carbon three-dimensional time-series dynamic coupling carbon emission metering method according to any one of claims 1-9, characterized in that, It includes a data acquisition module, a unified carbon data acquisition module, a correction item data acquisition module, a carbon emission differentiation module, and a carbon emission verification and storage module. The data acquisition module samples the original computing load and original real-time power consumption of each rack, and samples the original regional power grid carbon intensity to obtain the original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence. The original computing load sequence, the original real-time power consumption sequence, and the original regional power grid carbon intensity sequence are then corrected by the UTC global clock, converted by unit, and deduplicated to obtain a three-dimensional time series dataset. The unified carbon data module for computing power uses a three-dimensional time series dataset to construct an equally spaced standard time axis. The computing power load time series of the three-dimensional time series dataset is downsampled at high frequency through the equally spaced standard time axis to obtain the average load time series. The real-time power consumption time series and the regional power grid carbon intensity time series are interpolated through the equally spaced standard time axis to obtain the cabinet power consumption and the granular regional dynamic carbon intensity. The correction data acquisition module obtains green electricity correction items based on the dynamic carbon intensity of the granular region and the national standard basic carbon emission coefficient, obtains load efficiency correction items using the average load time sequence, sets peak and valley time correction items according to the peak and valley time periods of the power grid, and obtains real-time dynamic carbon emission conversion factors using the national standard basic carbon emission coefficient, green electricity correction items, load efficiency correction items and peak and valley time correction items. The carbon emission differentiation module obtains total carbon emissions, benchmark steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions based on cabinet power consumption, real-time dynamic carbon emission conversion factor, national standard basic carbon emission coefficient, load efficiency correction item, green electricity correction item, and peak-valley time period correction item. The carbon emission verification and storage module performs threshold boundary verification, timing smoothing verification, and coupling logic verification, and stores the verified total carbon emissions, benchmark steady-state carbon emissions, load fluctuation-related carbon emissions, and grid carbon intensity deviation carbon emissions.
