A method and system for carbon emission monitoring for data centers

Abstract

The application discloses a kind of carbon emission monitoring method and system for data center, comprising: real-time acquisition of the power generation proportion data of different energy types, combine weather information to predict renewable energy power supply trend, calculate each energy type power supply proportion and construct dynamic weight matrix;Using historical and real-time renewable energy data to construct time series prediction model, predict future renewable energy power supply capacity;Calculate the carbon emission factor and priority of different data centers, construct task migration feasibility matrix, according to the dispatching benefit function dynamically migrates calculation task to low carbon emission data center, optimizes task scheduling path, reduces cross-regional data transmission delay and energy consumption, realizes the cross-regional optimization scheduling of computing power;Build a multi-dimensional time series prediction model to predict carbon emission trends and monitor carbon emission anomalies in real time.The application can solve the problem of matching difficulty between computing power load and renewable energy faced by data center.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission technology, and in particular to a carbon emission monitoring method and system for a data center. Background Art

[0002] With the acceleration of global digital transformation, data centers have become crucial infrastructure supporting key technologies such as cloud computing, big data, and artificial intelligence. However, this rapid growth is accompanied by extremely high energy consumption, and reducing their carbon emissions has become a pressing issue.

[0003] Currently, an increasing number of data centers are integrating renewable energy sources, such as wind and photovoltaic power, to reduce their reliance on traditional fossil fuels. However, due to the volatility and uneven geographical distribution of renewable energy generation, there is a significant mismatch between data center load demand and green electricity supply. Most data centers tend to be located in core areas with convenient communications and low network latency, while areas rich in renewable energy resources are often located in remote areas. This creates a conflict between data center load distribution and the geographical distribution of renewable energy. Furthermore, wind and photovoltaic power generation are significantly affected by factors such as weather, seasons, and diurnal variations, resulting in significant fluctuations in their power supply capacity. Data center computing tasks generally require a highly stable and reliable power supply. This temporal mismatch between supply and demand makes it difficult for data centers to fully utilize renewable energy, thereby limiting their potential for carbon reduction.

[0004] Traditional data center energy consumption monitoring methods, primarily based on total power consumption statistics, cannot accurately assess carbon emissions under different energy structures, nor can they dynamically adjust computing power to adapt to changes in renewable energy. Therefore, an innovative carbon emission monitoring method is needed that can not only perceive data center energy consumption and carbon emissions in real time, but also dynamically optimize computing load scheduling strategies based on renewable energy supply forecasts to achieve adaptive matching of computing power demand with renewable energy supply, thereby effectively reducing carbon emissions and improving green energy utilization efficiency. Summary of the Invention

[0005] In order to solve the above problems, the present invention provides a carbon emission monitoring method and system for a data center.

[0006] To achieve the above object, the technical solutions adopted by the present invention are as follows:

[0007] In one aspect, the present invention discloses a method for monitoring carbon emissions in a data center, comprising:

[0008] Step 1: Collect power generation data of different energy types in real time, calculate the power supply proportion of each energy type, and use this proportion as the element of the dynamic weight matrix to construct a dynamic weight matrix;

[0009] Step 2: Build a time-series forecasting model using historical and real-time renewable energy data to predict future renewable energy supply capacity. Combined with the data center's computing load characteristics, this model dynamically adjusts load distribution, optimizes task execution time and resource allocation, and achieves dynamic matching with renewable energy supply levels.

[0010] Step 3: Calculate the carbon emission factors and priorities of different data centers, build a task migration feasibility matrix, and dynamically migrate computing tasks to low-carbon emission data centers based on the scheduling benefit function. This optimizes the task scheduling path, reduces cross-regional data transmission latency and energy consumption, and achieves cross-regional optimized scheduling of computing power.

[0011] Step 4: Build a multi-dimensional time series prediction model to predict carbon emission trends, monitor carbon emission anomalies in real time, divide warning levels and take corresponding measures to ensure the stable operation of the data center under the low-carbon operation target.

[0012] Further: Step 1 includes:

[0013] Obtain real-time power data for each energy type through the grid dispatch data interface, and combine smart meters with the SCADA system to monitor the proportion of power input sources in the data center area.

[0014] Calculate the power generation proportion of each energy type and construct a dynamic weight matrix for the power grid generation structure;

[0015] Establish a time series change trend analysis, calculate the first-order derivative of each energy proportion with respect to time, construct the power generation structure change rate vector, quantify the changing trend of different energy proportions, and provide a dynamic correction factor for adaptive carbon emission calculation.

[0016] Further: Step 2 includes:

[0017] Extract historical renewable energy generation data from the time series database to form a time series. Use the meteorological API to obtain real-time meteorological data on current solar radiation intensity, wind speed, and reservoir water level to construct a prediction input matrix.

[0018] Use multimodal long short-term memory network for prediction, and dynamically adjust the prediction value by combining historical error correction coefficient and trend correction coefficient to improve prediction accuracy;

[0019] Build a computational load weight vector based on the data center task type, calculate the load adjustment coefficient, and dynamically match the renewable energy supply level;

[0020] Based on renewable energy forecasts and load adjustment coefficients, the execution time of low-priority tasks is adjusted, the AI ​​computing load distribution is optimized, and task migration is performed when necessary. Computing tasks are scheduled to data centers with a higher proportion of renewable energy, achieving dynamic matching of data center load and renewable energy supply.

[0021] Further: Step 3 includes:

[0022] Based on the dynamic weight matrix of the power grid's power generation structure, the carbon emission factor of each data center at different times is calculated. The low-carbon computing power scheduling priority of the data center is measured using the carbon emission priority factor. Data centers with higher priorities are more suitable for receiving computing power tasks.

[0023] Construct a task migration feasibility matrix to evaluate the task migration latency between data centers. Combined with the scheduling benefit function, determine the feasibility of migrating tasks from high-carbon emission data centers to low-carbon emission data centers.

[0024] With the goal of minimizing the combined weight of carbon emissions and network latency during task scheduling, the optimal task scheduling path is calculated to avoid overload of a single data center due to a surge in tasks.

[0025] Data sharding is performed on migration tasks to reduce transmission bandwidth usage and improve parallelism. Task execution is monitored through edge computing nodes, and blockchain is used to record task migration paths to ensure the traceability and security of scheduling and achieve cross-regional optimized scheduling of computing power.

[0026] Further: Step 4 includes:

[0027] Combining data center load, renewable energy supply, grid carbon emission factors, task migration traffic, and historical carbon emission data, a multidimensional time series prediction model was constructed using a two-layer time series prediction network. The short-term trend prediction model is used to detect carbon emission fluctuations at the hourly level, while the long-term trend prediction model is used to predict carbon emission changes at the daily / weekly level.

[0028] Set carbon emission warning thresholds based on the historical carbon emission mean and standard deviation; use a two-layer adaptive anomaly detection model, combined with local anomaly factor detection and Bayesian mutation detection, to monitor carbon emission trends in real time and identify anomalies;

[0029] Warning levels are divided into mild, moderate and severe levels according to the degree of carbon emission abnormality; in mild warnings, lightweight load adjustments are performed; in moderate warnings, the proportion of low-carbon power usage is optimized by combining load adjustment and cross-regional computing power scheduling; in severe warnings, high-energy-consuming tasks are offloaded to low-carbon data centers, backup clean energy is enabled, and management is notified to lower the priority of non-critical tasks.

[0030] In another aspect, the present invention discloses a carbon emission monitoring system for a data center, comprising:

[0031] The module for constructing a dynamic weight matrix for the power generation structure of the power grid: This module collects data on the power generation share of different energy types in real time, combines meteorological information to predict the changing trend of renewable energy power supply, calculates the power supply share of each energy type, and constructs a dynamic weight matrix.

[0032] Renewable Energy Prediction and Load Adjustment Module: This module uses historical and real-time renewable energy data to build a time series prediction model to predict future renewable energy power supply capacity. This module, combined with the computing load characteristics of the data center, dynamically adjusts load distribution, optimizes task execution time and resource allocation, and achieves dynamic matching with renewable energy supply levels.

[0033] Cross-regional low-carbon computing power scheduling module: Calculates the carbon emission factors and priorities of different data centers, builds a task migration feasibility matrix, dynamically migrates computing tasks to low-carbon emission data centers based on the scheduling benefit function, optimizes task scheduling paths, reduces cross-regional data transmission latency and energy consumption, and realizes cross-regional optimized scheduling of computing power;

[0034] Carbon emission early warning module: Build a multi-dimensional time series prediction model to predict carbon emission trends, monitor carbon emission anomalies in real time, divide warning levels and take corresponding measures to ensure the stable operation of the data center under the low-carbon operation goal.

[0035] Compared with the prior art, the present invention has the following technical advances:

[0036] The present invention adopts an adaptive carbon emission calculation algorithm based on real-time power structure, which can obtain the energy structure of the power grid to which the data center is connected in real time and dynamically calculate the carbon emission factor. By adaptively adjusting the calculation model, the carbon emission calculation can accurately reflect the changes in the proportion of renewable energy, providing a reliable data basis for subsequent optimization. It solves the problem that the traditional method is only based on total power consumption statistics and cannot accurately calculate carbon emissions, making the carbon emission assessment more refined and dynamic. The present invention adopts a load adjustment algorithm based on renewable energy prediction, uses AI to predict the power supply capacity of renewable energy such as wind energy and photovoltaics, and combines the adjustability of data center computing tasks to increase the computing load during energy peaks and reduce the load or perform task migration during energy troughs. By intelligently adjusting the execution time of computing tasks, dynamic matching of computing power demand and renewable energy supply is achieved, thereby improving the utilization rate of renewable energy and reducing carbon emissions. It solves the problem that traditional load management methods cannot take into account the matching of renewable energy supply and demand, making data centers more low-carbon and intelligent.

[0037] The present invention adopts a cross-regional low-carbon computing power scheduling algorithm to perform intelligent task migration between multiple data centers, dynamically allocate computing tasks to areas with a higher proportion of low-carbon electricity, and realize cross-regional green computing resource scheduling. Combined with edge computing and distributed computing architecture, it maximizes the reduction of high-carbon electricity use and avoids the increase of carbon emissions due to insufficient green electricity supply in a single data center. It solves the problem of mismatch between the load distribution of data centers and the geographical distribution of renewable energy, making data center operations more flexible, efficient and low-carbon. The present invention adopts a carbon emission early warning algorithm based on intelligent detection. Through time series analysis, machine learning and Bayesian anomaly detection technology, it predicts future carbon emission trends and issues early warnings before exceeding the preset threshold. Combined with a multi-level response mechanism, different levels of control measures are implemented according to the degree of carbon emission anomaly, such as load adjustment, task migration, clean energy storage release, etc., to ensure the low-carbon and stable operation of the data center. It solves the problems of strong lag and lack of early warning mechanism of traditional monitoring methods, so that the data center has adaptive control capabilities and effectively reduces carbon emission risks.

[0038] In summary, the present invention can solve the problem of matching computing load with renewable energy faced by data centers. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0040] In the attached figure:

[0041] Figure 1 is a flow chart of the present invention;

[0042] Figure 2 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0043] The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below with reference to the accompanying drawings.

[0044] Example 1

[0045] like Figure 1 As shown, the present invention discloses a carbon emission monitoring method for a data center, comprising:

[0046] Step 1: Collect power generation data of different energy types in real time, calculate the power supply proportion of each energy type, and use this proportion as the element of the dynamic weight matrix to construct a dynamic weight matrix;

[0047] Step 2: Build a time-series forecasting model using historical and real-time renewable energy data to predict future renewable energy supply capacity. Combined with the data center's computing load characteristics, this model dynamically adjusts load distribution, optimizes task execution time and resource allocation, and achieves dynamic matching with renewable energy supply levels.

[0048] Step 3: Calculate the carbon emission factors and priorities of different data centers, build a task migration feasibility matrix, and dynamically migrate computing tasks to low-carbon emission data centers based on the scheduling benefit function. This optimizes the task scheduling path, reduces cross-regional data transmission latency and energy consumption, and achieves cross-regional optimized scheduling of computing power.

[0049] Step 4: Build a multi-dimensional time series prediction model to predict carbon emission trends, monitor carbon emission anomalies in real time, divide warning levels and take corresponding measures to ensure the stable operation of the data center under the low-carbon operation target.

[0050] Specifically, step 1 includes:

[0051] In order to realize adaptive carbon emission calculation based on real-time power structure, it is necessary to first construct the dynamic weight matrix W of the power grid generation structure. t This matrix is ​​used to describe the power supply ratio of different energy types in the power grid at a certain time t, thereby providing accurate input for subsequent carbon emission calculations. The specific implementation steps are as follows:

[0052] 1.1 Data Collection: Obtaining the Real-time Power Generation Ratio of Different Energy Types

[0053] Through the grid dispatch data interface (such as the API or open data platform provided by the State Grid or regional power companies), the power supply data of each energy type (coal, gas, hydropower, wind power, photovoltaic, nuclear power, etc.) at the current time t is obtained, which is recorded as P i (t), where i represents different energy types.

[0054] By combining smart meters with the SCADA system, the power input sources in the data center area can be monitored in real time to obtain the actual proportion of various types of power supply to improve the accuracy and timeliness of the data.

[0055] Through meteorological data interfaces (such as satellite remote sensing, weather stations, etc.), information such as wind speed, solar radiation intensity, and reservoir water level is obtained, and combined with historical data, the power supply change trend of renewable energy in the short term in the future is predicted to provide auxiliary reference for subsequent optimization.

[0056] 1.2 Data Processing: Constructing a Dynamic Weight Matrix for the Power Grid Generation Structure

[0057] Calculate the share of electricity generation by individual energy types:

[0058] Assume that the total power generation power of the data center grid is Ptotal (t), then the power supply proportion of each energy source i at time t is R i (t) is calculated as follows:

[0059]

[0060] in, Represents the sum of all energy types, ensuring that the proportion of all energy satisfies the normalization constraint:

[0061]

[0062] Construct a dynamic weight matrix of the power grid generation structure:

[0063] After calculating the power supply proportion of each energy type, the power generation structure weight matrix W_t is constructed:

[0064] W t =[R coal (t) R gas (t) R hydro (t) R wind (t) R solar (t) R nuclear (t)]

[0065] in:

[0066] R coal (t) represents the proportion of coal-fired power generation;

[0067] R gas (t) represents the proportion of gas-fired power generation;

[0068] R hydro (t) represents the proportion of hydropower generation;

[0069] R wind (t) represents the proportion of wind power generation;

[0070] R solar (t) represents the proportion of photovoltaic power generation;

[0071] R nuclear (t) represents the proportion of nuclear power generation.

[0072] 1.3 Calculating the rate of change of power generation structure: Establishing time series trend analysis

[0073] Calculate the first-order time derivative of the power generation share:

[0074] In order to analyze the dynamic change trend of the power generation structure of the power grid, the change rate of each energy proportion over time is calculated:

[0075]

[0076] Here, Δt is the sampling time interval, which is usually set at the minute level to ensure the real-time calculation.

[0077] Construct the power generation structure change rate vector:

[0078]

[0079] This vector is used to quantify the changing trend of different energy proportions and provide a dynamic correction factor for subsequent adaptive carbon emission calculations.

[0080] This step forms a complete dynamic weight matrix W of the power generation structure through real-time data collection, power grid power supply ratio calculation, power generation structure dynamic weight matrix construction, time series trend analysis and data storage and call. t , providing accurate input data for subsequent adaptive carbon emission calculations.

[0081] Specifically, step 2 includes:

[0082] After completing step 1 (dynamic weight matrix W of power grid generation structure t After the construction of a data center, data centers can obtain the current power supply share of various energy sources in real time. However, due to the volatility of renewable energy, relying solely on the current power supply structure to optimize carbon emissions is still limited. Therefore, it is necessary to predict future renewable energy supply capacity and dynamically adjust load distribution based on the computing load characteristics of the data center to maximize the matching of renewable energy supply levels and thus optimize carbon emissions.

[0083] 2.1 Data Collection: Obtaining Historical and Real-time Data on Renewable Energy

[0084] In order to predict the renewable energy power supply capacity at the future time t+τ, the following key data needs to be collected:

[0085] Obtaining historical time series data

[0086] Extract renewable energy generation data of the past n time windows from a time series database (such as InfluxDB, TimescaleDB) to form a time series ε = {E solar (t),E wind (t),E hydro (t)}.

[0087] Among them, E solar (t) represents the photovoltaic power generation, E wind (t) represents the wind power generation, E hydro (t) represents the hydropower generation.

[0088] Get real-time weather data

[0089] Get the current solar radiation intensity S(t) and wind speed V through the meteorological API w (t), reservoir water level H w (t) and other data as prediction input.

[0090] Constructing the prediction input matrix

[0091] Set the prediction window length T f , define the input matrix X t :

[0092]

[0093] This matrix will be used as input for the renewable energy forecasting model.

[0094] 2.2 Renewable Energy Generation Forecasting: Building a Time Series Forecasting Model

[0095] Prediction using multimodal long short-term memory networks

[0096] Since renewable energy power supply is greatly affected by meteorological factors and the time series data characteristics of different energy types are different, a multimodal long short-term memory network model is adopted.

[0097] Setting up the prediction model:

[0098]

[0099] in, is the predicted power supply of the i-th energy source at time τ in the future, and θ represents the model parameters.

[0100] Construct a dynamic correction mechanism for prediction errors

[0101] Since prediction errors may lead to load adjustment deviations, a dynamic correction factor is introduced:

[0102]

[0103] in:

[0104] α is the historical error correction coefficient,

[0105] β is the trend correction coefficient,

[0106] Represents the rate of change of renewable energy generation.

[0107] Calculate the final revised forecast value:

[0108]

[0109] 2.3 Calculating the load adjustment factor: matching renewable energy supply levels

[0110] Construct a computational load weight vector and set the computational load proportion vectors for different types of tasks:

[0111] L t =[W batch (t)W realtime (t)W AI (t)]

[0112] in:

[0113] W batch (t) represents low-priority tasks that can be delayed (such as AI model training and data analysis).

[0114] W realtime (t) represents high-priority real-time computing tasks (such as video rendering and database transactions).

[0115] W AI (t) represents optimizable AI computing tasks (such as inference computing).

[0116] Calculate the load adjustment factor λ t , using the normalization method, calculate the ratio of renewable energy power supply to the historical maximum:

[0117]

[0118] in:

[0119] represents the revised forecast of total renewable energy,

[0120] γ controls the renewable energy weight,

[0121] δ controls the calculation load weight,

[0122] λ t Used to adjust mission loads to dynamically match renewable energy supply levels.

[0123] 2.4 Dynamic Load Adjustment: Task Migration and Power Optimization

[0124] Adjust the computation time of low-priority tasks

[0125] When λ t When >1 (sufficient renewable energy), batch tasks are executed in advance to take advantage of high green power periods.

[0126] When λ t When <1 (insufficient renewable energy), the computing resource allocation for batch tasks is postponed or reduced.

[0127] 2. Optimize AI computing load distribution

[0128] For AI computing tasks, according to λ t Adjust the operating mode of GPUs and TPUs to reduce computing loads during periods of high carbon emissions.

[0129] 3. Distributed Task Migration

[0130] Combined with subsequent cross-regional low-carbon computing power scheduling algorithms, some computing tasks will be migrated to data centers with a higher proportion of renewable energy.

[0131] This step is based on the grid power generation structure data from step 1, and utilizes time series forecasting technology, dynamic correction mechanism, and load adjustment coefficient calculation method to achieve dynamic matching of data center load and renewable energy supply, thereby reducing carbon emissions and improving energy utilization efficiency.

[0132] Specifically, step 3 includes:

[0133] After completing step 1 (dynamic weight matrix W of power grid generation structure t After implementing the load adjustment algorithm based on renewable energy forecasts (and step 2), data centers are now able to dynamically adjust computing loads based on local renewable energy supply capabilities. However, due to the spatial mismatch between renewable energy supply levels in data centers across different regions, optimizing a single data center still cannot fully reduce overall carbon emissions. Therefore, the core objectives of step 3 are:

[0134] 1. Calculate the carbon emission factors of different data centers based on the proportion of renewable energy in cross-regional power grids;

[0135] 2. Dynamically migrate computing tasks between multiple data centers, shifting computing power demands from high-carbon emission areas to low-carbon emission areas;

[0136] 3. Optimize task scheduling paths and reduce cross-regional data transmission latency and energy loss to ensure efficient execution.

[0137] 3.1 Calculation of carbon emission factors for cross-regional data centers

[0138] Establish a data center carbon emission factor calculation model and set the carbon emission factor C of the jth data center at time t j (t):

[0139]

[0140] in:

[0141] R i,j (t) represents the proportion of the i-th type of energy used by data center j at time t. This data comes from the dynamic weight matrix W of the power generation structure in step 1. t .

[0142] EF i is the unit carbon emission factor of category i energy (kg CO2 / kWh), which can be provided by the International Energy Agency (IEA) or local grid operators.

[0143] Calculate the carbon emission priority factor of the data center and set the carbon emission priority factor P of the data center j (t), used to measure the low-carbon computing power scheduling priority of data centers:

[0144]

[0145] in:

[0146] C max (t) is the highest carbon emission factor among all data centers at the current moment;

[0147] C min (t) is the lowest carbon emission factor among all data centers at the current moment.

[0148] When P j The larger (t) is, the more data center j has a cleaner power supply and is more suitable for receiving computing tasks.

[0149] 3.2 Computing Task Migration Decision: Cross-Region Computing Power Optimization

[0150] Construct the task migration feasibility matrix and set the data center set as D = {D1, D2, ..., D m The feasibility matrix M of task migration between different data centers is defined as follows:

[0151]

[0152] Where T jk Represents data center D j Migrate tasks to data center D k The transmission delay is determined by factors such as network bandwidth, link status between data centers, and task data size. jk Only migrations within the threshold range will be considered to ensure task execution efficiency.

[0153] Task migration decision function, computing tasks from high carbon emission data center D j Migrate to a low-carbon data center k The scheduling profit function is:

[0154] G jk (t) = α·(C j (t)-C k (t))-β·T jk

[0155] in:

[0156] α is the carbon emission reduction weight factor;

[0157] β is the weight factor of task transmission delay;

[0158] When G jk When (t)>0, task migration is beneficial to the overall carbon emission optimization, and task scheduling is allowed.

[0159] Optimal task scheduling path calculation, in order to perform optimal low-carbon computing power migration between multiple data centers, the following algorithm is used to minimize the combined weight of carbon emissions and network latency during task scheduling.

[0160] Set the objective function:

[0161]

[0162] in A set of task migration paths.

[0163] This step ensures that tasks can be executed on time in the target data center, and prioritizes scheduling tasks to data centers with lower carbon emission factors, preventing a single data center from being overloaded due to a surge in tasks.

[0164] 3.3 Task Migration Execution: Intelligent Distribution of Scheduling Instructions

[0165] Task packaging and splitting

[0166] Data sharding is performed on the computing tasks to be migrated to reduce bandwidth usage during data transmission and improve task parallelism. Task packaging solution:

[0167] T package ={T1,T2,...,T n}

[0168] in,

[0169] The scheduling instruction issuance mechanism monitors task execution through edge computing nodes to ensure that the scheduling process does not affect the execution of key tasks. It adopts a scheduling confirmation mechanism based on smart contracts and uses blockchain to record task migration paths to ensure the traceability and security of task scheduling.

[0170] This step ensures the efficiency of computing task execution while achieving dynamic cross-regional scheduling of computing power, maximizing the use of low-carbon electricity resources, and effectively reducing the overall carbon emissions of the data center.

[0171] Specifically, step 4 includes:

[0172] After completing Step 1 (Adaptive Carbon Emission Calculation Algorithm for Real-Time Power Structure), Step 2 (Load Adjustment Algorithm Based on Renewable Energy Forecast), and Step 3 (Cross-Region Low-Carbon Computing Power Dispatching Algorithm), the data center has been able to optimize in terms of carbon emission estimation, dynamic load adjustment, and cross-region low-carbon computing power dispatching. However, due to the instability of the power grid structure, the uncertainty of renewable energy supply, and the volatility of data center computing task load, there is still a risk of unexpected increase in carbon emissions. Therefore, a carbon emission early warning algorithm based on intelligent detection is needed to identify high-carbon emission risks in advance and provide automated intervention measures to ensure stable operation of the data center under the low-carbon operation target.

[0173] 4.1 Prediction Model Construction: Carbon Emission Trend Prediction Based on Multidimensional Data

[0174] To achieve accurate carbon emission early warning, a multidimensional time series prediction model needs to be constructed, which needs to consider factors such as data center load, renewable energy supply, power grid carbon emission factors, and task migration.

[0175] Input Variable Selection

[0176] Set the key factor set that affects the carbon emission of the data center at time t:

[0177] X t ={L t ,R t ,W t ,M t ,H t ,T t}

[0178] Where:

[0179] L t : Current computing load of the data center (CPU, GPU utilization).

[0180] R t : Renewable energy proportion (wind, photovoltaic, etc.).

[0181] W t : Carbon emission factor matrix in the current power grid structure (calculated by Step 1).

[0182] M t : Cross-region task migration flow (calculated by the Cross-Region Low-Carbon Computing Power Dispatching Algorithm in Step 3).

[0183] H t : Historical carbon emission data (used for time series modeling).

[0184] T t : External environmental factors such as ambient temperature and humidity.

[0185] A two-layer time series forecasting network is used for short-term trend forecasting (hourly level) and long-term trend forecasting (day / week level):

[0186]

[0187] Short-term trend prediction model (LSTM): used to predict carbon emission changes within the next 16 hours and suitable for short-term fluctuation detection.

[0188] Long-term trend prediction model (Transformer): used to predict carbon emission changes at the daily level to detect potential structural carbon emission anomalies.

[0189] Final carbon emission forecast:

[0190]

[0191] Among them, λ is the weighted coefficient of short-term prediction and long-term prediction, which is dynamically adjusted according to actual demand.

[0192] 4.2 Carbon Emission Anomaly Detection Mechanism

[0193] Set carbon emission warning thresholds. Based on historical data and policy goals, set warning thresholds for data center carbon emissions:

[0194] C thresh =C avg +γ·σ C

[0195] in:

[0196] C avg : Historical average carbon emissions.

[0197] σ C : Standard deviation of carbon emissions.

[0198] γ: Adjustment factor (can be dynamically adjusted according to data center strategy).

[0199] Intelligently detect carbon emission anomalies, using a two-layer adaptive anomaly detection model, combined with local anomaly factors and Bayesian-based mutation detection, to monitor carbon emission trends in real time:

[0200] Local anomaly factor detection formula:

[0201]

[0202] When LOF(C t )>δ(threshold), the current carbon emission is judged as an abnormal point.

[0203] Bayesian Mutation Detection:

[0204]

[0205] When P(change|C t )>τ, it is considered that the data center may enter an abnormal carbon emission state.

[0206] 4.3 Warning level classification and response mechanism

[0207] According to the degree of abnormal carbon emissions, warning levels are divided and different levels of response measures are taken:

[0208] Mild warning (Level 1):

[0209] Trigger condition: C t Beyond C thresh But not exceeding 1.2C thresh .

[0210] Treatment measures: Lightweight load adjustment (the load adjustment algorithm based on renewable energy forecast in step 2 reduces the CPU frequency); early warning information is sent to data center operations and maintenance personnel.

[0211] Moderate Warning (Level 2):

[0212] Trigger condition: 1.2C thresh <C t <1.5C thresh .

[0213] Treatment measures: Enhanced load adjustment (the load adjustment algorithm based on renewable energy prediction in step 2 is combined with the cross-regional low-carbon computing power scheduling algorithm in step 3 to migrate some computing tasks to low-carbon data centers); dynamically adjust the task scheduling strategy to increase the proportion of low-carbon electricity use.

[0214] Severe warning (Level 3):

[0215] Trigger Condition C t >1.5C thresh .

[0216] Handling measures: Trigger emergency task offloading and migrate high-energy-consuming tasks to low-carbon data centers; enable backup clean energy (such as energy storage systems); notify management and, if necessary, lower the execution priority of some non-critical computing tasks.

[0217] The carbon emission early warning algorithm based on intelligent detection mainly achieves real-time detection and intelligent early warning of carbon emissions through the following steps:

[0218] This step can issue an early warning when carbon emissions are about to exceed the standard, and combine with the previous algorithms (adaptive carbon emission calculation algorithm based on real-time power structure, load adjustment algorithm based on renewable energy prediction, and continuous cross-regional low-carbon computing power scheduling algorithm) to perform intelligent control, thereby achieving low-carbon operation of the data center while ensuring computing power requirements.

[0219] Example 2

[0220] like Figure 2 As shown, the present invention discloses a carbon emission monitoring system for a data center, comprising:

[0221] The module for constructing a dynamic weight matrix for the power generation structure of the power grid: This module collects data on the power generation share of different energy types in real time, combines meteorological information to predict the changing trend of renewable energy power supply, calculates the power supply share of each energy type, and constructs a dynamic weight matrix.

[0222] Renewable Energy Prediction and Load Adjustment Module: This module uses historical and real-time renewable energy data to build a time series prediction model to predict future renewable energy power supply capacity. This module, combined with the computing load characteristics of the data center, dynamically adjusts load distribution, optimizes task execution time and resource allocation, and achieves dynamic matching with renewable energy supply levels.

[0223] Cross-regional low-carbon computing power scheduling module: Calculates the carbon emission factors and priorities of different data centers, builds a task migration feasibility matrix, dynamically migrates computing tasks to low-carbon emission data centers based on the scheduling benefit function, optimizes task scheduling paths, reduces cross-regional data transmission latency and energy consumption, and realizes cross-regional optimized scheduling of computing power;

[0224] Carbon emission early warning module: Build a multi-dimensional time series prediction model to predict carbon emission trends, monitor carbon emission anomalies in real time, divide warning levels and take corresponding measures to ensure the stable operation of the data center under the low-carbon operation goal.

[0225] The modules in Example 2 are used to implement the functions in Example 1. This embodiment can be implemented via a system comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the method for monitoring carbon emissions for a data center according to Example 1 of this application is implemented. The system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are well known in the art and are therefore not described in detail here.

[0226] In the present application, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in this application may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0227] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A carbon emission monitoring method for a data center, characterized in that: include: Step 1: Collect power generation data of different energy types in real time, calculate the power supply proportion of each energy type, and use this proportion as the element of the dynamic weight matrix to construct a dynamic weight matrix; Step 2: Build a time series forecasting model using historical and real-time renewable energy data to predict future renewable energy supply capacity. Combined with the data center, calculate the load adjustment coefficient to dynamically adjust load distribution, optimize task execution time and resource allocation, and achieve dynamic matching with renewable energy supply levels. Calculating the load adjustment coefficient includes: Construct a computational load weight vector and set the computational load proportion vectors for different types of tasks: in: Represents a low-priority task that can be delayed; Represents high-priority real-time computing tasks; Represents optimizable AI computing tasks; Calculating Load Adjustment Factor , using the normalization method, calculate the ratio of renewable energy power supply to the historical maximum: in: represents the revised total renewable energy forecast; Control the weight of renewable energy, Control calculation load weight, Used to adjust mission loads to dynamically match renewable energy supply levels; Dynamically adjust load distribution including: Adjust the calculation time of low priority tasks: when When renewable energy is abundant, batch processing tasks are executed in advance to take advantage of high green power periods; when When (renewable energy is insufficient), postpone or reduce the allocation of computing resources for batch tasks; Step 3: Calculate the carbon emission factors and priorities of different data centers, build a task migration feasibility matrix, and dynamically migrate computing tasks to low-carbon emission data centers based on the scheduling benefit function. This optimizes the task scheduling path, reduces cross-regional data transmission latency and energy consumption, and achieves cross-regional optimized scheduling of computing power, including: Establish a data center carbon emission factor calculation model and set the Data centers at the time Carbon emission factor : in: Representative data center At the moment The used The proportion of energy sources; For the Unit carbon emission factor of energy type (kg CO2 / kWh); Calculate the carbon emission priority factor of the data center and set the carbon emission priority factor of the data center , used to measure the low-carbon computing power scheduling priority of data centers: in: It is the highest carbon emission factor among all data centers at the moment; It has the lowest carbon emission factor among all data centers at the moment; Construct a task migration feasibility matrix and set the data center set as , feasibility matrix of task migration between different data centers The definition is as follows: in Representative data center Migrate tasks to the data center transmission delay; Task migration decision function, computing tasks from high carbon emission data centers Migrate to a low-carbon data center The scheduling profit function is: in: Reduce weighting factors for carbon emissions; is the weight factor of task transmission delay; when When task migration is beneficial to the overall carbon emission optimization, task scheduling is allowed; Optimal task scheduling path calculation: To perform optimal low-carbon computing power migration between multiple data centers, the following algorithm is used to minimize the combined weight of carbon emissions and network latency during task scheduling. Set the objective function: in is a set of task migration paths; Step 4: Build a multidimensional time series prediction model, which includes: Set at time A collection of key factors affecting data center carbon emissions: in: : Current computing load of the data center; : proportion of renewable energy; : Carbon emission factor matrix in the current power grid structure; : Cross-region task migration traffic; : historical carbon emissions data; : External environmental factors such as ambient temperature and humidity; Predict carbon emission trends, monitor carbon emission anomalies in real time, classify warning levels and take corresponding measures to ensure stable operation of the data center under the low-carbon operation goal, including: Set carbon emission warning thresholds. Based on historical data and policy goals, set warning thresholds for data center carbon emissions: in: : historical carbon emission average; : carbon emission standard deviation; : regulatory factor; Intelligently detect carbon emission anomalies, using a two-layer adaptive anomaly detection model, combined with local anomaly factors and Bayesian-based mutation detection, to monitor carbon emission trends in real time: Local anomaly factor detection formula: when When the threshold is reached, the current carbon emissions are judged as abnormal points; Bayesian Mutation Detection: when When the data center enters an abnormal carbon emission state.

2. A carbon emission monitoring method for a data center according to claim 1, characterized in that: The step 1 comprises: Obtain real-time power data for each energy type through the grid dispatch data interface, and combine smart meters with the SCADA system to monitor the proportion of power input sources in the data center area. Calculate the power generation proportion of each energy type and construct a dynamic weight matrix for the power grid generation structure; Establish a time series change trend analysis, calculate the first-order derivative of each energy proportion with respect to time, construct the power generation structure change rate vector, quantify the changing trend of different energy proportions, and provide a dynamic correction factor for adaptive carbon emission calculation.

3. The carbon emission monitoring method for a data center according to claim 2, characterized in that: The step 2 includes: Extract historical renewable energy generation data from the time series database to form a time series. Use the meteorological API to obtain real-time meteorological data on current solar radiation intensity, wind speed, and reservoir water level to construct a prediction input matrix. Use multimodal long short-term memory network for prediction, and dynamically adjust the prediction value by combining historical error correction coefficient and trend correction coefficient to improve prediction accuracy; Build a computational load weight vector based on the data center task type, calculate the load adjustment coefficient, and dynamically match the renewable energy supply level; Based on renewable energy forecasts and load adjustment coefficients, the execution time of low-priority tasks is adjusted, the AI ​​computing load distribution is optimized, and task migration is performed when necessary. Computing tasks are scheduled to data centers with a higher proportion of renewable energy, achieving dynamic matching of data center load and renewable energy supply.

4. A carbon emission monitoring method for a data center according to claim 3, characterized in that: The step 3 comprises: Based on the dynamic weight matrix of the power grid's power generation structure, the carbon emission factor of each data center at different times is calculated. The low-carbon computing power scheduling priority of the data center is measured using the carbon emission priority factor. Data centers with higher priorities are more suitable for receiving computing power tasks. Construct a task migration feasibility matrix to evaluate the task migration latency between data centers. Combined with the scheduling benefit function, determine the feasibility of migrating tasks from high-carbon emission data centers to low-carbon emission data centers. With the goal of minimizing the combined weight of carbon emissions and network latency during task scheduling, the optimal task scheduling path is calculated to avoid overload of a single data center due to a surge in tasks. Data sharding is performed on migration tasks to reduce transmission bandwidth usage and improve parallelism. Task execution is monitored through edge computing nodes, and blockchain is used to record task migration paths to ensure the traceability and security of scheduling and achieve cross-regional optimized scheduling of computing power.

5. A carbon emission monitoring method for a data center according to claim 4, characterized in that: The step 4 comprises: Combining data center load, renewable energy supply, grid carbon emission factors, task migration traffic, and historical carbon emission data, a multidimensional time series prediction model was constructed using a two-layer time series prediction network. The short-term trend prediction model is used to detect carbon emission fluctuations at the hourly level, while the long-term trend prediction model is used to predict carbon emission changes at the daily / weekly level. Set carbon emission warning thresholds based on the historical carbon emission mean and standard deviation; use a two-layer adaptive anomaly detection model, combined with local anomaly factor detection and Bayesian mutation detection, to monitor carbon emission trends in real time and identify anomalies; Warning levels are divided into mild, moderate and severe levels according to the degree of carbon emission abnormality; in mild warnings, lightweight load adjustments are performed; in moderate warnings, the proportion of low-carbon power usage is optimized by combining load adjustment and cross-regional computing power scheduling; in severe warnings, high-energy-consuming tasks are offloaded to low-carbon data centers, backup clean energy is enabled, and management is notified to lower the priority of non-critical tasks.

6. A carbon emission monitoring system for a data center, characterized in that: include: The module for constructing a dynamic weight matrix for the power grid generation structure collects power generation data of different energy types in real time, calculates the power supply proportion of each energy type, and uses this proportion as an element of the dynamic weight matrix to construct a dynamic weight matrix. Renewable energy prediction and load adjustment module: This module uses historical and real-time renewable energy data to build a time series prediction model to predict future renewable energy power supply capacity. It then calculates the load adjustment coefficient based on the data center, dynamically adjusts load distribution, optimizes task execution time and resource allocation, and achieves dynamic matching with renewable energy supply levels. The calculation of the load adjustment coefficient includes: Construct a computational load weight vector and set the computational load proportion vectors for different types of tasks: in: Represents a low-priority task that can be delayed; Represents high-priority real-time computing tasks; Represents optimizable AI computing tasks; Calculating Load Adjustment Factor , using the normalization method, calculate the ratio of renewable energy power supply to the historical maximum: in: represents the revised total renewable energy forecast; Control the weight of renewable energy, Control calculation load weight, Used to adjust mission loads to dynamically match renewable energy supply levels; Dynamically adjust load distribution including: Adjust the calculation time of low priority tasks: when When renewable energy is abundant, batch processing tasks are executed in advance to take advantage of high green power periods; when When (renewable energy is insufficient), postpone or reduce the allocation of computing resources for batch tasks; Cross-regional low-carbon computing power scheduling module: Calculates the carbon emission factors and priorities of different data centers, builds a task migration feasibility matrix, dynamically migrates computing tasks to low-carbon emission data centers based on the scheduling benefit function, optimizes task scheduling paths, reduces cross-regional data transmission latency and energy consumption, and realizes cross-regional optimized scheduling of computing power, including: Establish a data center carbon emission factor calculation model and set the Data centers at the time Carbon emission factor : in: Representative data center At the moment The used The proportion of energy sources; For the Unit carbon emission factor of energy type (kg CO2 / kWh); Calculate the carbon emission priority factor of the data center and set the carbon emission priority factor of the data center , used to measure the low-carbon computing power scheduling priority of data centers: in: It is the highest carbon emission factor among all data centers at the moment; It has the lowest carbon emission factor among all data centers at the moment; Construct a task migration feasibility matrix and set the data center set as , feasibility matrix of task migration between different data centers The definition is as follows: in Representative data center Migrate tasks to the data center transmission delay; Task migration decision function, computing tasks from high carbon emission data centers Migrate to a low-carbon data center The scheduling profit function is: in: Reduce weighting factors for carbon emissions; is the weight factor of task transmission delay; when When task migration is beneficial to the overall carbon emission optimization, task scheduling is allowed; Optimal task scheduling path calculation: To perform optimal low-carbon computing power migration between multiple data centers, the following algorithm is used to minimize the combined weight of carbon emissions and network latency during task scheduling. Set the objective function: in is a set of task migration paths; Carbon emission early warning module: Build a multi-dimensional time series prediction model, which includes: Set at time A collection of key factors affecting data center carbon emissions: in: : Current computing load of the data center; : proportion of renewable energy; : Carbon emission factor matrix in the current power grid structure; : Cross-region task migration traffic; : historical carbon emissions data; : External environmental factors such as ambient temperature and humidity; Predict carbon emission trends, monitor carbon emission anomalies in real time, classify warning levels and take corresponding measures to ensure stable operation of the data center under the low-carbon operation goal, including: Set carbon emission warning thresholds. Based on historical data and policy goals, set warning thresholds for data center carbon emissions: in: : historical carbon emission average; : carbon emission standard deviation; : regulatory factor; Intelligently detect carbon emission anomalies, using a two-layer adaptive anomaly detection model, combined with local anomaly factors and Bayesian-based mutation detection, to monitor carbon emission trends in real time: Local anomaly factor detection formula: when When the threshold is reached, the current carbon emissions are judged as abnormal points; Bayesian Mutation Detection: when When the data center enters an abnormal carbon emission state.

Images