Joint Optimization Method for Shared Energy Storage and Computing Load in Data Center Microgrid Clusters
By utilizing carbon market volatility signals and hierarchical scheduling to optimize energy storage and computing load in data center microgrids, the problems of energy storage capacity allocation and computing power migration were solved, achieving coordinated carbon and electricity scheduling and improving system stability and economy.
Patent Information
- Application Number
- CN202611632484.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing data center microgrid clusters lack responsiveness in energy storage capacity allocation strategies when the carbon trading market fluctuates. The migration of computing load does not take into account the temperature changes and carbon emission costs of the receiving microgrid. Carbon finance market information is not effectively transmitted in multi-layer scheduling, resulting in energy storage equipment lifespan loss, scheduling oscillations, and distorted economic assessments.
By acquiring carbon market volatility signals, adjusting the intensity of energy storage capacity allocation, and combining carbon futures forward prices, carbon option implied volatility, and real-time carbon spot prices for hierarchical scheduling across multiple time scales, the joint management of energy storage and computing loads is optimized. The cooling energy consumption of the migrated loads and the carbon emission costs of the receiving microgrid are taken into account, forming a closed loop of carbon and electricity coordinated scheduling.
It achieves dynamic balance response of energy storage capacity during carbon market fluctuations, avoids frequent energy storage switching, optimizes computing load migration targets, improves the economy and carbon emission management efficiency of data center microgrids, and ensures system stability and reliability.
Smart Images

Figure CN123121732A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of energy management for data center microgrids, specifically a method for joint optimization of shared energy storage and computing load in data center microgrid clusters. Background Technology
[0002] As data centers continue to expand in scale, their high energy consumption and high carbon emissions are becoming increasingly prominent issues. Interconnecting multiple data center microgrids to form a cluster collaborative system, and configuring renewable energy units and shared energy storage systems, has become an important technological approach to reduce data center operating costs and carbon emissions.
[0003] In existing data center microgrid scheduling methods, the allocation of shared energy storage capacity is typically based solely on static or quasi-static divisions of load forecasts and renewable energy output forecasts for each microgrid. Real-time corrections primarily respond to instantaneous imbalances in power supply and demand. When carbon trading market prices fluctuate drastically, energy storage capacity allocation strategies lack the ability to perceive and respond to carbon finance risk signals. This leads to frequent adjustments to energy storage charging and discharging strategies during periods of short-term, sharp carbon price fluctuations, resulting in energy storage equipment lifespan degradation and scheduling oscillations.
[0004] Regarding computing load migration, existing methods primarily determine the migration direction of batch processing loads based on differences in electricity prices or the abundance of renewable energy. They fail to couple the heat load generated during migration with the temperature changes of the receiving microgrid server cluster, and also fail to convert the additional cooling energy consumption caused by migration into incremental carbon emission costs based on carbon allowance prices and incorporate this into the migration decision. When carbon prices rise, migration schemes cannot automatically avoid high-carbon-cost recipients, leading to distorted economic assessments of computing load migration.
[0005] Regarding the coordination between carbon trading and hierarchical scheduling, while existing methods have introduced multi-level carbon trading mechanisms and multi-layered distributed solution architectures, there is a lack of effective signal transmission channels between carbon market price signals and the various levels of hierarchical scheduling. Forward-looking carbon price signals, volatility signals, and real-time carbon price signals are not embedded into the corresponding time-scale scheduling levels, resulting in a failure to achieve hierarchical transmission of carbon financial market information between macro-level global allocation, meso-level regional coordination, and micro-level local tracking, leading to insufficient coordination among carbon, electricity, and computing flows.
[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0007] The purpose of this invention is to provide a method for jointly optimizing shared energy storage and computing load in data center microgrid clusters, so as to solve the problems mentioned in the background art.
[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for jointly optimizing shared energy storage and computing load in a data center microgrid cluster includes the following steps: Obtain carbon market volatility signals and determine the correction strength for energy storage capacity allocation based on these signals, wherein the correction strength decreases as the carbon market volatility signal increases. The initial energy storage capacity allocation plan is rolled over and adjusted using this adjustment intensity to obtain the energy storage power capacity and energy storage capacity allocated to each microgrid. Based on the energy storage power capacity and energy storage capacity, the migration direction and migration load amount of the batch processing load are determined. The additional cooling energy consumption required for the additional heat generation after migration to the receiving party is converted into carbon emission increment cost based on the internal carbon quota spot price. This carbon emission increment cost is included in the migration cost. Layered scheduling is performed on multiple time scales in descending order. Carbon futures forward price signals are used to correct global capacity allocation, carbon option implied volatility signals are used to adjust regional coordination penalty parameters, and carbon spot real-time price signals are used to correct the charging and discharging tracking deviation weights of local tracking. Output and execute energy storage charging and discharging power, main grid switching power, microgrid inter-trading power, migration load, and carbon quota trading volume.
[0009] Furthermore, this carbon market volatility signal is obtained by a weighted combination of carbon futures price volatility and carbon option implied volatility; The correction strength is set at a baseline value when the carbon market volatility signal is below a low threshold and at a minimum value when the carbon market volatility signal is above a high threshold. When the carbon market volatility signal is between the low threshold and the high threshold, it decreases monotonically as the carbon market volatility signal increases.
[0010] Furthermore, the initial energy storage capacity allocation plan is based on the weighted sum of the proportion of each microgrid's day-ahead forecast load in the cluster and the proportion of its day-ahead forecast renewable energy output in the cluster, forming the basic weight of each microgrid. The adjustment weight of each microgrid is determined by the proportion of the absolute value of the actual net load of each microgrid to the absolute value of the total net load of the cluster. The adjustment amount for energy storage capacity allocation is determined by the difference between the adjustment weight and the base weight, and the adjustment strength.
[0011] Furthermore, the energy storage capacity allocation correction includes a power capacity correction and an energy capacity correction. When the adjustment weight is greater than the base weight, both the power capacity correction and the energy capacity correction are positive, and both increase monotonically with the increase of the absolute value of the actual net load, with the growth rate decreasing and approaching saturation. When the adjustment weight is less than or equal to the base weight, both the power capacity correction and the energy capacity correction are negative, and both decrease monotonically with the increase of the absolute value of the actual net load and approach zero. The final comprehensive weight of the previous period and the current adjustment weight are smoothed and then normalized to obtain the final comprehensive weight of each microgrid in the current period. The final comprehensive weight of the previous period is the final comprehensive weight obtained in the previous period using the same smoothing and normalization method. The final energy storage power capacity and energy storage capacity allocated to each microgrid are determined by the final comprehensive weight, and the sum of the allocated capacities of all microgrids is equal to the total available capacity after deducting the system-level spinning reserve capacity from the total shared energy storage capacity.
[0012] Furthermore, the migration direction and migration load amount of this batch processing workload are determined in the following way: The migration load is coupled to the temperature change of the receiving server cluster, and the migration load does not exceed the thermal response capacity of the receiving cooling system. The additional cooling energy consumption resulting from the relocation is converted into the incremental cost of carbon emissions based on the spot price of the internal carbon quota, the carbon emission coefficient of the main grid electricity purchase, and the energy efficiency ratio of the cooling system. When selecting relocation targets, priority should be given to avoiding microgrids with high costs for incremental carbon emissions or weak thermal response capabilities.
[0013] Furthermore, the temperature change of the receiving server cluster is updated based on the recursive relationship between the temperature of the previous period, the difference between server heat generation and cooling, heat capacity, and time interval. The heat generation of the server and the total power of the server are determined according to the heat generation coefficient; The migration load increases the server's total power and thus its temperature; when the temperature reaches the safe limit or the rate of temperature rise exceeds the thermal response capability of the receiving party's cooling system, the receiving party rejects the migration load. The additional cooling energy consumption generated by the migration is determined by the migration load, the heat generation coefficient, the energy efficiency ratio of the cooling system, the carbon emission coefficient of the main grid electricity purchase, and the spot price of the internal carbon quota.
[0014] Furthermore, in this hierarchical scheduling: The longest timescale will be used to adjust the overall cost of shared energy storage capacity allocation using the carbon futures forward price signal. The intermediate timescale is used to determine the initial value of the regional coordination penalty parameter by using the implied volatility signal of the carbon option. The shortest time scale is used as the real-time spot price signal of carbon as the correction term for the charge and discharge tracking deviation weight in the local tracking optimization; The three time scales are updated sequentially in the direction of decreasing time.
[0015] Furthermore, the first level performs global capacity allocation at a first preset time interval, and applies the correction term of the ratio of the carbon futures forward price signal to the reference carbon price to the effective use cost of the shared energy storage power capacity and energy storage capacity. The second level performs regional coordination at the second preset time interval. The correction term of the ratio of the implied volatility signal of the carbon option to the reference volatility is used to determine the initial value of the regional coordination penalty parameter. Each microgrid solves the objective in parallel with the local cost term, the adjacent microgrid deviation term and the regional coordination penalty parameter until the original residual and the dual residual both meet the convergence condition. The third level performs local rolling tracking at the third preset time interval, at the cost of the deviation between the actual output of the local equipment and the reference trajectory and the control increment, and applies the charge and discharge tracking deviation weight after the real-time carbon spot price signal is corrected to the cost. Each solution is obtained by issuing only the most recent control increment and rolling forward.
[0016] Furthermore, the migration direction and migration load of this batch processing workload are also subject to the following constraints: The queued load accumulation represents the amount of batch processing load that has arrived but not yet been processed locally and transferred out as of the end of the current time period. It is updated according to the recursive relationship of the remaining amount in the previous time period, newly arrived load, locally processed load, transferred out load, and transferred in load, and ensures that each batch processing load is processed within the specified time limit after arrival. The amount transferred out in a single time period shall not exceed the allowable transfer ratio of the current batch processing load of the microgrid; The amount received by the receiver shall not exceed the idle processing capacity of its server. The amount of data transferred between any two micronets in a single time period shall not exceed the data transmission link capacity.
[0017] Furthermore, after the hierarchical scheduling converges: The energy storage charging and discharging power request obtained by solving the shortest time scale is sent to the energy storage converter and a drive signal is generated. The power request for inter-micronet transactions is sent to the bidirectional power flow controller at the public connection point; The main grid switching power request is sent to the grid-connected converter; After converting the migration load into a task migration quantity, it is sent to the data center task scheduling system; the carbon quota trading volume is then sent to the third-level carbon trading market interface to execute the actual trading operation. The final physical control command is generated through closed-loop feedback.
[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention uses carbon market volatility signals as input for the energy storage capacity allocation correction intensity, enabling the energy storage capacity allocation to fully respond to power imbalances when the carbon market is stable, and automatically converge the correction amplitude when the carbon market fluctuates sharply, avoiding frequent switching of energy storage charging and discharging strategies caused by short-term noise in carbon prices, and achieving a dynamic balance between power supply and demand response and carbon financial risk avoidance. This invention converts the additional cooling energy consumption generated by the migration load into carbon emission increment cost based on the spot price of internal carbon quotas and includes it in the migration cost. This makes the migration of computing load not only take into account the difference in electricity price, but also the heat dissipation capacity and carbon cost level of the receiving microgrid. When the carbon price changes, it automatically selects the microgrid with lower temperature or higher cooling efficiency as the migration target. This invention embeds carbon futures forward price signals, carbon option implied volatility signals, and carbon spot real-time price signals into a hierarchical scheduling system with progressively decreasing time scales. It uses forward carbon prices to correct global capacity allocation, implied volatility to adjust regional coordination penalty parameters, and spot real-time prices to correct the charging and discharging tracking deviation weights of local tracking. This allows carbon financial market information to be transmitted step by step among the three levels of global allocation, regional coordination, and local tracking, forming a closed loop of carbon and electricity coordinated scheduling from forward guidance to real-time correction. Attached Figure Description
[0019] Figure 1 A flowchart illustrating the optimization method provided by this invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0022] Example: Please see Figure 1The present invention provides a technical solution: It should be noted that this embodiment is applied to a cluster collaborative system consisting of N data center microgrids interconnected through a common connection point. Each data center microgrid is equipped with renewable energy units such as wind turbine generators and photovoltaic power generation arrays, server clusters containing interactive loads and batch processing loads, a centralized shared energy storage system, and a three-level carbon trading market interface.
[0023] Interactive workloads refer to latency-sensitive computing tasks, such as web page access and real-time communication, which must be processed immediately and locally, without the possibility of migration in time or space. Batch processing workloads refer to latency-tolerant computing tasks, such as data backup and model training, which can be completed within a preset maximum allowable latency time and have flexibility in terms of time and space migration. The three-tier carbon trading market interface connects to the internal carbon quota spot trading market, the external carbon futures trading market, and the carbon options trading market, respectively. Each data center microgrid obtains real-time price and volatility data from the carbon market through this interface and executes carbon quota trading instructions.
[0024] This embodiment provides a method for jointly optimizing shared energy storage and computing load in a data center microgrid cluster, the specific steps of which include: Step 1: Obtain the carbon market volatility signal and determine the correction strength for energy storage capacity allocation based on the carbon market volatility signal, wherein the correction strength decreases when the carbon market volatility signal increases.
[0025] In this embodiment, the carbon market volatility signal originates from the three-tiered carbon trading market linkage mechanism. The three-tiered carbon trading market comprises an internal carbon allowance spot trading market, an external carbon futures trading market, and a carbon options trading market. Each microgrid sequentially performs allowance sharing, futures hedging, and options hedging through the three-tiered carbon trading market interface. In the internal carbon allowance spot market, each microgrid obtains an initial free carbon allowance based on its historical emissions. Initial free carbon allowances are allocated based on historical emissions percentages, with a corresponding reduction in allowance percentages for microgrids with higher historical emissions. The carbon emissions generated by each microgrid when purchasing electricity from the main grid during time period t are also considered. Calculate using the following formula: in, The carbon emissions (tons of carbon dioxide) of microgrid i during time period t. The carbon emission factor (tons of carbon dioxide / megawatt-hour) for purchasing electricity from the main grid is set to 0.85 in this embodiment, with reference to the average carbon dioxide emission factor of the regional power grid in China. The power (megawatts) that microgrid i purchases from the main grid during time period t. For the duration of the time period (hours), the quota surplus or deficit of microgrid i (tons of carbon dioxide) are calculated using the following formula: when A positive value indicates a quota surplus, while a negative value indicates a quota deficit. Microgrids use a unified clearing mechanism to balance quotas: microgrids with a surplus sell quotas, while microgrids with a deficit buy quotas, with a unified clearing price. The price (yuan / ton) is determined by the price quoted by each microgrid and the supply-demand difference. The sum of the net purchase volume within all microgrids is zero. And internal net buying volume The absolute value does not exceed the absolute value of the microgrid's quota profit or loss. Internal carbon quota spot price. As the shortest timescale, it serves as a real-time spot price signal for carbon.
[0026] In the carbon futures market, microgrids with a quota shortfall buy futures contracts, while microgrids with a quota surplus sell futures contracts. The same microgrid cannot simultaneously buy and sell futures contracts. The carbon futures forward price curve is determined by the futures contract prices in the carbon futures market, and the carbon futures forward price signal... As the longest-term carbon price signal, carbon options markets use call or put options to hedge against carbon price volatility. The strike price is set slightly out-of-the-money to reduce option premium costs. (Carbon option implied volatility) This signal, derived from carbon options market data, serves as the implied volatility signal for carbon options. Price signals from the three-tiered carbon trading market are transmitted sequentially through a risk premium transfer function, using the internal carbon quota spot price as the benchmark. Based on this, plus the term structure premium of carbon futures And carbon option volatility premium To obtain the comprehensive risk premium carbon price : in, The term structure premium for carbon futures (in yuan / ton) represents the premium or discount of the forward price of carbon futures relative to the spot price. The carbon option volatility premium (in yuan / ton) represents the deviation of implied volatility from historical volatility. The risk market price coefficient reflects the unit price of volatility risk in the market; in this embodiment, it is preset to 0.5. Historical volatility is calculated using the settlement prices of the past 30 trading days. The weight of the carbon futures term structure premium is preset to 0.3 in this embodiment; The carbon option volatility premium weight is preset to 0.2 in this embodiment, and both can be adjusted according to actual market conditions. The technical effect of the risk premium transfer function is that: the internal carbon quota spot price reflects the immediate supply and demand relationship of carbon quotas; the carbon futures term structure premium reflects the market's expectation of the future carbon price term structure; and the carbon option volatility premium reflects the market's pricing of the uncertainty of future carbon prices. The three factors combined form a comprehensive risk premium carbon price, enabling scheduling decisions to simultaneously perceive carbon price information from three dimensions: immediate supply and demand, forward expectations, and volatility risk. Transformed into a dynamic boundary carbon price through nonlinear saturation mapping : in, and These are the lowest and highest historical prices in the national carbon market (yuan / ton), respectively. The bias (yuan / ton) is the midpoint between the base carbon price and the value of carbon futures. The basis and technical effect of this mapping are: using a sigmoid-type saturation function to monotonically map the comprehensive risk premium carbon price to... Within the range, the dynamic boundary carbon price has both upper and lower bounds, and can smoothly transition when the comprehensive risk premium carbon price changes, avoiding distorted boundary carbon prices produced by linear mapping under extreme carbon prices. The final effective carbon cost of each microgrid. The (yuan) consists of the internal carbon quota spot trading cost, carbon futures trading cost, and net carbon option cost: in, and These are the carbon futures buying and selling prices (in yuan / ton), respectively. and These are the carbon futures purchase and sale volumes (tons of carbon dioxide) of Microgrid i, respectively. The net cost of the option (in RMB) is determined by subtracting the option's expiration profit from the total option premium paid. The option premium is priced by the carbon options market based on factors such as implied volatility, strike price, and time to expiration. For any remaining quota shortfall after internal trading and futures hedging, the microgrid purchases quotas from the external carbon spot market to make up the shortfall, and applies a secondary penalty to any unfilled shortfall. : in, The actual quota shortfall (tons of carbon dioxide) for microgrid i after carbon futures hedging. This represents the actual purchase volume (tons of carbon dioxide) of microgrid i in the external carbon spot market; the coefficient 0.05 is the penalty coefficient. The construction basis and technical effect of the secondary penalty are as follows: the penalty amount is proportional to the square of the unfilled gap; the larger the gap, the faster the penalty increases. This creates increasing pressure on the quota gap in scheduling decisions, prompting microgrids to actively fill the quota gap through internal mutual assistance, futures, and options trading, thereby reducing the overall compliance cost. The internal carbon quota spot price generated by the above three-tier carbon trading market linkage mechanism... Carbon futures forward price signals Implied volatility of carbon options These signals are used as the real-time spot price signal, the forward price signal, and the implied volatility signal for carbon options in subsequent steps, respectively. In this embodiment, the carbon market volatility signal is obtained by a weighted combination of carbon futures price volatility and carbon option implied volatility.
[0027] In this embodiment, the annualized volatility of carbon futures prices is calculated using the daily settlement prices of carbon futures over the past 30 trading days, reflecting the degree of price volatility in the carbon futures market. The annualized volatility is expressed in a dimensionless decimal form, for example, 0.15 represents 15%. The implied volatility of carbon options is the average of the implied volatility of carbon options over the same time window, reflecting the options market's expectations of uncertainty regarding future carbon prices.
[0028] The implied volatility of carbon options is obtained from market data in the carbon options trading market through an implied volatility inverse calculation model. This inverse calculation model takes the option market price, strike price, time to expiration, and risk-free interest rate as inputs, and uses numerical methods to solve for the volatility value that makes the theoretical option price consistent with the actual market transaction price. That is, based on the option pricing model, the implied volatility is treated as an unknown quantity, and the volatility is obtained by solving the equation that the theoretical option price equals the actual market transaction price. The risk-free interest rate is taken as the yield of the corresponding government bonds.
[0029] The specific calculation method for carbon market volatility signals is as follows: in, This represents the carbon market volatility signal for time period t. For carbon futures price volatility; This represents the average implied volatility of carbon options. and For the weighting coefficients, satisfying .
[0030] Weighting coefficient and The value of carbon futures price volatility is based on the following: carbon futures price volatility reflects realized historical volatility, the data source is direct and the calculation is stable, but there is a certain lag; carbon option implied volatility reflects the market's forward-looking expectations of future carbon price uncertainty, and the response to sudden events is more timely, but it is greatly affected by the liquidity of the options market and investor sentiment.
[0031] Taking into account the complementary characteristics of both, this embodiment presupposes , This ensures that carbon market volatility signals maintain the stability of historical data while also incorporating forward-looking information. In other embodiments, if the carbon options market has high liquidity and the implied volatility data is of good quality, the implied volatility signal can be appropriately increased. Up to 0.5; if the carbon options market is sluggish and implied volatility data is noisy, then it can be reduced. Adjust accordingly to 0.3 .
[0032] The correction strength is set at a baseline value when the carbon market volatility signal is below a low threshold, at a minimum value when the carbon market volatility signal is above a high threshold, and monotonically decreases as the carbon market volatility signal increases when the carbon market volatility signal is between the low threshold and the high threshold.
[0033] Corrected strength coefficient The piecewise mapping function is as follows: in, Based on the correction strength coefficient, For the minimum corrected strength coefficient, For low threshold, The threshold is high.
[0034] The values and calibration methods for each preset parameter are as follows: low threshold and high threshold The volatility was determined by referencing the statistical distribution of historical volatility in the carbon market. Under normal market conditions, the annualized volatility of carbon prices is typically between 15% and 25%, while a volatility exceeding 35% is considered extreme.
[0035] Therefore, this embodiment presupposes , When the carbon market volatility signal is below 0.15, the carbon market operates smoothly, and the predictability of carbon price signals is relatively high. At this time, the correction intensity coefficient takes its maximum value, and the energy storage capacity allocation fully responds to real-time power imbalances. When the carbon market volatility signal is above 0.35, the carbon market is in a state of violent fluctuation, and the predictability of short-term carbon price trends decreases significantly. At this time, the correction intensity coefficient takes its minimum value, and the real-time adjustment range of energy storage capacity is limited to the greatest extent. When the carbon market volatility signal is between 0.15 and 0.35, the correction intensity coefficient decreases linearly with the increase of volatility, and the adjustment range of energy storage capacity gradually narrows.
[0036] Basic correction strength coefficient The value of this coefficient is based on the fact that it determines the magnitude of the response of energy storage capacity allocation to real-time power imbalances under stable carbon market conditions. If... If the value is too large, the energy storage capacity allocation will be overly sensitive to short-term power fluctuations, easily triggering allocation oscillations; if... If the value is too small, the real-time correction stage will not respond adequately to power imbalances, making it difficult to perform the rolling correction function. This embodiment presupposes... This ensures that the correction amount has a moderate response range under stable carbon market conditions.
[0037] Minimum Correction Strength Coefficient The value of this coefficient is based on the fact that it determines the minimum corrective capability that energy storage capacity allocation can retain under conditions of severe carbon market volatility. If If the value is zero, energy storage capacity allocation will be completely frozen during periods of severe carbon market volatility, making it impossible to respond to any real-time power imbalances, which may lead to severe power deficits or surpluses in individual microgrids; if If the value is too large, it will be unable to effectively suppress the interference of short-term carbon price noise on energy storage allocation strategies. This embodiment assumes... This allows energy storage capacity allocation to retain approximately 15% of its base adjustment capability during periods of severe carbon market volatility, thus avoiding a complete freeze and effectively suppressing the interference of carbon price noise.
[0038] This step uses carbon futures price volatility and implied volatility of carbon options as input signals for the correction strength coefficient, enabling energy storage capacity allocation to not only respond to changes in electricity supply and demand but also proactively respond to risk signals from the carbon financial market. When carbon market volatility increases, the technical effect of attenuating the correction strength is that the predictability of short-term carbon price movements decreases. If energy storage capacity allocation continues to be adjusted with a large correction strength at this time, frequent switching of energy storage charging and discharging strategies can easily be triggered by short-term noise in carbon price signals, leading to wear and tear on energy storage equipment and scheduling oscillations. By allowing the correction strength to decrease with increasing volatility, the energy storage allocation strategy automatically becomes more conservative during periods of sharp carbon market fluctuations and regains full responsiveness during periods of calm, thereby achieving a dynamic balance between electricity supply and demand response and carbon financial risk avoidance.
[0039] Step 2: The initial energy storage capacity allocation plan is rolled over and adjusted using the adjustment intensity to obtain the energy storage power capacity and energy storage capacity allocated to each microgrid.
[0040] In this embodiment, the initial energy storage capacity allocation plan is based on the weighted sum of the proportion of the day-ahead predicted load of each microgrid in the cluster and the proportion of the day-ahead predicted renewable energy output in the cluster, forming the basic weight of each microgrid.
[0041] In this embodiment, during the day-ahead phase 24 hours before the start of the scheduling cycle, the predicted load is calculated based on historical data compiled by each microgrid. With forecasts of renewable energy output Calculate the basic weights of each microgrid : in, As the load weighting factor, As a weighting factor for renewable energy, satisfying .
[0042] Load weighting factor This reflects the principle of fairness that microgrids with high load demand should receive more energy storage and discharge support to meet their electricity needs; renewable energy weighting factor This reflects the low-carbon guiding principle that microgrids with high renewable energy output should receive more energy storage and charging capacity to absorb their excess output.
[0043] In the allocation of energy storage capacity, the need to ensure the reliability of electricity supply to the load takes a slight priority over the low-carbon need to promote the consumption of renewable energy. It should be slightly larger than This embodiment is pre-set. , Depending on the policy guidance in different regions, if renewable energy resources are abundant but the pressure to absorb them is significant, the allowance can be appropriately increased. Values of 0.5 or higher will decrease accordingly. value.
[0044] A proportion reserved from the total shared energy storage capacity As a system-level spinning reserve capacity, a balance is struck between ensuring security and avoiding excessive use of available capacity. The reserve ratio is typically set at a midpoint between 5% and 15% of the total system capacity, as per industry practice. This embodiment presupposes... The power capacity allocated recently. and energy capacity They are respectively: in, To share the total power capacity of the energy storage system, To share the total energy capacity of the energy storage system, This represents the percentage of usable capacity after deducting reserved reserves; the unit for total power capacity is kilowatts (kW), and the unit for total energy capacity is kilowatt-hours (kWh). During the real-time phase, the actual total load of each microgrid is collected. The actual renewable energy output is calculated based on the sum of wind power output and photovoltaic (PV) power output. Wind power output is calculated according to the measured wind speed and the turbine power characteristic curve. Output is zero when the wind speed is below the cut-in wind speed or above the cut-out wind speed. Output is calculated linearly between the cut-in wind speed and the rated wind speed, and outputs at rated power between the rated wind speed and the cut-out wind speed. PV power output is calculated based on measured irradiance, ambient temperature, and the temperature correction coefficient of the PV modules. A rolling correction is used, adjusting the weights based on the net load of the current period and using exponential smoothing by incorporating the comprehensive weights from the previous period to suppress weight oscillations and avoid over-correction caused by relying solely on the current instantaneous value. Positive net load indicates the power deficit of the microgrid, while negative net load indicates the power surplus of the microgrid. The absolute value of these values reflects the urgency of the microgrid's demand for energy storage charging and discharging.
[0045] The adjustment weight for each microgrid is determined by its proportion of the absolute value of the actual net load to the absolute value of the total net load of the cluster. The specific calculation method is as follows: when At this time, it indicates that all microgrid power is balanced and there is no urgent need for energy storage. That is, the weights are adjusted and distributed equally among the microgrids.
[0046] The energy storage capacity allocation correction amount is determined by the difference between the adjustment weight and the basic weight and the correction intensity. The energy storage capacity allocation correction amount includes the power capacity correction amount and the energy capacity correction amount. Both the power capacity correction amount and the energy capacity correction amount increase monotonically with the increase of the absolute value of the actual net load, and the growth rate decreases and approaches saturation.
[0047] Power capacity correction and energy capacity correction The specific calculation method is as follows: in, The power reference value is as follows: That is, the average value of the absolute value of the net load of all microgrids is taken and used to normalize the correction range so that the dimension of the correction amount is consistent with that of power.
[0048] Saturation function The construction basis and technical effect of this function are as follows: when the adjustment weight is greater than the base weight, the correction amount increases monotonically with the increase of the absolute value of the net load and the growth rate decreases and approaches saturation; when the adjustment weight is less than or equal to the base weight, the correction amount decreases monotonically with the increase of the absolute value of the actual net load and approaches zero.
[0049] The reason for using an exponential saturation function instead of a nonlinear function is that a linear function can lead to an unbounded increase in the correction amount when the net load increases to an extreme extent, which may cause over-adjustment of energy storage capacity allocation; an exponential saturation function is approximately linear when the net load is small, can respond sensitively to power imbalance, and automatically converges when the net load is large, avoiding over-correction under extreme conditions, thereby ensuring the robustness of the energy storage capacity allocation strategy.
[0050] Corrected strength coefficient Its role in the correction is that this coefficient, as an overall scaling factor, directly controls the response magnitude of the real-time correction to power imbalances. When carbon market volatility increases... The reduction in volume and the narrowing of overall corrections indicate a more conservative approach to energy storage capacity allocation; when carbon market volatility decreases... Increase, corrective measures are restored, and energy storage capacity allocation fully responds to real-time power imbalance.
[0051] The comprehensive weights of the previous period and the current adjustment weights are smoothed and then normalized to obtain the final comprehensive weights of each microgrid. The final comprehensive weights of the previous period are the final comprehensive weights obtained in the previous period using the same smoothing and normalization method. The final energy storage power capacity and energy storage capacity allocated to each microgrid are determined by the final comprehensive weights, and the sum of the allocated capacities of all microgrids is equal to the total available capacity after deducting the system-level spinning reserve capacity from the total shared energy storage capacity.
[0052] Calculate the target comprehensive weights that require smoothing suppression. Specifically: Exponential smoothing is used to suppress weight oscillations, and the unnormalized comprehensive weights are calculated. : in, For smoothing coefficients, This represents the final overall weight for the previous period.
[0053] Smoothing coefficient The basis for its value and its effect are as follows: The larger the value, the more the overall weight of the current period depends on the historical information of the previous period, the smoother the weight changes, but the slower the response speed. The smaller the value, the more the overall weight of the current period depends on the adjustment weight of the current period, the faster the response speed, but the more susceptible it is to interference from instantaneous power fluctuations.
[0054] This embodiment is preset This ensures that the overall weight of the current period mainly depends on the adjusted weight of the current period, while retaining less historical information from the previous period. The allocation strategy can respond quickly to changes in the system state, while retaining a certain degree of smoothing effect to suppress weight oscillations.
[0055] The smoothed composite weights are then normalized to obtain the final composite weights for each microgrid. : The final energy storage capacity allocated to microgrid i in time period t and energy storage capacity is for: because Therefore, it satisfies , That is, the sum of the capacity allocated to all microgrids equals the total available capacity after deducting the system-level spinning reserve capacity from the total shared energy storage capacity.
[0056] This step uses the carbon market volatility signal as the input signal for the correction strength coefficient, enabling energy storage capacity allocation to not only respond to changes in electricity supply and demand, but also proactively respond to risk signals from the carbon finance market, thus achieving the integration of the dual objectives of responding to electricity supply and demand and avoiding carbon finance risks.
[0057] Step 3: Based on the energy storage power capacity and energy storage capacity, determine the migration direction and migration load of the batch processing load. The additional cooling energy consumption required for the additional heat generation after migration to the receiving party is converted into carbon emission increment cost based on the internal carbon quota spot price. This carbon emission increment cost is included in the migration cost.
[0058] In this embodiment, the migration direction and migration load amount of the batch processing load are determined by coupling the migration load amount with the temperature change of the receiving server cluster, and the migration load amount does not exceed the thermal response capability of the receiving cooling system.
[0059] In this embodiment, migration power is defined. Let t be the batch load power migrating from microgrid i to microgrid j, and let the migration power satisfy antisymmetry. This means that the net migration amounts between the same pair of microgrids are equal in size and opposite in direction. This constraint ensures that the migration amounts can be transmitted bidirectionally in physical terms and that there will be no invalid migrations between the same pair of microgrids.
[0060] The temperature change of the receiving server cluster is updated based on the recursive relationship between the temperature of the previous period, the difference between server heat generation and cooling, heat capacity, and time interval.
[0061] The specific changes in server cluster temperature are represented as follows: in, This indicates the server cluster temperature of data center micronet j during time period t; This indicates the server cluster temperature in the previous time period; This indicates a time interval; in this embodiment, it is taken as 1 hour. The heat capacity of data center micronet j reflects the magnitude of temperature change inertia. It is estimated by the building area and total mass of servers in each micronet data center. The larger the heat capacity, the slower the temperature rises for the same amount of heat generated. The heat generation coefficient of the server represents the conversion ratio of electrical energy to heat energy, which is determined by the physical characteristics of the server chip. Based on the average coefficient of server chips on the market, this embodiment sets it to 0.92. This represents the total server power of data center micronet j during time period t, which is the sum of interactive load and actual batch processing load. This indicates the cooling power of the server cooling system in data center micronet j during time period t, which is affected by the rated power of the actual configuration.
[0062] The above temperature recursion is based on the principle that the temperature change of the server cluster follows the law of energy conservation and thermal balance. The heat generated by the conversion of electrical energy from the server minus the heat dissipation of the cooling system equals the product of the server cluster's heat capacity and the rate of temperature change. Discretizing the continuous differential relationship according to the scheduling period yields the above recursive formula. When heat generation exceeds heat dissipation, the temperature rises; the larger the heat capacity, the slower the temperature rises. In the above parameters, the physical unit of temperature is degrees Celsius, the physical unit of heat capacity is kilojoules per degree Celsius, the physical unit of total server power and cooling power is kilowatts, and the physical unit of time interval is hours.
[0063] The heat generated by the server is determined by the heat generation coefficient of the server's total power. This migrating load increases the server's total power and thus its temperature. When the migrating load reaches the receiving micronet, it will increase the total processing power of the micronet's servers, thereby increasing heat generation and causing the server cluster temperature to rise. If the temperature rise exceeds the safe limit or the rate of rise exceeds the cooling system's response capability, it may lead to server overheating and shutdown.
[0064] When the temperature reaches the safety limit or the rate of temperature rise exceeds the thermal response capability of the receiver's cooling system, the receiver will reject the migration load.
[0065] In this embodiment, the upper limit of temperature safety is determined based on the maximum allowable operating temperature of the server chip, which is preset to 35 degrees Celsius. The upper limit of temperature rise rate is determined based on the thermal response capability of the cooling system, that is, the temperature rise rate corresponding to the additional heat that the cooling system can remove per unit time. When the temperature of the receiving server cluster is close to the upper limit of safety, or when the temperature rise rate caused by the migration load exceeds the thermal response capability of the cooling system, the receiving party rejects the migration load to ensure the safe operation of the server.
[0066] The additional cooling energy consumption resulting from the relocation is converted into the incremental cost of carbon emissions based on the spot price of the internal carbon quota, the carbon emission coefficient of the main grid electricity purchase, and the energy efficiency ratio of the cooling system.
[0067] The additional cooling energy consumption generated by the migration is determined by the migration load, the heat generation coefficient, the energy efficiency ratio of the cooling system, the carbon emission coefficient of the main grid electricity purchase, and the spot price of the internal carbon quota.
[0068] The specific method for calculating the incremental cost of carbon emissions is as follows: in, The current spot trading price of internal carbon allowances is determined by the unified clearing mechanism of the internal carbon allowance spot market. The carbon emission factor for electricity purchased from the main grid is set to 0.85, referencing the average carbon dioxide emission factor of regional power grids in China. The heat generation coefficient of the server is preset to 0.92 in this embodiment; This represents the total load power migrated to microgrid j; The energy efficiency ratio of the refrigeration system is affected by the actual configuration of the refrigeration system. Referring to the typical energy efficiency ratio of most general refrigeration systems, it is preset to 3.5 in this embodiment.
[0069] The step-by-step construction of the above carbon emission increment cost conversion formula is as follows: First, the migrated load to the receiving party is converted into additional heat generation using the heat generation coefficient; second, the additional heat generation is divided by the energy efficiency ratio of the cooling system to obtain the cooling power consumption required to offset this heat generation; third, the cooling power consumption is multiplied by the time period duration to obtain the cooling power consumption; fourth, the cooling power consumption is multiplied by the carbon emission coefficient of the main grid electricity purchase to obtain the corresponding carbon emission amount; fifth, the carbon emission amount is multiplied by the internal carbon quota spot price to obtain the carbon emission increment cost. The effect of this step-by-step construction is that the heat load of the migrated load is transmitted to the carbon economic cost step by step through heat generation, cooling, power consumption, and carbon emissions, so that the carbon price forms an economic constraint on computing power migration through temperature constraints.
[0070] When selecting migration targets, priority is given to avoiding microgrids with high carbon emission increment costs or weak thermal response capabilities. In this embodiment, when the internal carbon quota spot price increases, the carbon emission increment cost also increases, and the optimized scheduling will automatically tend to select microgrids with lower temperatures or higher cooling efficiency as migration targets. Microgrids with weak thermal response capabilities refer to microgrids with small heat capacity or low cooling system heat dissipation power. These microgrids experience a rapid temperature rise after receiving the migrated load, which can easily trigger temperature safety constraints.
[0071] The migration direction and amount of batch processing load are also subject to the following constraints: the accumulated amount of queued load is updated according to the recursive relationship of the remaining amount of the previous period, newly arrived load, local processing load, outgoing load and incoming load, and it is guaranteed that each batch processing load will be processed within a specified time limit after arrival. The specified time limit is set according to the delivery time limit agreed upon in the service level agreement.
[0072] The specific equation for the dynamic accumulation of queuing load is as follows: in, The cumulative queued load of microgrid i at the end of time period t represents the batch processing load in microgrid i that has arrived but has not yet been processed locally and has not been transferred out, converted to kilowatt-hours by power integral over time; This represents the accumulated queue load at the end of the previous time period. The batch load power arriving at time period t; The time interval is 1 hour in this embodiment; This refers to the load power of batch processing handled locally. This refers to the amount transferred out. This refers to the inbound volume. The time window constraint ensures that any batch workload must be processed within a specified time limit after it arrives, guaranteeing the basic service quality of the data center.
[0073] The amount transferred out in a single time period shall not exceed the allowable transfer ratio of the batch processing load of the micronet. The batch processing load transferred out by each micronet in a single time period shall not exceed the maximum ratio of the batch processing load of the micronet. The maximum ratio can be determined with reference to the engineering experience of data center load scheduling. In this embodiment, it is preset to 0.3, that is, the amount transferred out in a single time period shall not exceed 30% of the current total batch processing load of the micronet.
[0074] The receiving load cannot exceed the idle processing capacity of its servers, and the transferred load received by each micronet cannot exceed the idle capacity equal to the difference between the maximum processing power of the micronet's server cluster and the current actual load. This constraint ensures that the migrated load will not cause the receiving server to exceed its maximum processing capacity. The mathematical expression of this constraint is: in, This is the maximum processing power (kilowatts) of the microgrid i server cluster. This represents the actual total load (kW) of microgrid i during time period t.
[0075] The amount of data transferred between any two micronets in a single time period shall not exceed the data transmission link capacity.
[0076] The power transfer between any two micronets in a single time period shall not exceed the maximum transmission power of the data transmission link. The maximum transmission power can be determined by referring to the typical bandwidth capacity of a direct fiber optic link between data centers and adjusted according to the actual network configuration. The mathematical expression of this constraint is: in, This represents the maximum transmission power (kilowatts) of the data transmission link between micronet i and micronet j, with reference to the typical bandwidth capacity of a direct fiber optic link between data centers.
[0077] This step establishes a complete cross-domain transmission chain from thermodynamic physical constraints to cooling power consumption, then to carbon emissions, and finally to carbon economic costs by converting the additional heat load of the relocated load into incremental carbon emission costs and linking these costs to the internal carbon spot price in real time. When the internal carbon price rises, the constraint of temperature on the relocation scheme is amplified through economic costs, and the optimization model will automatically tend to select microgrids with lower temperatures or higher cooling efficiency as relocation targets.
[0078] Step 4: Perform hierarchical scheduling according to multiple time scales in descending order. Among them, the carbon futures forward price signal is used to correct the global capacity allocation, the carbon option implied volatility signal is used to adjust the regional coordination penalty parameter, and the carbon spot real-time price signal is used to correct the charging and discharging tracking deviation weight of local tracking.
[0079] The optimization problem solved by the hierarchical scheduling in this step is a mixed-integer linear programming problem with the objective of minimizing the overall operating cost of the cluster, and with constraints on the power and computing power operation of each device and carbon trading constraints. The objective function is: The meaning and composition of each cost item are as follows: Main Grid Electricity Purchase Cost (Yuan) represents the difference between the cost of purchasing electricity and the revenue from selling it. in, and These are the time-of-use electricity purchase price and the time-of-use electricity sales price (yuan / kWh), which are determined by the time-of-use electricity pricing policy of the superior power grid. and These represent the electricity purchased and sold (in kilowatts) by microgrid i during time period t. Energy storage usage cost. (Yuan) represents the sum of the cost of energy storage capacity occupancy and the cost of charging and discharging losses: in, and These are the unit power capacity occupancy cost and unit energy capacity occupancy cost (yuan / kW, yuan / kWh), respectively, which are calculated from the investment discount and operation and maintenance costs of the shared energy storage system. The unit charge / discharge loss cost (yuan / kW) reflects the energy loss during the energy storage charge / discharge process. and These represent the energy storage charging power and discharging power (kW), respectively. Microgrid inter-transaction costs. (Yuan) represents the sum of the settlement cost and grid access fee for inter-microgrid electricity transactions: in, This is the settlement price for inter-microgrid transactions (yuan / kWh), which falls between the on-grid electricity price and the purchase price. The grid access fee (yuan / kWh) is determined with reference to the provincial transmission and distribution price. The power transaction (kW) between microgrid i and microgrid j during time period t. Batch load hopping cost. (Yuan) represents the sum of the unit transfer base cost and the incremental cost of carbon emissions driven by temperature constraints: in, The unit transfer base cost (RMB / kW) includes network transmission energy consumption and additional overhead for computing power scheduling, and is shared with reference to the bandwidth rental cost of data transmission between data centers and the operation and maintenance cost of cross-data center scheduling systems; The incremental cost of carbon emissions driven by temperature constraints in the aforementioned S300 (RMB). Service level agreement breach cost. (RMB) represents the penalty for failure to process batches exceeding the threshold due to untimely processing, calculated based on the average compensation standard under the service level agreement. in, The average compensation standard for service level agreements is preset to 2.0 in this embodiment; For batch load power (kW) that was not processed on time; The allowable unprocessed threshold is set at 5% of the total batch processing load of the microgrid. The basis and technical effect of this default cost are as follows: default cost is only calculated when the unprocessed load exceeds the allowable threshold, thus reserving a certain scheduling flexibility for the delayed processing of batch processing load. At the same time, a penalty that increases linearly with the excess is imposed on serious defaults, prompting the scheduler to prioritize ensuring the time limit requirements of batch processing load.
[0080] The final effective carbon cost (in yuan) determined by the aforementioned three-tier carbon trading linkage mechanism. The carbon risk value cost (in yuan) is measured using the conditional risk value index to assess the uncertainty risk of carbon prices. Conditional risk value is obtained by generating multiple carbon price scenarios and calculating the conditional expectation of the excess carbon cost in each scenario. Its construction basis and technical effect are as follows: ordinary expected cost only reflects the average carbon price level, while conditional risk value additionally measures the tail risk under extremely unfavorable carbon price scenarios. This allows the objective function to impose a budget constraint on the worst-case scenario when carbon prices fluctuate significantly, avoiding high compliance costs for scheduling schemes when carbon prices rise sharply.
[0081] The constraints of the mixed-integer linear programming problem include: power balance constraints of each microgrid, charging and discharging power and state of charge constraints of the energy storage system, and integer variables of the energy storage charging and discharging state. and Mutual exclusion constraints, namely, the ability of storage devices to charge and discharge simultaneously within the same time period, the range constraints of power trading between microgrids and power exchanged with the main grid, the three-dimensional constraints of batch load migration in S300, and the aforementioned three-level carbon trading constraints. Integer variables. and The charge / discharge state of energy storage is characterized by a value of 0 or 1, which are mutually exclusive to ensure the physical realizability of the charge / discharge state. In this embodiment, the correspondence between the three time scales and the three carbon price signals in the hierarchical scheduling is as follows.
[0082] The longest timescale uses the carbon futures forward price signal to correct the global cost of shared energy storage capacity allocation; the medium timescale uses the carbon option implied volatility signal to determine the initial value of the regional coordination penalty parameter; and the shortest timescale uses the carbon spot real-time price signal as a correction term for the charge and discharge tracking deviation weight in local tracking optimization.
[0083] The three time scales are updated sequentially in a decreasing direction. In this embodiment, the three time scales are as follows: the first level performs global capacity allocation at a first preset time interval, which is 1 hour in this embodiment; the second level performs regional coordination at a second preset time interval, which is 15 minutes in this embodiment; and the third level performs local rolling tracking at a third preset time interval, which is 5 minutes in this embodiment. The three time scales are updated sequentially in a decreasing direction, that is, the first level updates once every 1 hour, the second level updates once every 15 minutes, and the third level updates once every 5 minutes.
[0084] The first level performs global capacity allocation at a first preset time interval, and applies a correction term to the ratio of the carbon futures forward price signal to the reference carbon price to the effective usage cost of the shared energy storage power capacity and energy storage capacity.
[0085] In this embodiment, the first level performs global capacity allocation on a 1-hour cycle, solving the main problem with the objective of minimizing the sum of shared energy storage capacity usage costs and proxy variables. The specific way the ratio of the carbon futures forward price signal to the reference carbon price affects the effective usage cost is as follows: in, and These are the effective usage costs of shared energy storage power capacity and energy storage capacity, respectively. and These are the basic unit usage costs for shared energy storage power capacity and energy storage capacity, respectively. The carbon price impact coefficient represents the degree to which changes in carbon prices affect the perceived cost of energy storage; in this embodiment, it is preset to 0.5. Let be the price of the carbon futures forward contract for time period t; For reference carbon prices, this example uses 50 yuan / ton.
[0086] Carbon price impact coefficient The value of this coefficient is based on the fact that it determines the extent to which the carbon futures forward price signal corrects for the effective cost of energy storage. If... If the value is too large, even small fluctuations in the carbon price signal can lead to significant changes in the perceived cost of energy storage, potentially causing oscillations in global capacity allocation; if... If the value is too small, the carbon price signal will not have a sufficient guiding effect on global capacity allocation. This embodiment assumes... This allows the carbon price signal to have a moderate correction range for the perception of energy storage costs.
[0087] Reference carbon price The value of this parameter is based on the fact that it serves as a normalized benchmark for carbon futures forward price signals, making the carbon price correction term dimensionless. The reference carbon price is the average transaction price of the national carbon market over the past year, which is updated on a rolling basis. This reflects the long-term central level of carbon prices and avoids the correction term being systematically too large or too small due to improper selection of the reference value.
[0088] The first level uses Benders decomposition to solve the global allocation scheme of shared energy storage capacity. The main problem aims to minimize the sum of the energy storage capacity usage cost and proxy variables for each microgrid in each time period. in, For Benders proxy variables, it represents the upper bound approximation of the optimal total cost returned by the regional aggregation layer; and Let be the energy storage power capacity and energy storage capacity allocated to microgrid i in time period t, respectively. After solving the main problem in each iteration to obtain the capacity allocation scheme, the scheme is distributed to the sub-problems of each microgrid in the regional aggregation layer for solution. Each microgrid sub-problem returns the optimal total cost. And the corresponding Lagrange multipliers, from which the main problem generates Benders cutting plane constraints: in, The optimal total cost returned by the region aggregation layer in the m-th iteration; and Let be the Lagrange multipliers corresponding to the power capacity and energy capacity constraints in the m-th iteration, and let represent the marginal cost of the optimal total cost for each capacity allocation; and The capacity allocation value given for the main problem in the m-th iteration.
[0089] The technical effect of tangent plane constraints is that by feeding back the cost information of the region aggregation layer to the main problem in the form of a linear tangent plane, the proxy variables are gradually tightened. The lower bound of the boundary value allows the main problem to approach the global optimum without directly solving the subproblems of each micronet. As the number of iterations increases, the main problem accumulates multiple cutting plane constraints. To control the problem size, a multi-cutting plane aggregation strategy is adopted, aggregating historical cutting planes according to their weights: in, The aggregation weights of the historical cutting plane, The preset baseline weight for this embodiment is 1.0. The attenuation coefficient for this embodiment is preset to 0.3. This represents the current iteration number. The iteration number corresponds to the historical cutting plane. The aggregated weights decay exponentially with increasing iteration distance; cutting planes closer to the current iteration have larger weights, and those farther away have smaller weights. In practical numerical computation, these aggregated weights can be calculated in the logarithmic field or normalized to the maximum value to avoid exponential underflow. The technical advantage of multi-cutting plane aggregation is that the earlier the historical cutting planes are generated, the lower their contribution to the current solution. Exponentially decaying weighting preserves effective cutting plane information while avoiding excessive interference from historical information in the current solution, balancing convergence and computational efficiency. The convergence condition for Benders decomposition is that the relative gap between the upper and lower bounds of the main problem is less than a preset tolerance.
[0090] The objective function value of the main problem is the lower bound of the globally optimal cost, while the actual total cost obtained by substituting the solution of the main problem into the regional aggregation layer is the upper bound. The iteration terminates when the relative gap between the two is less than a preset tolerance. When the long-term carbon price rises, the perceived cost of energy storage usage increases for microgrids, leading them to reduce their reliance on main grid electricity purchases and increase the intensity of energy storage usage. This, in turn, guides energy storage capacity towards low-carbon cost allocation at the macro level.
[0091] The second level performs regional coordination at the second preset time interval. The correction term of the ratio of the implied volatility signal of the carbon option to the reference volatility is used to determine the initial value of the regional coordination penalty parameter. Each microgrid solves the objective in parallel with the local cost term, the adjacent microgrid deviation term and the regional coordination penalty parameter until both the original residual and the dual residual satisfy the convergence condition.
[0092] In this embodiment, the second level performs regional coordination on a 15-minute cycle. The specific method for determining the initial value of the penalty parameter based on the ratio of the implied volatility signal of carbon options to the reference volatility is as follows: in, and These are the initial values of the penalty parameters for power trading and load transfer, respectively. The penalty parameter is a dimensionless scaling factor used to scale the magnitude of the deviation penalty term in the objective function of the subproblem. The volatility impact coefficient is preset to 2.0 in this embodiment; Implied volatility of carbon options; For reference volatility, this embodiment is preset to 0.20.
[0093] Volatility Impact Coefficient The value of this coefficient is determined by the extent to which the implied volatility of carbon options adjusts the initial value of the penalty parameter. If If the value is too large, the penalty parameter becomes overly sensitive to changes in volatility, potentially causing significant fluctuations in the convergence speed of regional coordination during carbon market volatility; if... If the value is too small, the volatility signal will not be able to adequately regulate regional coordination. This embodiment assumes... This allows the penalty parameter to adjust appropriately when carbon market volatility changes.
[0094] Reference volatility The value of this parameter is based on the fact that it serves as a normalized benchmark for the implied volatility of carbon options. It takes the annualized volatility center value of 0.20 under normal market conditions, so that the initial value of the penalty parameter is about 1.0 under normal market conditions, and increases accordingly when the volatility of the carbon market intensifies.
[0095] The higher the implied volatility of carbon options, the larger the initial value of the penalty parameter. The technical effect is that the larger the penalty parameter, the stronger the penalty for the deviation of adjacent microgrids in regional coordination. Each microgrid tends to converge to the consensus scheme more quickly, thereby accelerating the convergence speed of regional coordination when the carbon market is highly uncertain and reducing the delay in consensus coordination caused by carbon price fluctuations.
[0096] Each microgrid solves its objective function in parallel, comprising a local cost term, a neighboring microgrid deviation term, and a regional coordination penalty parameter. In this embodiment, each microgrid solves its own local operation optimization subproblem in parallel, under a predetermined capacity quota allocated globally. The objective function of the local subproblem includes a local operating cost term, a penalty term for power trading deviations with neighboring microgrids, and a penalty term for load transfer deviations with neighboring microgrids. The neighboring microgrid deviation term is composed of dual multipliers and a penalty parameter, used to penalize the deviation between each microgrid's proposed value and the global consensus value.
[0097] The second level employs the alternating direction multiplier method to solve the regional coordination problem, decomposing the global problem into local subproblems of each microgrid and solving them in parallel. The objective function of the local subproblem of microgrid i is: in, The local operating cost of microgrid i (in yuan); The set of microgrids adjacent to microgrid i; and These are dual multipliers for power trading and load shifting, used to penalize deviations between each microgrid's proposed value and the global consensus value. The power (kW) exchanged between microgrid i and microgrid j; and These are globally consistent variables for power trading and load transfer, respectively. and These are the penalty parameters for power trading and load shifting, respectively.
[0098] The convergence condition for the alternating direction multiplier method is that both the original residual and the dual residual are less than a preset tolerance. Original residual Dual residuals measure the deviation between the proposed values of each microgrid and the global consensus value. Measuring the magnitude of change in the global consensus value between two consecutive iterations: in, and These represent the power trading power and load transfer power proposed by microgrid i in the k-th iteration; and These are the globally consistent variables for the k-th iteration; and These are the global consensus variables for the (k-1)th iteration. The original residuals reflect the deviation of each microgrid's proposed value from the global consensus value, while the dual residuals reflect the change in the global consensus value between adjacent iterations. When both are less than a preset tolerance... At that time, the local solutions of each microgrid tend to be consistent, and regional coordination converges.
[0099] After all micronets have completed their local solutions, the global consensus variable, dual multipliers, and penalty parameters are updated. The iteration continues until both the original residual and the dual residual are less than a preset tolerance, at which point convergence occurs. The original residual measures the deviation between each micronet's proposed value and the global consensus value, while the dual residual measures the change in the global consensus value between two adjacent iterations. In this embodiment, the preset tolerance is [value missing]. .
[0100] The third level performs local rolling tracking at a third preset time interval, at the cost of the deviation between the actual output of the local equipment and the reference trajectory and the control increment, and applies the charge and discharge tracking deviation weight after the real-time carbon spot price signal is corrected to this cost.
[0101] In this embodiment, the third level performs local rolling tracking with a 5-minute cycle, receives the scheduling reference trajectory from the second level, and solves a quadratic programming problem in the prediction time domain that minimizes the deviation of the local equipment's actual output from the tracking reference trajectory and minimizes the control increment. The third level uses model predictive control to track the reference trajectory, and the quadratic programming objective function is: in, To predict the actual output (kilowatts) of the local equipment at the k-th step in the time domain. The scheduling reference trajectory (kilowatts) issued at the second level; To control the increment (kilowatts); A deviation weighting matrix is used to penalize deviations between actual output and reference trajectory; The incremental weight matrix is used to control the frequent changes in control actions. To predict the length of the time domain.
[0102] In this embodiment, the prediction time domain is 30 minutes, combined with the third-level control step size of 5 minutes, corresponding to 6 prediction step sizes, i.e. The reason for choosing a 30-minute prediction time domain is to predict system state changes within the next half hour while avoiding increased computational burden and accumulated prediction errors caused by an excessively long prediction time domain. (Tracking deviation weight matrix) With control increment weight matrix The trade-off between local tracking accuracy and control stability is determined as follows: relatively The larger the value, the more accurately the local device tracks the reference trajectory, but the more frequent the control actions become. relatively The larger the value, the smoother the control action, but the tracking deviation may increase. The tracking deviation weight of the energy storage charging and discharging power is as described above. and The tracking deviation weights and control increment weights of the remaining devices are selected according to the device response characteristics.
[0103] The specific method for correcting the charge / discharge tracking deviation weight in real-time carbon spot price signals is as follows: in, and These are the tracking deviation weights for energy storage charging power and discharging power, respectively; and The reference tracking deviation weight for energy storage charging and discharging power is set to 100 in this embodiment; The carbon price impact coefficient is preset to 0.5 in this embodiment; Real-time clearing price for internal carbon quota spot trading; For reference carbon price.
[0104] Benchmark tracking deviation weight and The value of this value is determined by the fact that this weight determines the required tracking accuracy of the local device's actual output relative to the reference trajectory. In this embodiment, the preset value is 100, which gives the tracking deviation a high weight in the objective function, ensuring that the local device can accurately execute the scheduling instructions issued by the upper layer.
[0105] When the real-time carbon price is high, a greater penalty is imposed on the tracking deviation of the energy storage discharge power, so that the local execution layer prioritizes the execution of energy storage discharge commands and avoids purchasing electricity from the main grid during periods of high carbon price.
[0106] In this embodiment, after each solution is completed, only the most recent control increment is issued and the process rolls forward. The third level only issues the first control action obtained from the solution to the device execution layer after each solution is completed, and then updates the prediction time domain and optimization problem in the next moment and solves the problem again. This rolling execution method can respond to changes in system state in a timely manner and avoid control deviations caused by the accumulation of prediction errors.
[0107] This step embeds price signals from three different time scales in the three-tiered carbon trading market into three different levels of hierarchical scheduling. The forward carbon price guides the global capacity allocation, the implied volatility adjusts the regional consistency constraint strength, and the real-time carbon price correction prioritizes local tracking. This enables the carbon financial market signals to penetrate and be transmitted in a differentiated manner across the three levels, forming a complete carbon and electricity coordinated scheduling link from macro-level forward guidance to micro-level real-time correction.
[0108] Step 5: Output and execute the energy storage charging and discharging power, main grid switching power, microgrid inter-trading power, migration load, and carbon quota trading volume.
[0109] In this embodiment, after the hierarchical scheduling converges, the following control operation is performed: the energy storage charging and discharging power request obtained by solving the shortest time scale is sent to the energy storage converter and a drive signal is formed.
[0110] In this embodiment, the energy storage charging and discharging power request is obtained by the third-level local rolling tracking solution and sent to the energy storage converter. The energy storage converter generates a drive signal based on the power request to control the charging and discharging power of the energy storage system.
[0111] In this embodiment, the inter-microgrid power trading request is obtained by the second-level regional coordination solution and sent to the bidirectional power flow controller at the common connection point. The bidirectional power flow controller controls the power trading power between each microgrid.
[0112] In this embodiment, the main grid switching power request is obtained by the second-level regional coordination solution and sent to the grid-connected converter. The grid-connected converter controls the power purchase and sales between each microgrid and the main grid.
[0113] After converting the migration load into a number of task migrations, it is sent to the data center task scheduling system. In this embodiment, the migration load is expressed in units of power and needs to be converted into a specific number of task migrations. The conversion method is as follows: define the average processing power of a single batch processing task. (kW / unit), determined by historical statistics from the data center task scheduling system, is calculated by dividing the migration load power by the average processing power of that single batch task to directly obtain the specific number of task migrations, which is then sent to the data center task scheduling system for execution. Number of Task Migrations Determined by the ratio of migration load power to the average processing power of a single batch task: The carbon quota trading volume is distributed to the interface of the tertiary carbon trading market to execute actual trading operations. In this embodiment, the carbon quota trading volume includes the internal carbon quota spot trading volume, carbon futures trading volume, and carbon options trading volume, which are distributed to the corresponding tertiary carbon trading market interfaces to execute actual trading operations.
[0114] As the underlying physical execution link for the above three time-scale optimized scheduling, the device response layer operates with a 1-second cycle and uses local closed-loop control to execute the final command. The controller of the device response layer adopts proportional-integral-derivative control. The controller input is the power setpoint issued by the upper layer, and the output control signal is amplified by the drive circuit and directly drives each device. The energy storage converter controls the charging and discharging power of the energy storage system by forming a duty cycle drive signal through pulse width modulation according to the energy storage charging and discharging power request; the grid-connected converter controls the power purchase and sales through the current inner loop according to the main grid switching power request; the bidirectional power flow controller controls the bidirectional power flow at the common connection point according to the inter-microgrid trading power request; the server cluster controls the number of active servers according to the server start and stop status. The number of active servers is the ratio of the total power of the servers to the rated power consumption of a single server. Unused servers enter soft shutdown or hibernation state to reduce standby power consumption and are remotely woken up when needed. The equipment response layer operates on a timescale far exceeding that of the upper-level scheduling layer. Through local closed-loop control, it rapidly tracks the power setpoints issued from the upper layer, suppressing the impact of measurement noise and equipment parameter perturbations on execution accuracy. This forms a complete closed loop from optimization calculation to physical control, ensuring that the scheduling strategy is accurately executed on the actual equipment. The final physical control command is generated through closed-loop feedback.
[0115] In this embodiment, after each control command is issued to the corresponding device, the device executes the actual control action and simultaneously feeds back the actual execution status to the scheduling system, forming a closed-loop feedback. The scheduling system corrects subsequent control commands based on the feedback status to ensure that the actual execution result is consistent with the scheduling scheme. This step, by issuing the control quantities obtained from hierarchical scheduling to the corresponding devices for execution and correcting subsequent commands through closed-loop feedback, transforms the scheduling scheme from optimized calculation results into the actual physical actions of each device, forming a complete closed loop from carbon price signal sensing, energy storage capacity allocation, computing load migration to hierarchical solution and then to physical execution, achieving coordinated control of carbon, electricity, and computing in a carbon price fluctuation environment.
[0116] The specific values of the above preset parameters can be adjusted according to the actual application scenario. For example, the low threshold and high threshold can be recalibrated according to the historical volatility statistical distribution of different carbon markets. The basic correction strength coefficient and the minimum correction strength coefficient can be adjusted according to the response characteristics of the energy storage system and the conservatism of the scheduling strategy. The smoothing coefficient can be selected according to the system's trade-off between response speed and stability. The carbon price influence coefficient and the volatility influence coefficient can be calibrated according to the degree of influence of carbon price signals on scheduling decisions.
[0117] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.
[0118] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.
[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A method for joint optimization of shared energy storage and computing load in a data center microgrid cluster, characterized in that, Includes the following steps: Obtain carbon market volatility signals and determine the correction strength for energy storage capacity allocation based on these signals, wherein the correction strength decreases as the carbon market volatility signal increases. The initial energy storage capacity allocation plan is rolled over and adjusted using this adjustment intensity to obtain the energy storage power capacity and energy storage capacity allocated to each microgrid. Based on the energy storage power capacity and energy storage capacity, the migration direction and migration load amount of the batch processing load are determined. The additional cooling energy consumption required for the additional heat generation after migration to the receiving party is converted into carbon emission increment cost based on the internal carbon quota spot price. This carbon emission increment cost is included in the migration cost. Layered scheduling is performed on multiple time scales in descending order. Carbon futures forward price signals are used to correct global capacity allocation, carbon option implied volatility signals are used to adjust regional coordination penalty parameters, and carbon spot real-time price signals are used to correct the charging and discharging tracking deviation weights of local tracking. Output and execute energy storage charging and discharging power, main grid switching power, microgrid inter-trading power, migration load, and carbon quota trading volume.
2. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 1, characterized in that, This carbon market volatility signal is obtained by a weighted combination of carbon futures price volatility and carbon option implied volatility. The correction strength is set at a baseline value when the carbon market volatility signal is below a low threshold and at a minimum value when the carbon market volatility signal is above a high threshold. When the carbon market volatility signal is between the low threshold and the high threshold, it decreases monotonically as the carbon market volatility signal increases.
3. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 2, characterized in that, The initial energy storage capacity allocation plan is based on the weighted sum of the proportion of each microgrid's day-ahead forecast load in the cluster and the proportion of day-ahead forecast renewable energy output in the cluster, forming the basic weight of each microgrid. The adjustment weight of each microgrid is determined by the proportion of the absolute value of the actual net load of each microgrid to the absolute value of the total net load of the cluster. The adjustment amount for energy storage capacity allocation is determined by the difference between the adjustment weight and the base weight, and the adjustment strength.
4. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 3, characterized in that, The energy storage capacity allocation correction includes a power capacity correction and an energy capacity correction. When the adjustment weight is greater than the base weight, both the power capacity correction and the energy capacity correction are positive, and both increase monotonically with the increase of the absolute value of the actual net load, with the growth rate decreasing and approaching saturation. When the adjustment weight is less than or equal to the base weight, both the power capacity correction and the energy capacity correction are negative, and both decrease monotonically with the increase of the absolute value of the actual net load and approach zero. The final comprehensive weight of the previous period and the current adjusted weight are smoothed and then normalized to obtain the final comprehensive weight of each microgrid in the current period. The final comprehensive weight of the previous period is the final comprehensive weight obtained from the previous period using the same smoothing normalization method; The final energy storage power capacity and energy storage capacity allocated to each microgrid are determined by this final comprehensive weight, and the sum of the allocated capacities of all microgrids is equal to the total available capacity after deducting the system-level spinning reserve capacity from the total shared energy storage capacity.
5. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 1, characterized in that, The migration direction and migration load amount of this batch processing load are determined in the following way: The migration load is coupled to the temperature change of the receiving server cluster, and the migration load does not exceed the thermal response capacity of the receiving cooling system. The additional cooling energy consumption resulting from the relocation is converted into the incremental cost of carbon emissions based on the spot price of the internal carbon quota, the carbon emission coefficient of the main grid electricity purchase, and the energy efficiency ratio of the cooling system. When selecting relocation targets, priority should be given to avoiding microgrids with high costs for incremental carbon emissions or weak thermal response capabilities.
6. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 5, characterized in that, The temperature change of the receiving server cluster is updated based on the recursive relationship between the temperature of the previous period, the difference between server heat generation and cooling, heat capacity, and time interval. The heat generation of the server and the total power of the server are determined according to the heat generation coefficient; The migration load increases the server's total power and thus its temperature; when the temperature reaches the safe limit or the rate of temperature rise exceeds the thermal response capability of the receiving party's cooling system, the receiving party rejects the migration load. The additional cooling energy consumption generated by the migration is determined by the migration load, the heat generation coefficient, the energy efficiency ratio of the cooling system, the carbon emission coefficient of the main grid electricity purchase, and the spot price of the internal carbon quota.
7. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 1, characterized in that, In this hierarchical scheduling: The longest timescale will be used to adjust the overall cost of shared energy storage capacity allocation using the carbon futures forward price signal. The intermediate timescale is used to determine the initial value of the regional coordination penalty parameter by using the implied volatility signal of the carbon option. The shortest time scale is used as the real-time spot price signal of carbon as the correction term for the charge and discharge tracking deviation weight in the local tracking optimization; The three time scales are updated sequentially in the direction of decreasing time.
8. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 7, characterized in that, The first level performs global capacity allocation at the first preset time interval, and applies the correction term of the ratio of the carbon futures forward price signal to the reference carbon price to the effective use cost of shared energy storage power capacity and energy storage capacity. The second level performs regional coordination at the second preset time interval. The correction term of the ratio of the implied volatility signal of the carbon option to the reference volatility is used to determine the initial value of the regional coordination penalty parameter. Each microgrid solves the objective in parallel with the local cost term, the adjacent microgrid deviation term and the regional coordination penalty parameter until the original residual and the dual residual both meet the convergence condition. The third level performs local rolling tracking at the third preset time interval, at the cost of the deviation between the actual output of the local equipment and the reference trajectory and the control increment, and applies the charge and discharge tracking deviation weight after the real-time carbon spot price signal is corrected to the cost. Each solution is obtained by issuing only the most recent control increment and rolling forward.
9. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 5, characterized in that, The migration direction and amount of the batch processing load are also subject to the following constraints: The queued load accumulation represents the amount of batch processing load that has arrived but not yet been processed locally and transferred out as of the end of the current time period. It is updated according to the recursive relationship of the remaining amount in the previous time period, newly arrived load, locally processed load, transferred out load, and transferred in load, and ensures that each batch processing load is processed within the specified time limit after arrival. The amount transferred out in a single time period shall not exceed the allowable transfer ratio of the current batch processing load of the microgrid; The amount received by the receiver shall not exceed the idle processing capacity of its server. The amount of data transferred between any two micronets in a single time period shall not exceed the data transmission link capacity.
10. The method for joint optimization of shared energy storage and computing load in a data center microgrid cluster according to claim 8, characterized in that, After the hierarchical scheduling converges: The energy storage charging and discharging power request obtained by solving the shortest time scale is sent to the energy storage converter and a drive signal is generated. The power request for inter-micronet transactions is sent to the bidirectional power flow controller at the public connection point; The main grid switching power request is sent to the grid-connected converter; The migration load is converted into a number of task migrations and then sent to the data center task scheduling system. The carbon quota trading volume will be distributed to the interface of the tertiary carbon trading market to execute the actual trading operation. The final physical control command is generated through closed-loop feedback.
