A distributed energy aggregator double-time-scale closed-loop scheduling method and system

Abstract

The present application relates to the technical field of energy resource scheduling, and particularly relates to a distributed energy aggregator double-time-scale closed-loop scheduling method and system. The method comprises unified mapping based on obtained distributed power consumption data; constructing a regional target net power curve based on the unified mapped data; generating a real-time control signal based on the distributed power consumption data; generating a regional side local scheduling instruction according to the real-time control signal; calculating a tracking deviation and a resource pressure state based on an energy distribution state under the local scheduling instruction and the net power curve; slow-time-scale LS-Corr double-channel feedback regulation based on the tracking deviation and the resource pressure state; the present application improves the matching of the regional target curve and the local regulation capacity through the CDL two-stage decomposition method of "natural net load baseline + system adjustment amount distribution + heterogeneity correction".

Description

Technical Field

[0001] This invention relates to the field of energy resource scheduling technology, and in particular to a method and system for closed-loop scheduling of distributed energy aggregators with dual time scales. Background Technology

[0002] With the rapid development of cloud computing and artificial intelligence training, data centers are increasingly exhibiting trends of multi-regional deployment and cross-domain collaborative operation. Their electricity load fluctuates significantly and is closely related to grid peak shaving, demand response, and renewable energy consumption. The grid side typically guides multi-regional data centers to track the target net power curve within the scheduling cycle by issuing Customer Guided Load (CDL) curves, in order to achieve system load shaping and renewable energy consumption.

[0003] In multi-regional data center scenarios, different regions differ in terms of time-of-use electricity pricing, local photovoltaic output, IT load levels, cooling load, task migration capabilities, batch processing delay capabilities, and the availability of distributed energy resources. If the overall CDL target is directly distributed to each region, or if it is only decomposed according to the baseline load ratio, some regions may be burdened with tasks beyond their actual adjustment capabilities, resulting in uneven distribution of regional targets, increased tracking deviations, and higher adjustment costs.

[0004] Existing multi-region load coordination methods typically employ optimal scheduling, game-theoretic coordination, or price incentives to allocate target curves, incorporating local energy storage or flexible loads for response. While these methods can improve target tracking performance to some extent, they lack comprehensive consideration of regional natural net power baselines, directional correlation adjustability margins, cooling-side flexibility, and historical tracking performance, making it difficult to ensure that regional target curves match local actual execution capabilities. Furthermore, distributed energy resource pools are constrained by maximum charging / discharging power, energy capacity, state of charge boundary, and charging / discharging efficiency. When multiple regions compete to utilize distributed energy resources in the same time slot, the lack of real-time coordination signals reflecting resource scarcity can easily lead to problems such as total requests exceeding the resource pool's executable power limit, infeasible request results, and unstable resource allocation.

[0005] Existing cross-cycle feedback mechanisms primarily rely on deviation penalties or unified control quantity corrections, making it difficult to distinguish between two types of problems: insufficient target tracking capability and long-term shortage of distributed energy resources. If target tracking deviations cannot be fed back to the target decomposition stage of the next cycle, and resource pressures cannot be fed back to the distributed energy allocation stage, it will be difficult to achieve synergistic optimization between target tracking and resource constraints.

[0006] Therefore, a hierarchical collaborative scheduling method for distributed energy resources in multi-regional data centers is needed. This method should consider the natural net power baseline, regional heterogeneity, and execution capability when constructing regional target curves. In real-time scheduling, it should use scarcity signals to coordinate distributed energy resource requests. At the cross-cycle level, it should achieve rolling correction through target burden reduction and resource pressure feedback, thereby improving target tracking performance, the feasibility of distributed energy resource allocation, and the stability of system operation. Summary of the Invention

[0007] To address the following issues in demand response scenarios involving multiple data centers: 1. The target curve struggles to balance regional electricity price differences, renewable energy fluctuations, and shared resource capacity constraints; 2. The lack of a mechanism to integrate shared energy storage into the unified scheduling closed loop, resulting in limited peak shaving, absorption, and target tracking capabilities; 3. The lack of a cross-cycle feedback mechanism to separately handle target tracking deviations and distributed energy resource pressures, this invention provides a dual-timescale closed-loop scheduling method and system for distributed energy aggregators.

[0008] Firstly, the present invention provides a dual-time-scale closed-loop scheduling method for distributed energy aggregators, which adopts the following technical solution: A distributed energy aggregator dual-timescale closed-loop scheduling method includes: Acquire distributed power consumption data from regional data centers; A unified mapping is performed based on the acquired distributed electricity consumption data; Construct a regional target net power curve based on the unified mapped data; Real-time control signals are generated based on distributed power consumption data; Generate local dispatch instructions on the regional side based on real-time control signals; Calculate tracking deviation and resource pressure status based on energy allocation status and net power curve under local dispatch instructions; Slow-timescale LS-Corr dual-channel feedback regulation based on tracking deviation and resource pressure status; Output feedback adjustment results.

[0009] Secondly, a distributed energy aggregator dual-timescale closed-loop scheduling system includes: The data acquisition module is configured to acquire distributed power consumption data from the regional data center. The mapping module is configured to perform unified mapping on the acquired distributed electricity consumption data; The net power module is configured to construct the regional target net power curve from the uniformly mapped data. The control module is configured to generate real-time control signals from distributed power consumption data; The instruction module is configured to generate local scheduling instructions on the regional side in real time; The deviation module is configured to calculate and track deviations and resource pressure status based on the energy allocation status and net power curve under local scheduling commands. The feedback module is configured to track deviations and resource pressure status using a slow-timescale LS-Corr dual-channel feedback adjustment. The output module is configured to provide feedback on the adjustment results.

[0010] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned distributed energy aggregator dual-time-scale closed-loop scheduling method.

[0011] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a distributed energy aggregator dual-time-scale closed-loop scheduling method.

[0012] In summary, the present invention has the following beneficial technical effects: 1. This invention improves the matching between regional target curves and local regulation capabilities through a two-stage CDL decomposition method of "natural net load baseline + system regulation allocation + heterogeneity correction".

[0013] 2. This invention generates a real-time scarcity signal through projection-dual updates, which suppresses excessive requests and ensures that the allocation results meet power constraints when distributed energy resources are limited.

[0014] 3. This invention improves the overall coordination effect between target tracking and distributed energy resource constraints by forming a slow time-scale feedback through the Corr target burden reduction channel and the LS resource priority channel. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a distributed energy aggregator dual-time-scale closed-loop scheduling method according to Embodiment 1 of the present invention; Figure 2 This is a hierarchical collaborative operation modeling framework diagram of distributed energy aggregators according to Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the two-stage decomposition and heterogeneity correction of the CDL target curve in Embodiment 1 of the present invention; Figure 4 This is a diagram illustrating the effect of CDL two-stage heterogeneity decomposition on tracking bias in Embodiment 1 of the present invention. Figure 5 This is a graph showing the effect of fast timescale scarcity signals on distributed energy application rates in Embodiment 1 of the present invention. Figure 6 This is a diagram illustrating the mechanism of dual-channel LS+Corr under a slow timescale in Embodiment 1 of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings.

[0017] Example 1 Reference Figure 1 This embodiment presents a dual-timescale closed-loop scheduling method for distributed energy aggregators. Its core lies in constructing a two-part periodic correction mechanism of "real-time signal generation + post-event periodic correction," forming a closed-loop coordination architecture based on unified metering results. During the real-time scheduling phase, the upper-level distributed energy aggregator resource coordination layer acts as the leader, generating a real-time scarcity signal and issuing control signals based on the total available charging and discharging power limit, energy capacity boundary, and state constraints of the distributed energy resource pool, according to the aggregated power requests from each region. Each regional data center aggregator acts as the follower, generating distributed energy request power and a set of local execution instructions based on preset local response rules and heuristic scheduling criteria, given the control signals and the regional target net power curve. The request results are then fed back to the upper layer to update the real-time scarcity signal, thus forming a master-slave iterative coordination closed loop at a fast timescale.

[0018] In the post-event periodic correction phase, the upper layer recalculates the target tracking deviation of each region based on unified metering records, generates periodic feedback correction quantities, and combines long-term scarcity memory states to continuously update the benchmark control components for the next scheduling cycle, forming a dual-state feedback adjustment closed loop under a slow time scale. Thus, the fast time scale achieves immediate coordination of the physical constraints of the distributed energy resource pool through real-time scarcity signals, while the slow time scale achieves continuous correction of the region's historical operating status and cross-cycle resource tension through periodic feedback correction and long-term scarcity memory. Together, they constitute a dual-time scale leader-follower master-slave coordination closed loop. Unlike conventional demand response methods that primarily rely on capacity signals or single post-event compensation, this invention combines "real-time coordination of resource scarcity" with "cross-cycle correction of target tracking performance," realizing a collaborative scheduling mechanism for distributed energy resource pools that can be periodically updated, uniformly metered and recalculated, and adapted to regional heterogeneity differences.

[0019] S1 is a dual-time-scale collaborative scheduling system for distributed energy aggregators across multiple data centers. S1.1 Construct a collaborative scheduling system consisting of a grid-side target release unit, a distributed energy resource coordination layer, multiple regional data center aggregators, local renewable energy units, and a metering and energy management unit; in some embodiments, the collaborative scheduling system further includes a thermal inertia auxiliary regulation unit, which is used to provide additional flexible regulation capabilities when the response of electrochemical energy storage is limited.

[0020] S1.2 The grid-side target issuing unit is used to issue the customer guidance load curve (CDL), i.e., the target total net power curve, to the system. The distributed energy resource coordination layer is used to uniformly manage the total available charging and discharging power, remaining available energy capacity, state boundaries and related metering information of the distributed energy resource pool; the multiple regional data center aggregators correspond to data center nodes in different regions and are used to generate local scheduling instructions by combining local load, renewable energy, computing power task constraints and distributed energy call conditions.

[0021] S1.3 The information interaction process of the system includes: the power grid side sends the CDL target curve to the distributed energy resource coordination layer; the distributed energy resource coordination layer sends control signals to the aggregators of each regional data center; the aggregators of each regional data center report the power requested by distributed energy and necessary status information to the distributed energy resource coordination layer; the metering and energy management unit records and transmits back the actual net power curve of each region according to the time slot; after the scheduling cycle ends, the distributed energy resource coordination layer uniformly calculates the tracking deviation and generates the periodic feedback correction status quantity. and execute long-term scarce memory state quantity Update.

[0022] S1.4 Distributed Integrated Energy Aggregator is used to uniformly aggregate power resources, cooling and heating load resources, computing load resources, and renewable energy consumption resources in multi-regional data centers. Among them, power resources include distributed energy storage, backup power supplies, UPS, and power resources that can participate in power regulation equivalently; cooling and heating load resources include chillers, air conditioning systems, cold storage capacity, and building thermal inertia; computing load resources include interactive task migration, batch processing task postponement, and server load rearrangement; renewable energy consumption resources include local photovoltaic power consumption, curtailment suppression, and load absorption capacity during low-price periods.

[0023] To facilitate unified scheduling, distributed integrated energy aggregators map different types of resources to equivalent net power regulation capabilities, and use available power, duration, response speed, regulation cost, and operational constraints as unified modeling dimensions. For resource type r in region i, its equivalent adjustable capability is defined as: , in, This represents the maximum adjustable power of resource type r in time slot t. This indicates sustainable capacity or available energy. Indicates response time. Indicates the cost of invocation. Let represent the set of resource operation constraints. Furthermore, the comprehensive adjustability of region i can be written as: , in, A collection of resource types managed by distributed integrated energy aggregators. This represents the weight of the resource type.

[0024] The controlled object of the distributed energy resource coordination layer described in S1.5 is not an abstract price entity, but a distributed energy resource pool that is competitively called by multiple regional data centers. Therefore, the core of this invention is to construct a dual-time-scale closed-loop scheduling mechanism that combines real-time coordination and cross-cycle correction around shared physical resources constrained by power limits, energy capacity, state boundaries and metering results. The dual-time-scale closed loop includes fast-time-scale real-time scarcity coordination and slow-time-scale normalized deviation correction update.

[0025] S2 Establish a regional net power model and construct a regional target net power curve. S2.1 For the i-th regional data center, predict the power based on the service load. Cooling baseline power and the available power of local renewable energy Establish a regional baseline net power model: , in, This represents the natural net power baseline of region i in time slot t before considering the power output of distributed energy sources and active cooling adjustments. The cooling baseline power can be generated from the region IT power according to a preset cooling coefficient, i.e.: , in, It is the regional cooling power factor, used to characterize the differences in cooling efficiency, climate conditions, and equipment status of data centers in different regions.

[0026] S2.2 In each time slot t, obtain the customer guidance load curve issued by the grid side. And calculate the total adjustment of the system relative to the regional natural net power baseline: , in, This indicates the net power regulation that the system needs to share among all regions in this time slot. When the target net power of the system is higher than the current total natural net power, it means that each region needs to increase its net power or decrease the downward adjustment amount; when When this occurs, it indicates that the system's target net power is lower than the current total natural net power, requiring each region to reduce its net power or increase its flexible response.

[0027] Compared to directly The power distribution varies across different regions. This invention first preserves the natural net power baseline for each region, and then adjusts only the system power. A two-stage allocation is implemented to avoid confusing the regional base load level with the actual adjustment task.

[0028] S2.3 Determine the adjustability margin for each region based on the adjustment direction. For region i, let its feasible range for target net power be: , when At that time, the upsizing margin for region i is: , when At that time, the downsizing margin for region i is: , Based on this definition, the direction-dependent adjustable margin is: , This approach ensures that different regions only handle system regulation within their feasible adjustment space, avoiding the assignment of regulation tasks beyond the region's physical or operational capabilities to lower-level entities.

[0029] S2.4 Based on the directional correlation adjustable margin, a regional heterogeneity correction factor is introduced. This heterogeneity correction factor consists of normalized indicators such as electricity price level, local renewable energy output, distributed energy availability, task migration elasticity, and batch processing delay elasticity. , in, For normalized electricity price indicators, To normalize renewable energy output targets, This is an indicator of the availability and status of distributed energy resources. For task migration elasticity indicators, This is a flexible indicator for batch processing delays; The weight coefficients are non-negative and satisfy the following conditions: , To avoid over-amplifying the distribution differences between regions by heterogeneity correction, the original heterogeneity factor can be further weakened to: , And restrict it to a preset range: , in, This is the intensity coefficient for heterogeneity correction. and These are the lower and upper limits of the heterogeneity factor, respectively.

[0030] S2.5 To further improve the matching degree between target decomposition and actual regional execution capabilities, an execution attainability factor is introduced. and historical tracking of burden reduction factors Among them, the execution reachability factor is used to reflect the region's executable adjustment capability under the current time slot, which is jointly affected by cooling flexibility, distributed energy availability, task migration capability, and batch processing delay capability; the historical tracking burden reduction factor is generated by the normalized tracking deviation of the previous scheduling cycle, and is used to appropriately reduce the adjustment tasks of regions that were difficult to track in the previous cycle in the next cycle.

[0031] The effective allocation weight of region i is defined as: , when At that time, the proportion of system regulation undertaken by region i is: , when At that time, take: , Where N represents the number of data centers in the region. Based on this, the initial target net power for the region is generated: , S2.6 Perform feasible region projection and total closure check on the initial target net power of the region. First, [the following steps are taken]. Projected onto the feasible region: , Then calculate the target total residual: , If r(t) > 0, the residuals are further allocated according to the remaining upward adjustment margin of each region; if r(t) < 0, the residuals are further allocated according to the remaining downward adjustment margin of each region; if all regions reach the feasible boundary but cannot be completely closed, the unclosed residuals are recorded and the target power is kept within the feasible range. After projection and residual redistribution, the final regional target net power curve is obtained. The final target net power curve for the region satisfies the feasible region constraint, and under feasible conditions, satisfies: , Therefore, this step achieves a two-stage decomposition from the total CDL target curve on the grid side to the target net power curves for multiple regions. This decomposition method does not simply divide the total system target power proportionally, but rather, while preserving the regional natural net power baseline, it allocates the system regulation quantities under directional correlation, heterogeneity correction, and execution capability constraints, thereby improving the matching degree between the regional target curves and local resource conditions, flexibility capabilities, and historical tracking performance.

[0032] S3 establishes a distributed energy resource pool coordination model and generates real-time control signals. S3.1 Establish a two-layer coordination model consisting of a distributed energy resource coordination layer and multiple regional data center aggregators. The distributed energy resource coordination layer acts as the leader, and the multiple regional data center aggregators act as followers, forming a two-layer master-slave coordination model. The upper layer generates control signals based on the total available charging and discharging power limit of the distributed energy resource pool, energy capacity boundaries, and state constraints. The lower layer, given the control signals and the regional target net power curve, completes local scheduling and reports the distributed energy power requests.

[0033] S3.2 The control signal issued by the upper layer consists of a reference control component and a real-time scarcity signal: , in, For region i, the reference control component, This is a real-time scarcity signal that reflects the marginal tension of distributed energy resource pools.

[0034] S3.3 Within each time slot t, the distributed energy resource coordination layer aggregates the distributed energy request power reported by the aggregators of data centers in each region. Let the distributed energy request power of region i in the k-th round of interaction be... The total number of applications is: , To meet the maximum executable power constraint of the distributed energy resource pool By performing feasible region projection on the total application amount, the executable power of the distributed energy resource pool in the current round is obtained: , Furthermore, a real-time scarcity signal is generated based on the deviation between the total application amount and the upper limit of distributed energy power. This real-time scarcity signal is iterated according to the following projection-dual update rule: , in, This is the real-time scarcity signal in the k-th round of interaction. To update the step size, This represents the nonnegative projection operator. If an upper limit for the scarcity signal is set... Then the update formula can be written as: , when When the total number of applications in each region exceeds the power limit of the distributed energy resource pool, the real-time scarcity signal increases and is fed back to the application rules of each region in the next round of interaction to suppress excessive applications; when As time progresses, the real-time scarcity signal gradually falls back to the non-negative feasible region. The iterative process satisfies: , Or reach the maximum number of iterations Stop when the time slot is reached, and output the real-time scarcity signal for the current time slot. \ and distributed energy's executable power z(t).

[0035] Data center aggregators in each region adjust the power allocation for the next round of distributed energy applications based on the updated real-time scarcity signal. For example: , in, This represents the basic application items determined by the regional target net power deviation, local flexibility adjustment capability, and application ratio coefficient. This forms a fast-timescale coordination closed loop of "application aggregation - feasible region projection - scarcity update - regional application revision".

[0036] S3.4 Regional data center aggregators, given a control signal and regional target net power curve Under these conditions, the power request for distributed energy resources is generated according to the local response rules. and the local execution instruction set; the distributed energy resource coordination layer determines the actual execution power of each region based on the application results of each region and the physical constraints of the resource pool. This will serve as the basis for subsequent measurement, deviation recalculation, and periodic settlement.

[0037] S4 generates regional-side local dispatch instructions and forms distributed energy resource requests. S4.1 Regional data center aggregators, given a control signal and regional target net power curve Under these conditions, based on preset rules for task migration, batch processing extension, priority consumption of renewable energy, and distributed energy dispatch, the power requested by distributed energy is output. And a set of locally executed instructions. The locally executed instructions include: migrating cross-regional tasks under network bandwidth and service constraints, adjusting the timing of batch processing tasks under time window and quality of service constraints, increasing the consumption of local renewable energy during periods of photovoltaic surplus or low prices, and executing charging and discharging plans based on the distributed energy allocation results.

[0038] After receiving the control signal from the coordination layer of the distributed integrated energy aggregator, the S4.2 regional data center aggregator calculates the resource call score for different resource types based on local target deviations and the status of various resources. For resource type r in region i, the resource call score is defined as: , in, Indicates resource availability. Indicates the cost of resource allocation. This indicates a resource response delay. This indicates potential service quality, temperature out-of-bounds, or operational risks that may arise from using this resource. This indicates the scarcity signal corresponding to this resource type. to These are non-negative weighting coefficients.

[0039] The regional side determines resource allocation priorities based on the scoring results, and generates power adjustment requests for various resources under the conditions of meeting service quality, temperature boundaries, task completion time limits, power boundaries, and energy boundaries. .

[0040] The regional integrated regulation application power is: , When the target deviation is small or time permits, low-cost resources, such as batch processing delays, renewable energy consumption, and cold load regulation, should be prioritized. When the target deviation is large or the response time is short, fast power resources, including distributed energy storage or UPS regulation capabilities, should be prioritized. When a single resource is insufficient, a combined regulation request should be formed according to the scoring results.

[0041] S4.3 For computing power tasks in region i at time slot t, define the interactive task arrival quantity. Batch processing task arrival volume Local execution volume and Migration task volume and batch processing delay All tasks must satisfy task conservation constraints, service quality constraints, and migration resource constraints, among which batch processing delay must meet the following requirements. , Cross-region migration task volume meets bandwidth limit constraint , The IT power consumption in region i of S4.4 is determined by the base power, task execution power, and migration-related power: , in, Based on the no-load power, Additional power for migration, Server utilization is affected by the combined volume of interactive tasks, batch tasks, and migration tasks. The region's actual net power is determined by IT power consumption, cooling power, local renewable energy output, and distributed energy output, and is denoted as... , Among them, cooling power This includes the baseline cooling power and the amount of change due to adjustments.

[0042] S4.5 Each region will report the requested power and necessary status information of distributed energy to the distributed energy resource coordination layer, which will be used by the upper layer to update the total request amount, generate a real-time scarcity signal and complete resource allocation within the current time slot.

[0043] S5 calculates tracking deviation and resource pressure status based on operation records. S5.1 During real-time scheduling, the distributed energy resource coordination layer records the target net power, actual net power, distributed energy requested power, distributed energy actual executed power, and resource allocation-related status of each region according to time slots. For region i in time slot t, the target net power of the region is recorded. Actual net power Distributed energy application power Distributed energy execution power And the results of distributed energy allocation. The actual net power is calculated according to the following formula: , in, To take into account the actual cooling power after the cooling side adjustment.

[0044] After the S5.2 scheduling cycle ends, the distributed energy resource coordination layer recalculates the target tracking error of each region based on the operation records. For region i in time slot t, its time slot tracking error is defined as: , To avoid the impact of power magnitude differences in different regions on the deviation evaluation results, the regional periodic tracking deviation index is calculated using the normalized root mean square form: , Where T is the set of time slots within the scheduling period, and |T| is the number of time slots. To prevent extremely small positive numbers with a denominator of zero. This is used to characterize the normalized tracking deviation of region i from the target net power curve within the current scheduling period.

[0045] S5.3 characterizes the resource scarcity of the distributed energy resource pool during the scheduling cycle by calculating the system-level application pressure and the proportion of unmet applications at the regional level. Let the total application volume for each region in the current time slot be: , System-level resource pressure is defined as: , in, This represents the maximum executable power of the distributed energy resource pool. This is the threshold triggered by resource pressure. The regional-level unmet application ratio is defined as: , in, This is used to indicate the degree to which distributed energy applications in region i have not been fulfilled.

[0046] S5.4 Based on the combined system-level resource pressure and the proportion of unmet applications at the regional level, a regional resource tension level is constructed: , in, This is a weighting coefficient between system-level pressure and unmet regional requests. The regional periodic resource stress is obtained by averaging the stress across all time slots within the scheduling cycle. , The Used to drive long-term scarce memory state quantities Update.

[0047] After the S5.5 scheduling cycle ends, the distributed energy resource coordination layer generates a set of cycle operation records: , in, This is the target burden reduction factor for region i. The periodic operation record is used for target decomposition correction and distributed energy allocation priority correction in the next scheduling cycle. If necessary, the operation record can be checked for time integrity and physical consistency; if data is missing or abnormal, the corresponding record will not participate in subsequent cycle feedback updates.

[0048] S6 performs slow-timescale LS-Corr dual-channel feedback regulation. S6.1 At a slow time scale, track deviation indicators based on regional cycles. Generate a Corr tracking correction channel. This channel does not directly update the distributed energy request suppression amount, but instead generates the target load reduction factor for the next scheduling cycle. This is then fed back to the CDL two-stage target decomposition process. For region i, the original target burden reduction factor is first calculated: , in, To reduce the target load intensity coefficient, To preset a normalized tracking deviation threshold. To ensure that the overall adjustment responsibility of each region does not shift systematically, the following is implemented: Perform mean normalization: , And limit it to a preset range: , in, These represent the lower and upper limits of the target burden reduction factor, respectively. If a region had a large tracking deviation in the previous period, then its... The relative decrease results in a lower adjustment weight in the next cycle's CDL decomposition; if the tracking deviation in a certain region is small, then its With a relatively larger size, it can undertake more regulatory tasks.

[0049] S6.2 Simultaneously, based on regional cyclical resource tension... Update long-term scarce memory state quantity The aforementioned No longer driven directly by tracking deviation, but by the pressure of distributed energy application and the degree of unmet applications, its update rule is as follows: , in, Forgetting coefficient, To enhance the resource scarcity factor, These represent the lower and upper limits of long-term scarce memory, respectively. When a region is in a state of high resource application pressure or a high degree of unmet application needs for an extended period... Increase; when resource shortages ease. It gradually decays according to the forgetting mechanism.

[0050] S6.3 In the CDL two-stage target decomposition of the next scheduling cycle, the target burden reduction factor generated by the Corr channel will be... Introduce effective allocation weights for regions. For region i, its effective allocation weight is written as: , in, For directionally adjustable margin, This is a weakened heterogeneity correction factor. To implement the reachability factor, the proportion of system regulation undertaken by region i is: , Therefore, the Corr channel passes through It is applied to the generation process of the net power curve of the target in the next cycle region, thereby reducing the burden on the target layer in areas where tracking is difficult.

[0051] S6.4 In the distributed energy allocation process, the long-term scarce memory generated by the LS channel is introduced into the distributed energy allocation priority. First, based on the relationship between the total request amount in the current time slot and the maximum power of the distributed energy, a resource scarcity gating factor is constructed: , in, This is the LS gating trigger threshold. When the pressure from distributed energy resource requests is low... When the value is close to zero, the LS channel basically does not interfere with the allocation; when the pressure of distributed energy resource applications is high, Increase the priority of LS channels in allocation.

[0052] based on and Effective application power in the construction region: , in, For resource type weights, Let be the LS gating factor for resource type r. The distributed energy resource coordination layer is based on the effective requested power in each region. Perform distributed energy power allocation and ensure that the actual power output meets the following requirements: , , Therefore, when distributed energy resources are scarce, the LS channel can conditionally prioritize and suppress areas with high long-term resource pressure to prevent them from continuously occupying distributed energy resources.

[0053] In S6.5, at a slow timescale, the LS channel is used not only to correct resource allocation priorities in different regions but also to correct the call weights of different resource types in the next scheduling cycle. The distributed integrated energy aggregator coordination layer updates the resource type weights based on the scarcity, call cost, and target tracking contribution of each resource type in the previous cycle. For resource type r, its periodic scarcity is defined as: , Its target tracking contribution is defined as: , Where ΔΦ_r represents the contribution of resource type r to the improvement of regional target tracking deviation. This represents the call cost for this resource type within the period. The resource type weight update rule is as follows: , Where γ is the weight update step size, and F(·) is the weight correction function constructed based on resource contribution, resource scarcity, and call cost. If a certain type of resource has a high contribution to target tracking and low resource pressure in the previous cycle, its call weight will be appropriately increased in the next cycle; if a certain type of resource is in a state of long-term tension, has high call cost, or has limited improvement on target tracking, its call weight will be decreased in the next cycle.

[0054] In the next scheduling cycle of S6.6, the distributed energy resource coordination layer simultaneously uses fast-timescale real-time scarcity signals. The target burden reduction factor generated by the Corr channel And the amount of long-term scarce memory generated by LS channels Rolling coordination is performed. This includes coordination for the real-time feasibility of requests in the current time slot. Used for CDL target decomposition correction in the next cycle. This is used for priority adjustment in distributed energy allocation. This forms a closed-loop collaborative scheduling mechanism with two time scales: "fast time-scale scarcity coordination + slow time-scale target burden reduction feedback + slow time-scale resource pressure feedback."

[0055] Experimental verification The following describes the implementation of this invention using a distributed energy collaborative scheduling example for six regional data centers. In this embodiment, the system includes six regional data center nodes: IA, SC, OR, NC, OK, and GA. The scheduling cycle is 24 hours, with a time step of 1 hour. The grid side issues the Customer Guided Load (CDL) curve. Input data for each region includes IT load forecast, available local photovoltaic output, cooling baseline power, time-of-use pricing, outdoor temperature, task migration capability, batch processing delay capability, and distributed energy application and execution records. The distributed energy resource pool is configured with maximum charging and discharging power, energy capacity, charging and discharging efficiency, state of charge boundary, and executable power boundary, and is uniformly managed by the distributed energy resource coordination layer. This experiment follows... Figure 1 The simulation scenario is constructed using the dual-time-scale closed-loop scheduling process shown, and... Figure 2 The hierarchical collaborative operation framework shown organizes the processes of target modeling, coordination and control, regional response, and measurement correction.

[0056] During the regional basic load modeling phase, the IT power for each region is generated using a lightweight load-power coupling model, consisting of idle power, load utilization power, task migration additional power, and batch processing additional power. Cooling baseline power is generated as a proportion of IT power and combined with local photovoltaic output to form the regional natural net power baseline. This process is used to construct 24-hour simulation inputs for multiple regions, primarily serving the verification of distributed energy coordination mechanisms, rather than high-fidelity data center energy consumption modeling.

[0057] In the regional target curve generation stage, this embodiment adopts a two-stage method of "natural net load baseline + system adjustment allocation + heterogeneity and execution capability correction", the specific process of which is as follows: Figure 3 As shown. First, the difference between the grid-side CDL curve and the sum of the natural net power baselines for each region is calculated, and this difference is used as the system adjustment amount. Then, based on the adjustment direction, the adjustable margin for each region is determined by whether it is upward or downward. Finally, considering regional electricity prices, photovoltaic output, the availability of distributed energy resources, task migration flexibility, batch processing delay capability, execution accessibility, and historical tracking load reduction factors, a regional target net power curve is generated. This method does not directly allocate the total CDL target proportionally, but rather allocates actual adjustment tasks while preserving the region's natural load level, thereby improving the matching degree between target allocation and local regional adjustment capabilities. Figure 4 It is evident that after adopting the CDL two-stage decomposition, the overall target tracking deviation in each region is reduced, indicating that this decomposition method improves the matching degree between the target curve and the local adjustment capability.

[0058] During the real-time coordination phase of distributed energy resources, the distributed energy resource coordination layer acts as the leader, and regional data center aggregators act as followers. The regional side generates distributed energy request power based on target net power, natural net load, local flexibility, and real-time scarcity signals. After the distributed energy resource coordination layer aggregates the requests from each region, it performs feasible region projection based on the distributed energy power ceiling and generates real-time scarcity signals through projection-dual updates. When the total request amount exceeds the distributed energy executable power ceiling, the scarcity signal increases and suppresses excessive requests in the next round; when the total request amount is below the power ceiling, the scarcity signal falls back to the non-negative feasible region. Figure 5 This indicates that when the total number of regional applications approaches the upper limit of the resource pool, the real-time scarcity signal can suppress excessive applications, making the resource application process gradually more feasible and stable. This forms a fast-timescale coordination closed loop of "regional application - application aggregation - scarcity update - regional response correction".

[0059] During the regional response phase, data center aggregators in each region generate local responses using heuristic rules, without solving complex optimization problems. The regional side generates distributed energy request power and local execution instructions based on target deviation, available local photovoltaic output, cooling flexibility, task migration, and batch processing delay status. The distributed energy resource coordination layer determines the actual distributed energy execution power based on the request volume in each region, scarcity signals, resource pool power boundaries, and allocation priorities. Simultaneously, this embodiment introduces a virtual thermal energy storage-assisted regulation model, using a first-order equivalent thermal model to describe indoor temperature changes and providing additional net power flexibility through cooling regulation. During the simulation, the indoor temperature remained within the constraint range of 18°C ​​to 24°C and met the end-of-day temperature recovery requirement, demonstrating that cooling flexibility can participate in regional net power regulation without exceeding temperature boundaries.

[0060] After the scheduling cycle ends, the distributed energy resource coordination layer calculates the normalized tracking deviation between the actual net power and the target net power of each region based on the operation records, and calculates the resource pressure status based on the total distributed energy application volume, actual executed power, and unmet application status of the region. The tracking deviation is used in the Corr channel, and the resource pressure status is used in the LS channel. The Corr channel generates a target burden reduction factor based on the target tracking performance of the previous cycle and feeds it back to the CDL two-stage target decomposition of the next cycle, so that regions with tracking difficulties will bear a lower adjustment weight in the next cycle; the LS channel updates the long-term scarcity memory based on the resource tension and corrects the distributed energy allocation priority through gating factors and effective application power when distributed energy resources are scarce, thereby conditionally suppressing regions with high long-term resource pressure.

[0061] To verify the effectiveness of the proposed slow timescale feedback mechanism, this embodiment sets up four ablation modes: no slow timescale, Corr channel only, LS channel only, and LS–Corr dual-channel. Experimental results are as follows: Figure 6The results show that the Corr channel can improve regional target tracking through the target burden reduction factor, the LS channel can suppress distributed energy resource pressure through resource gating and effective power application, and the LS-Corr dual channel forms a comprehensive coordination between target tracking and resource constraints. Specifically, in this embodiment, under the no-slow-time-scale mode, the average tracking deviation at the end of the cycle is 0.178, the total number of unmet requests is 119.513, and the average resource pressure is 0.120; under the LS-Corr dual-channel mode, the average tracking deviation at the end of the cycle is reduced to 0.178, the total number of unmet requests is reduced to 119.308, and the average resource pressure is reduced to 0.198. At the same time, the comprehensive evaluation indices of deviation priority, balance, and resource priority are 0.999, 0.998, and 0.998, respectively, all lower than the baseline value of 1.000 without a slow-time-scale, indicating that the proposed dual-channel feedback mechanism can reduce resource pressure while maintaining target tracking performance.

[0062] As can be seen from the above embodiments, the present invention can realize IT load modeling, cooling flexibility modeling, CDL two-stage target decomposition, distributed energy fast-timescale scarcity coordination, regional heuristic response, virtual thermal energy storage auxiliary regulation, and slow-timescale LS-Corr dual-channel feedback in a six-region data center scenario. This embodiment verifies the feasibility of the present invention in a multi-region data center distributed energy collaborative scheduling scenario; simultaneously, ablation experiment data shows that the proposed LS-Corr dual-channel mechanism can achieve better overall coordination results compared to no slow-timescale and single-channel configurations, demonstrating the advantages of target tracking and distributed energy resource constraint collaborative optimization.

[0063] Example 2 This embodiment provides a distributed energy aggregator dual-timescale closed-loop scheduling system, including: The data acquisition module is configured to acquire distributed power consumption data from the regional data center. The mapping module is configured to perform unified mapping on the acquired distributed electricity consumption data; The net power module is configured to construct the regional target net power curve from the uniformly mapped data. The control module is configured to generate real-time control signals from distributed power consumption data; The instruction module is configured to generate local scheduling instructions on the regional side in real time; The deviation module is configured to calculate and track deviations and resource pressure status based on the energy allocation status and net power curve under local scheduling commands. The feedback module is configured to track deviations and resource pressure status using a slow-timescale LS-Corr dual-channel feedback adjustment. The output module is configured to provide feedback on the adjustment results.

[0064] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned distributed energy aggregator dual-timescale closed-loop scheduling method.

[0065] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a distributed energy aggregator dual-timescale closed-loop scheduling method.

[0066] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A distributed energy aggregator dual-time-scale closed-loop scheduling method, characterized in that, include: Acquire distributed power consumption data from regional data centers; A unified mapping is performed based on the acquired distributed electricity consumption data; Construct a regional target net power curve based on the unified mapped data; Real-time control signals are generated based on distributed power consumption data; Generate local dispatch instructions on the regional side based on real-time control signals; Calculate tracking deviation and resource pressure status based on energy allocation status and net power curve under local dispatch instructions; Slow-timescale LS-Corr dual-channel feedback regulation based on tracking deviation and resource pressure status; Output feedback adjustment results; The aforementioned slow-timescale LS-Corr dual-channel feedback regulation based on tracking deviation and resource pressure status includes, at the slow timescale, adjusting the regional periodic tracking deviation index... Generate a Corr tracking correction channel to generate the target load reduction factor for the next scheduling cycle. And feed it back to the CDL two-stage target decomposition process. For region i, first calculate the original target burden reduction factor: in, To reduce the target load intensity coefficient, To preset a normalized tracking deviation threshold and to ensure that the overall adjustment responsibility of each region does not shift systematically, the following measures are taken: Perform mean normalization: And limit it to a preset range: ,in, These represent the lower and upper limits of the target burden reduction factor, respectively; and are also based on the regional cyclical resource tension. Update long-term scarce memory state quantity The update rules are as follows: in, Forgetting coefficient, To enhance the resource scarcity factor, These represent the lower and upper limits of long-term scarce memory, respectively; in the CDL two-stage objective decomposition of the next scheduling cycle, the objective burden reduction factor generated by the Corr channel will be... Introducing the effective allocation weight for region i, its effective allocation weight is: ,in, For directionally adjustable margin, This is a weakened heterogeneity correction factor. To implement the reachability factor, the proportion of system regulation undertaken by region i is: Therefore, the Corr channel passes through It is applied to the generation process of the net power curve of the target in the next cycle region, thereby reducing the burden on the target layer in areas where tracking is difficult.

2. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 1, characterized in that, The unified mapping based on the acquired distributed energy data includes using a distributed integrated energy aggregator to uniformly aggregate electrical resources, heating and cooling load resources, computing load resources, and renewable energy consumption resources from multiple regional data centers. For unified scheduling, the distributed integrated energy aggregator maps different types of resources to equivalent net power regulation capabilities, using available power, duration, response speed, regulation cost, and operational constraints as unified modeling dimensions. For resource type r in region i, its equivalent adjustable capability is defined as: ,in, This indicates the maximum adjustable power of resource type r in time slot t. This indicates sustainable capacity or available energy. Indicates response time. Indicates the cost of invocation. The set of resource operation constraints is represented as follows: The overall adjustability of region i is expressed as: ,in, A collection of resource types managed by distributed integrated energy aggregators. This represents the weight of the resource type.

3. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 2, characterized in that, The construction of the regional target net power curve based on the unified mapped data includes, for the i-th regional data center, predicting power based on service load. Cooling baseline power and the available power of local renewable energy Establish a regional baseline net power model: ,in, This represents the natural net power baseline of region i in time slot t, before considering the shared energy storage execution power and active cooling adjustment. The cooling baseline power is generated by the region IT power according to a preset cooling coefficient, i.e.: ,in, The regional cooling power factor is determined; then, at each time slot t, the customer guidance load curve issued by the grid side is obtained. And calculate the total adjustment of the system relative to the regional natural net power baseline: ,in, This represents the net power adjustment, achieved by first preserving the natural net power baseline for each region, and then adjusting only the system level. A two-stage allocation is performed to avoid confusing the regional base load level with the actual adjustment task. Finally, the adjustment margin of each region is determined according to the adjustment direction. For region i, its feasible range for target net power is set as follows: ,when At that time, the upsizing margin for region i is: ,when At that time, the downsizing margin for region i is: Based on this, the direction-dependent adjustable margin is defined as follows: .

4. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 3, characterized in that, The construction of the regional target net power curve based on the unified mapping data also includes introducing a regional heterogeneity correction factor on the basis of the directional correlation adjustable margin, expressed as: ,in, For normalized electricity price indicators, To normalize renewable energy output targets, For shared energy storage available status indicators, For task migration elasticity indicators, This is a flexible indicator for batch processing delays; The weight coefficients are non-negative and satisfy the following conditions: To avoid over-amplifying inter-regional allocation differences in heterogeneity correction, the original heterogeneity factor is weakened to: And restrict it to a preset range: ,in, This is the intensity coefficient for heterogeneity correction. and These represent the lower and upper limits of the heterogeneity factor, respectively; then, to improve the matching degree between the target decomposition and the actual execution capability of the region, an execution reachability factor is introduced. and historical tracking of burden reduction factors The effective allocation weight for region i is defined as follows: ,when At that time, the proportion of system regulation undertaken by region i is: ,when At that time, take Where N is the number of regional data centers, and the initial target net power for the region is generated accordingly: Finally, feasible region projection and total closure verification are performed on the initial target net power of the region. Projected onto the feasible region: And calculate the target total residual: If r(t) > 0, the residuals are further allocated according to the remaining upward adjustment margin of each region; if r(t) < 0, the residuals are further allocated according to the remaining downward adjustment margin of each region; after projection and residual redistribution, the final regional target net power curve is obtained. The final target net power curve for the region satisfies the feasible region constraint, and under feasible conditions, satisfies: .

5. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 4, characterized in that, The real-time control signal generated based on distributed electricity consumption data includes a control signal composed of a reference control component and a real-time scarcity signal published from the upper layer: ,in, For region i, the reference control component, To provide a real-time scarcity signal reflecting the marginal tension of the distributed energy resource pool, the distributed energy request power reported by aggregators in each regional data center is summarized within each time slot t. Let the distributed energy request power of region i in the k-th round of interaction be . The total number of applications is: To meet the maximum executable power constraint of the distributed energy resource pool By performing feasible region projection on the total application amount, the executable power of the distributed energy resource pool in the current round is obtained: Then, a real-time scarcity signal is generated based on the deviation between the total application amount and the upper limit of distributed energy power. The real-time scarcity signal is iterated according to the projection-dual update rule. ,in, This is the real-time scarcity signal in the k-th round of interaction. To update the step size, This represents the nonnegative projection operator, if an upper limit for the scarcity signal is set. The update formula is: ,when When the total number of applications in each region exceeds the power limit of the distributed energy resource pool, it will be fed back to the application rules of each region in the next round of interaction to curb excessive applications; when As the real-time scarcity signal gradually falls back to the non-negative feasible region, the iterative process satisfies... Or reach the maximum number of iterations Stop when the time slot is reached, and output the real-time scarcity signal for the current time slot. and the executable power z(t) of distributed energy; data center aggregators in each region adjust the power of distributed energy application in the next round based on the updated real-time scarcity signal. ,in, This indicates the basic application items; finally, the regional data center aggregators, given the control signals... and regional target net power curve Under these conditions, the power request for distributed energy resources is generated according to the local response rules. and the set of locally executed instructions.

6. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 5, characterized in that, The process of generating regional-side local scheduling instructions based on real-time control signals includes the regional data center aggregator, after receiving the control signal issued by the distributed integrated energy aggregator coordination layer, calculating the call score for different resource types based on local target deviation and various resource statuses. For resource type r in region i, the resource call score is defined as: ,in, Indicates resource availability. Indicates the cost of resource allocation. This indicates a delay in resource response. Indicates operational risk. This indicates the scarcity signal corresponding to this resource type. to The weighting coefficients are non-negative; the regional side determines the resource allocation priority based on the scoring results and generates adjustment request power for various types of resources. : For computing power tasks in region i at time slot t, define the interactive task arrival quantity. Batch processing task arrival volume Local execution volume and Migration task volume and batch processing delay The IT power consumption in region i is determined by the base power, task execution power, and migration-related power. ,in, Based on the no-load power, Additional power for migration, For server utilization, the actual net power of the region is jointly determined by IT power consumption, cooling power, local renewable energy output, and distributed energy execution power, and is written as: Among them, cooling power This includes the baseline cooling power and the amount of change due to adjustments.

7. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 6, characterized in that, The calculation of tracking deviation and resource pressure status based on the energy allocation status and net power curve under local scheduling instructions includes recording the target net power of region i in time slot t. Actual net power Distributed energy application power Distributed energy execution power And the distributed energy allocation results, the actual net power is expressed as: ,in, To account for the actual cooling power after cooling side adjustment; after the scheduling cycle ends, the target tracking error of each region is recalculated based on the operation record. For region i in time slot t, its time slot tracking error is defined as: To avoid the impact of power magnitude differences in different regions on the deviation evaluation results, a normalized root mean square (RMS) form is used to calculate the regional periodic tracking deviation index, which characterizes the normalized tracking deviation of region i on the target net power curve within the current scheduling period, and is expressed as: Where T is the set of time slots within the scheduling period, and |T| is the number of time slots. To prevent extremely small positive numbers with a denominator of zero; to characterize the resource scarcity of the distributed energy resource pool during the scheduling cycle, and to calculate the system-level application pressure and the proportion of unmet applications at the regional level, let the total application amount for each region in the current time slot be: System-level resource pressure is defined as: in, This represents the maximum executable power of the distributed energy resource pool. The resource pressure trigger threshold is defined as the proportion of unmet applications at the regional level as follows: ,in, The degree to which distributed energy applications in region i are not met is used to represent the overall situation. Finally, by combining system-level resource pressure and the proportion of unmet applications in the region, the regional resource stress is constructed. in, Assuming a weighting coefficient between system-level pressure and unmet regional requests, the average value is calculated over each time slot within the scheduling cycle to obtain the regional periodic resource stress level. After the scheduling cycle ends, the distributed energy resource coordination layer forms a set of cycle operation records: , in, The target negative factor for region i.

8. The distributed energy aggregator dual-time-scale closed-loop scheduling method according to claim 7, characterized in that, The slow-timescale LS-Corr dual-channel feedback regulation based on tracking deviation and resource pressure status also includes introducing the long-term scarce memory generated by the LS channel into the distributed energy allocation priority during the distributed energy allocation process. First, a resource tension gating factor is constructed based on the relationship between the total request amount in the current time slot and the maximum power of the distributed energy. in, The LS-gated trigger threshold is based on and Effective application power in the construction region: ,in, For resource type weights, The distributed energy resource coordination layer uses the LS gating factor for resource type r as the basis for determining the effective power requested in each region. Perform distributed energy power allocation and ensure that the actual power output meets the following requirements: , Finally, based on the scarcity, invocation cost, and target tracking contribution of each resource type in the previous period, the resource type weights are updated. For resource type r, its periodic scarcity is defined as: Its target tracking contribution is defined as: Where ΔΦ_r represents the contribution of resource type r to the improvement of regional target tracking deviation. This represents the call cost of this resource type within the period. The resource type weight update rule is as follows: , Where γ is the weight update step size, and F(·) is the weight correction function; in the next scheduling cycle, the distributed energy resource coordination layer simultaneously uses the fast timescale real-time scarcity signal. The target burden reduction factor generated by the Corr channel And the amount of long-term scarce memory generated by LS channels Perform rolling coordination.

9. A distributed energy aggregator dual-time-scale closed-loop scheduling system, executing the distributed energy aggregator dual-time-scale closed-loop scheduling method as described in claim 1, characterized in that, include: The data acquisition module is configured to acquire distributed power consumption data from the regional data center. The mapping module is configured to perform unified mapping on the acquired distributed electricity consumption data; The net power module is configured to construct the regional target net power curve from the uniformly mapped data. The control module is configured to generate real-time control signals from distributed power consumption data; The instruction module is configured to generate local scheduling instructions on the regional side in real time; The deviation module is configured to calculate and track deviations and resource pressure status based on the energy allocation status and net power curve under local scheduling commands. The feedback module is configured to track deviations and resource pressure status using a slow-timescale LS-Corr dual-channel feedback adjustment. The output module is configured to provide feedback on the adjustment results.

Images