Design of Virtual Reserve Services Based on Spatially Transferable Loads in Data Centers

1. Introduction

The large-scale integration of renewable energy has substantially increased reserve capacity requirements in power systems. Cross-regional reserve resource allocation has therefore become an important issue for maintaining secure and economic system operation. Existing studies have mainly focused on reserve demand assessment, regional reserve sharing, and multi-energy coordinated dispatch. References [1,2,3,4,5] improve reserve allocation efficiency from perspectives such as dynamic reserve configuration and regional joint optimization. However, these studies primarily focus on supply-side flexible resources and have not fully exploited the cross-regional reserve potential of demand-side adjustable loads.

Compared with conventional adjustable loads, data center loads have high power density and strong spatial-temporal transferability. By migrating computing tasks, data centers can reshape the distribution of electricity demand across regions and form cross-regional virtual reserve resources. This provides a new path for alleviating transmission congestion and improving reserve resource utilization. Data centers are therefore important demand-side flexible resources in new power systems, and their participation in reserve optimization and market clearing requires load models and coordination mechanisms that capture spatial-temporal migration characteristics.

For data center load modeling, references [6,7,8] study workload allocation optimization, cross-regional migration, and carbon-oriented regulation, respectively, demonstrating the feasibility of data centers participating in optimized grid operation.

For scheduling optimization and market mechanisms, references [9,10,11] further investigate incentive allocation, computing-power–electricity coordination, and carbon-oriented demand response mechanisms for data centers participating in grid operation. Nevertheless, existing work mainly focuses on demand response, energy consumption optimization, and carbon emission reduction. Joint modeling, joint quantity–price bidding, and market clearing for spatially transferable data center loads participating in cross-regional reserve services remain insufficiently studied. In particular, a declaration and trading framework is still needed for data centers and reserve generators to jointly form virtual reserve resources.

To address these gaps, this paper proposes a virtual reserve service design based on spatially transferable loads in data centers. The main contributions are as follows: (1) a migration capability and constraint model is constructed for data centers; and (2) a thermal generator–data center joint virtual reserve service is designed, together with a joint bidding model for multiple data centers and thermal generating units.

2. Data Center Load Modeling for Virtual Reserve Services

2.1. Migration Capability and Constraint Modeling

The spatial-temporal transferability of data center loads is the basis of virtual reserve services. According to delay sensitivity, data center loads can be divided into online and offline loads. Online loads have limited temporal elasticity and therefore limited virtual reserve potential. Offline loads have stronger temporal flexibility; among them, compute-intensive tasks have relatively low migration costs and high spatial-temporal elasticity, making them the main resources for virtual reserve services.

This paper focuses on spatially migratable compute-intensive tasks. The task migration volume is modeled as a decision variable, and the unit server processing capability is used to characterize task processing capability. Under scheduling periods $t=1,2,\dots ,T$, data center load migration is subject to the following constraints:

$$s·y_{u\to v,t}\le {\overline{B}}_{uv},\forall \left(u,v\right)\in B_{\mathrm{IDC}}^{ij}$$

(1)

$$P_{IDC,i,t}=\sum _{u\in U_i}{\alpha }_{u,t}{q}_{u,t}+{\beta }_{u,t},\forall i\in IDC$$

(2)

$$p_{u,t}=q_{u,t}+\sum _{v:\left(v,u\right)\in \mathcal{E}}{y}_{v\to u,t}-\sum _{v:\left(u,v\right)\in \mathcal{E}}{y}_{u\to v,t}$$

(3)

$$p_{u,t}\le {\overline{p}}_u,\forall u\in U,t\in T$$

(4)

$$p_{u,t}\ge 0,\forall u\in U,t\in T$$

(5)

where s is the bandwidth occupied by a unit task volume; ${\overline{B}}_{uv}$ is the bandwidth limit between data centers u and v; $B_{\mathrm{IDC}}^{ij}$ is the set of connected data center pairs between regions i and j; $P_{IDC,i,t}$ is the power consumption of data centers at node i in period t; ${\alpha }_{u,t}$ and ${\beta }_{u,t}$ are power consumption coefficients; $U_i$ is the set of data centers at node i; $q_{u,t}$ and $p_{u,t}$ are the task arrival and processing volumes of data center u in period t, respectively; ${\overline{p}}_u$ is the maximum task processing volume of data center u; $y_{v\to u,t}$ is the task volume migrated from data center v to data center u; and $y_{u\to v,t}$ is the task volume migrated from data center u to data center v.

Based on these constraints, the cross-regional migration capability of data centers is defined. Consider regions i and j, where the data center sets are $U_i$ and $V_j$, respectively. The migration capability between two data centers and the total interregional migration capability are:

$$M_{u\to v}^{i\to j}\left(t\right)=y_{u\to v,t}·{\alpha }_u,u\in U_i,v\in V_j$$

(6)

$$M_{\mathrm{total}}^{i\to j}\left(t\right)=\sum _{\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}}{y}_{u\to v,t}·{\alpha }_u$$

(7)

where $M_{u\to v}^{i\to j}\left(t\right)$ is the migratable load capacity from data center u in region i to data center v in region j at time t; ${\alpha }_u$ is the power conversion coefficient of data center u; and $M_{\mathrm{total}}^{i\to j}\left(t\right)$ is the total migratable load capacity from region i to region j at time t.

2.2. Cross-Regional Generator–Data Center Joint Reserve Capacity Modeling

The virtual reserve service proposed in this paper migrates data center loads from region i to region j, thereby reducing the load in region i. Meanwhile, reserve generating units in region j support the additional migrated load in region j. This process provides cross-regional virtual reserve support for region i, and the released load in region i is the virtual reserve capacity obtained by region i. The following constraints are established:

$$R_{\mathrm{release}}^{i,u}\left(t\right)={\alpha }_{u,t}·y_{u\to v,t}$$

(8)

$$R_{\mathrm{release}}^i\left(t\right)=\sum _{\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}}{\alpha }_{u,t}·y_{u\to v,t}$$

(9)

$$R_{\mathrm{consume}}^{j,v}\left(t\right)={\alpha }_{v,t}·\sum _{\begin{array}{c}\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}\\ u\in U_i\end{array}}{y}_{u\to v,t}$$

(10)

$$R_{\mathrm{consume}}^j\left(t\right)=\sum _{v\in V_j}{\alpha }_{v,t}·\sum _{\begin{array}{c}\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}\\ u\in U_i\end{array}}{y}_{u\to v,t}$$

(11)

$$R_{\mathrm{consume}}^j\left(t\right)\le \sum _{g\in G_j}{R}_{g,t}^{\mathrm{gen}}$$

(12)

$$s·y_{u\to v,t}\le {\overline{B}}_{uv},\forall \left(u,v\right)\in B_{\mathrm{IDC}}^{ij}$$

(13)

where $R_{\mathrm{release}}^{i,u}\left(t\right)$ and $R_{\mathrm{release}}^i\left(t\right)$ are the releasable load of data center u and all data centers in region i at time t, respectively; $R_{\mathrm{consume}}^{j,v}\left(t\right)$ and $R_{\mathrm{consume}}^j\left(t\right)$ are the additional power consumption caused by migrated tasks at data center v and all data centers in region j, respectively; ${\alpha }_{v,t}$ is the unit task power consumption coefficient of data center v; $G_j$ is the set of reserve generating units in region j; $R_{g,t}^{\mathrm{gen}}$ is the available reserve capacity of unit g; and ${\overline{B}}_{uv}$ is the bandwidth limit between data centers u and v.

Considering these constraints, the maximum reserve capacity optimization model for cross-regional virtual reserve services is:

$$\underset{y_{u\to v,t}}{\max }{R}_{\mathrm{virtual}}^{i\leftarrow j}\left(t\right)=\sum _{\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}}{\alpha }_{u,t}·y_{u\to v,t}$$

(14)

subject to:

$$s·y_{u\to v,t}\le {\overline{B}}_{uv},\forall \left(u,v\right)\in B_{\mathrm{IDC}}^{ij}$$

(15)

$$\sum _{\begin{array}{c}\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}\\ u\in U_i\end{array}}{y}_{u\to v,t}\le {\overline{p}}_v-q_{v,t},\forall v\in V_j$$

(16)

$$\sum _{\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}}{\alpha }_{v,t}·y_{u\to v,t}\le \sum _{g\in G_j}{R}_{g,t}^{\mathrm{gen}}$$

(17)

$$\sum _{\begin{array}{c}\left(u,v\right)\in B_{\mathrm{IDC}}^{ij}\\ v\in V_j\end{array}}{y}_{u\to v,t}\le q_{u,t},\forall u\in U_i$$

(18)

$$y_{u\to v,t}\ge 0,\forall \left(u,v\right)\in B_{\mathrm{IDC}}^{ij}$$

(19)

where $R_{\mathrm{virtual}}^{i\leftarrow j}\left(t\right)$ is the virtual reserve capacity obtained by region i from region j; and $q_{v,t}$ is the task arrival volume of data center v in period t.

3. Virtual Reserve Service Design Based on Spatially Transferable Data Center Loads

3.1. Product Design

This paper proposes a cross-regional virtual reserve consortium (CRVRC) business model that participates in the reserve market through a joint quantity–price bidding mechanism.

3.1.1. Organizational Structure of the CRVRC

The CRVRC consists of a data center operator (DCO) and a reserve generation operator in region j (GO–j). The DCO owns data center facilities in both regions i and j. Contracts within the consortium define the rights and responsibilities of each participant.

The DCO provides the transferable task volume ${\displaystyle \sum _{u\in U_i}\sum _{v\in V_j}}{y}_{u\to v,t}$ from the data center set $U_i$ in region i to the data center set $V_j$ in region j. It receives migrated tasks in region j, provides bandwidth resources ${\overline{B}}_{uv}$, implements data migration, guarantees service quality, shares reserve service revenue, and bears migration and operating costs.

GO–j provides generation reserve capacity ${\displaystyle \sum _{g\in G_j}}{R}_{g,t}^{\mathrm{gen}}$ to support the increased power load in region j. It helps ensure secure grid operation, shares reserve service revenue, and receives reserve capacity payments.

3.1.2. Joint Quantity–Price Bidding Mechanism

The CRVRC participates in the reserve market of region i as a unified entity. Its quantity and price bids follow the principles below.

First, for the joint quantity constraint, the declared reserve capacity $R_{\mathrm{bid}}^{i\leftarrow j}\left(t\right)$ of the consortium must satisfy the solution of Equation (14):

$$R_{\mathrm{bid}}^{i\leftarrow j}\left(t\right)\le R_{\mathrm{virtual}}^{i\leftarrow j*}\left(t\right)$$

(20)

where $R_{\mathrm{virtual}}^{i\leftarrow j*}\left(t\right)$ is the optimal solution of Equation (14) subject to Equations (15)–(19).

Second, for tiered quantity–price bidding, the consortium submits a reserve capacity–price curve for period t to the dispatcher for market clearing:

$$\mathrm{Curv}{\mathrm{e}}_m\left(t\right)=\{\left(R_m^k\left(t\right),{\lambda }_m^k\left(t\right)\right)\mid k=1,2,\dots ,K_m\}$$

(21)

where $R_m^k\left(t\right)$ is the cumulative reserve capacity of tier k of consortium m, and ${\lambda }_m^k\left(t\right)$ is the corresponding bid price.

4. Pricing Method for Virtual Reserve Services

4.1. Comprehensive Cost Analysis of Virtual Reserve Services

The total cost of cross-regional virtual reserve services consists of data center migration cost, opportunity cost, risk cost, and reserve generation cost:

$$C_{\mathrm{total}}\left(t\right)=C_{\mathrm{mig}}\left(t\right)+C_{\mathrm{opp}}\left(t\right)+C_{\mathrm{risk}}\left(t\right)+C_{\mathrm{gen}}\left(t\right)$$

(22)

where $C_{\mathrm{mig}}\left(t\right)$ is the migration cost, including migration action cost and energy cost; $C_{\mathrm{opp}}\left(t\right)$ is the opportunity cost of participating in electricity ancillary services; $C_{\mathrm{risk}}\left(t\right)$ is the risk cost caused by potential service quality degradation during migration; and $C_{\mathrm{gen}}\left(t\right)$ is the reserve generation cost paid by the CRVRC to generation operators.

4.2. Joint Pricing Model for Data Centers and Reserve Generators

Because the reliability of virtual reserve services decreases as capacity increases, operators adopt tiered pricing strategies.

For DCO tiered pricing, the same data center u may appear in multiple transmission paths. The construction of tiers must therefore consider the cumulative occupation of each data center.

As shown in Figure 1, N is the number of transmission paths. ${\overline{y}}_{u_k}^{\mathrm{out},k}\left(t\right)=q_{u_k,t}-{\displaystyle \sum _{n=1}^{k-1}}{y}_{u_n\to v_n,t}^n·{\mathbf{1}}_{\{u_n=u_k\}}$ is the remaining outbound capability of data center $u_k$ in tier k; ${\overline{y}}_{v_k}^{\mathrm{in},k}\left(t\right)={\overline{p}}_{v_k}-q_{v_k,t}-{\displaystyle \sum _{n=1}^{k-1}}{y}_{u_n\to v_n,t}^n·{\mathbf{1}}_{\{v_n=v_k\}}$ is the remaining receiving capability of data center $v_k$ in tier k; ${\overline{y}}_{u_k\to v_k}$ is the bandwidth capacity limit of path $\left(u_k,v_k\right)$; and ${\mathbf{1}}_{\{·\}}$ is an indicator function, which equals one when the condition is satisfied and zero otherwise.

For GO–j tiered pricing, each unit $g\in G_j$ divides its reserve capacity into several tiers according to operating cost and capacity constraints, forming a unit-level tiered bidding curve:

$$\mathrm{Curv}{\mathrm{e}}_g\left(t\right)=\left\{\left(\Delta R_g^n\left(t\right),{\lambda }_g^n\left(t\right)\right)|n=1,2,\dots ,N_g\right\}$$

(23)

where $\Delta R_g^n\left(t\right)$ is the incremental reserve capacity of unit g in tier n; ${\lambda }_g^n\left(t\right)=C_{\mathrm{gen}}\left(t\right)$ is the unit reserve cost of tier n; and ${\lambda }_g^1\left(t\right)\le {\lambda }_g^2\left(t\right)\le \dots \le {\lambda }_g^{N_g}\left(t\right)$, meaning that tier prices are nondecreasing.

As shown in Figure 2, $K_{\mathrm{GO}-\mathrm{j}}={\displaystyle \sum _{g\in G_j}}{N}_g$ is the total number of tiers, and ${\tau }_k=\left(g_k,n_k,\Delta R_k,{\lambda }_k\right)$ denotes the unit, internal tier index, incremental capacity, and unit cost corresponding to tier k.

For the CRVRC tiered pricing, the capacity tiers in the DCO and GO–j quantity–price curves are not in one-to-one correspondence. A curve aggregation algorithm is therefore required to build a unified quantity–price curve for the consortium.

As shown in Figure 3, $K_m\le K_{\mathrm{DCO}}+K_{\mathrm{GO}-\mathrm{j}}+1$ is the number of distinct breakpoints after deduplication. ${\lambda }_{\mathrm{DCO}}\left(b_k,t\right)={\lambda }_{\mathrm{DCO}}^i\left(t\right)$ and ${\lambda }_{\mathrm{GO}-\mathrm{j}}\left(b_k,t\right)={\lambda }_{\mathrm{GO}-\mathrm{j}}^j\left(t\right)$ are the marginal costs of the DCO and GO–j at capacity $b_k$, respectively. They satisfy $R_{\mathrm{DCO}}^{i-1}\left(t\right)<b_k\le R_{\mathrm{DCO}}^i\left(t\right)$ and $R_{\mathrm{GO}-\mathrm{j}}^{j-1}\left(t\right)<b_k\le R_{\mathrm{GO}-\mathrm{j}}^j\left(t\right)$, respectively. ${\lambda }_{\mathrm{profit}}^k\left(t\right)$ is the profit margin of the consortium in tier k; $R_m^k\left(t\right)=b_k$ is the cumulative reserve capacity in tier k; and ${K_m}^{\prime }$ is the number of valid tiers satisfying the capacity constraint. The tiered bids satisfy ${\lambda }_m^1\left(t\right)\le {\lambda }_m^2\left(t\right)\le \dots \le {\lambda }_m^{{K_m}^{\prime }}\left(t\right)$.

If consortium m clears capacity $r_m^k\left(t\right)$ in tier k, the revenue of that tier is:

$$R_{\mathrm{revenue}}^{m,k}\left(t\right)={\lambda }_{\mathrm{clear}}\left(t\right)·r_m^k\left(t\right)$$

(24)

where ${\lambda }_{\mathrm{clear}}\left(t\right)$ is the uniform market clearing price.

Assuming a perfectly competitive market, all parties bid based on cost. The tier-k revenue of the DCO and GO–j is allocated according to their cost contribution ratios:

$$\begin{array}{cc}\hfill R_{\mathrm{revenue}}^{\mathrm{DCO},m,k}\left(t\right)& ={\varphi }_1^{m,k}·R_{\mathrm{revenue}}^{m,k}\left(t\right)\\ \hfill R_{\mathrm{revenue}}^{\mathrm{GO}-\mathrm{j},m,k}\left(t\right)& ={\varphi }_2^{m,k}·R_{\mathrm{revenue}}^{m,k}\left(t\right)\end{array}$$

(25)

where ${\varphi }_1^{m,k}+{\varphi }_2^{m,k}=1$.

The allocation coefficients of tier k are:

$$\begin{array}{c}{\varphi }_1^{m,k}={\displaystyle \frac{{\lambda }_{\mathrm{DCO}}\left(b_k,t\right)}{{\lambda }_{\mathrm{DCO}}\left(b_k,t\right)+{\lambda }_{\mathrm{GO}-\mathrm{j}}\left(b_k,t\right)}},\\ {\varphi }_2^{m,k}={\displaystyle \frac{{\lambda }_{\mathrm{GO}-\mathrm{j}}\left(b_k,t\right)}{{\lambda }_{\mathrm{DCO}}\left(b_k,t\right)+{\lambda }_{\mathrm{GO}-\mathrm{j}}\left(b_k,t\right)}}\end{array}$$

(26)

The total revenue of each party in the consortium is:

$$\begin{array}{cc}\hfill R_{\mathrm{revenue}}^{\mathrm{DCO},m}\left(t\right)& ={\displaystyle \sum _{k=1}^{{K_m}^{\prime }}}{R}_{\mathrm{revenue}}^{\mathrm{DCO},m,k}\left(t\right)\\ \hfill R_{\mathrm{revenue}}^{\mathrm{GO}-\mathrm{j},m}\left(t\right)& ={\displaystyle \sum _{k=1}^{{K_m}^{\prime }}}{R}_{\mathrm{revenue}}^{\mathrm{GO}-\mathrm{j},m,k}\left(t\right)\end{array}$$

(27)

5. Simulation Analysis

5.1. Simulation Model Design

To verify the economics of cross-regional virtual reserve services, a two-region virtual reserve simulation model is constructed [12,13].

Assume that M cross-regional virtual reserve consortia $\mathcal{M}=\left\{1,2,\dots ,M\right\}$ and N traditional reserve generating units in region i, $G_{\mathrm{trad}}^i=\left\{1,2,\dots ,N\right\}$, participate in market bidding. Only virtual reserve services from region j to region i are considered. After all participants submit tiered bidding curves, market clearing is optimized by minimizing the total cost in region i [14].

The objective function of the simulation model is:

$$\min \sum _{m\in \mathcal{M}}{\displaystyle \sum _{k=1}^{K_m}}{\lambda }_m^k\left(t\right)·r_m^k\left(t\right)+\sum _{g\in G_{\mathrm{trad}}^i}{\displaystyle \sum _{n=1}^{N_g}}{\lambda }_g^n\left(t\right)·r_g^n\left(t\right)$$

(28)

where $\mathcal{M}$ is the set of all CRVRCs; $r_m^k\left(t\right)$ and ${\lambda }_m^k\left(t\right)$ are the cleared reserve capacity and bid price of tier k of consortium m, respectively; $G_{\mathrm{trad}}^i$ is the set of traditional reserve generating units in region i; and ${\lambda }_g^n\left(t\right)$ and $r_g^n\left(t\right)$ are the unit reserve cost and reserve capacity of tier n of traditional reserve unit g, respectively.

The constraints are:

$$\sum _{m\in \mathcal{M}}{\displaystyle \sum _{k=1}^{K_m}}{r}_m^k\left(t\right)+\sum _{g\in G_{\mathrm{trad}}^i}{\displaystyle \sum _{n=1}^{N_g}}{r}_g^n\left(t\right)\ge D_i^{\mathrm{res}}\left(t\right)$$

(29)

$$r_m^k\left(t\right)\ge 0r_g^n\left(t\right)\ge 0$$

(30)

$$r_m^k\left(t\right)\le R_m^k\left(t\right)-R_m^{k-1}\left(t\right),\forall m,k,(R_m^0=0)$$

(31)

$$\sum _k{r}_m^k\left(t\right)\le R_m^{\max }\left(t\right),\forall m\in \mathcal{M}$$

(32)

$$r_g^n\left(t\right)\le \Delta R_g^n\left(t\right),\forall g\in G_{\mathrm{trad}}^i,\forall n=1,..,N_g$$

(33)

where $D_i^{\mathrm{res}}\left(t\right)$ is the reserve demand of region i at time t; $R_m^k\left(t\right)$ is the cumulative reserve capacity of tier k of consortium m; $R_m^{\max }\left(t\right)=\min \left\{R_{\mathrm{DCO},\mathrm{m}}^{K_{\mathrm{DCO},\mathrm{m}}}\left(t\right),R_{\mathrm{GO}-\mathrm{j}}^{K_{\mathrm{GO}-\mathrm{j}}}\left(t\right)\right\}$ is the maximum available reserve capacity of the consortium; and $\Delta R_g^n\left(t\right)$ is the maximum available reserve capacity of tier n of traditional reserve unit g at time t.

The clearing price is:

$${\lambda }_{\mathrm{clear}}\left(t\right)=max\{\underset{m,k:r_m^k>0}{\max }{\lambda }_m^k\left(t\right),\underset{g,n:r_g^n>0}{\max }{\lambda }_g^n\left(t\right)\}$$

(34)

The revenue obtained by winning consortium m in tier k is:

$$R_{\mathrm{revenue}}^{m,k}\left(t\right)={\lambda }_{\mathrm{clear}}\left(t\right)·r_m^k\left(t\right)$$

(35)

5.2. Simulation Scenario Settings

The scenario settings are shown in Table 1, and the ranges of key parameters are shown in Table 2. Detailed parameter settings are provided in Appendix A. The reserve demand is uniformly set to 100 MW, and ${\lambda }_{\mathrm{profit}}^k\left(t\right)$ is uniformly set to 0.5.

5.3. Simulation Results

5.3.1. Scenario 1: Market Clearing Results in the Baseline Scenario

As shown in Figure 4, Figure 5 and Figure 6, the clearing price is 68 CNY/MWh. Units G1–1 and G1–2 of S1 clear 40 MW and 35 MW, respectively, accounting for 75% in total. Unit G2–1 of S2 clears 25 MW, accounting for 25%. The total supplier revenue is CNY 6800, including CNY 5100 for S1 and CNY 1700 for S2. The total supplier cost, here and below referring to cost bids, is CNY 5640, including CNY 4060 for S1 (72.8%) and CNY 1580 for S2 (27.2%). This scenario serves as the baseline for comparing the effects of introducing virtual reserve services.

5.3.2. Scenario 2: Market Clearing Results with a Single CRVRC

As shown in Figure 7, when the reserve demand is 100 MW, the clearing price is 58.00 CNY/MWh. CRVRC–1 clears its full capacity of 44.20 MW, and S1 clears 55.80 MW. Thus, CRVRC–1 is cleared first, while S1 supplements the remaining demand.

As shown in Figure 8, the total service cost is CNY 4661.60. CRVRC–1 and S1 bear costs of CNY 1745.2 and CNY 2916.4, respectively. The revenue of CRVRC–1 is CNY 2563.6, while the revenue of S1 is CNY 3236.4.

5.3.3. Scenario 3: Market Clearing Results with Multiple CRVRCs

As shown in Figure 9, when the reserve demand is 100 MW, the clearing price is 43.01 CNY/MWh. CRVRC–1 clears its full capacity of 44.20 MW, CRVRC–2 clears 35.80 MW, CRVRC–3 clears 20.00 MW, and S1 does not clear. Compared with the clearing price of 58 CNY/MWh in Scenario 2, competition among multiple consortia reduces the clearing price by 14.99 CNY/MWh, corresponding to a decrease of 25.8%.

As shown in Figure 10, the costs and revenues of CRVRC–1 are CNY 1745.2 and CNY 1901.07, respectively; those of CRVRC–2 are CNY 1503.51 and CNY 1539.78, respectively; and those of CRVRC–3 are CNY 845.51 and CNY 860.21, respectively. The total system cost is CNY 4094.23, which is 12.2% lower than that in Scenario 2. However, marginal profits decline across the three consortia, and CRVRC–2 and CRVRC–3 approach zero profit. This indicates that competition among consortia significantly compresses profit margins, while the high bid of S1 causes it to be fully displaced from the market. Traditional reserves therefore face displacement risk when virtual reserve supply is sufficient.

6. Conclusions

This paper proposes a cross-regional virtual reserve service method based on spatially transferable loads in data centers. A data center migration model considering bandwidth, power, and task processing constraints is constructed to quantitatively characterize cross-regional migratable loads. Based on this model, a CRVRC market mechanism integrating data centers and reserve generators is designed. Joint quantity–price bidding and tiered pricing are used to enable unified reserve resource clearing.

The simulation results show that the proposed method can effectively reduce system reserve costs and improve market clearing efficiency in a multi-participant competitive environment while maintaining the basic revenue of participating entities.

Future work will focus on incorporating power flow and transmission congestion constraints, as well as modeling data center reserve reliability and task uncertainty, to further improve the engineering applicability and robustness of the proposed model.