Virtual Transmission Lines from Data Center Workload Migration: Coordinated Multi-Year Transmission Expansion Planning

Sets and indices

• Grid and planning notation covers buses, existing and candidate lines, generators, years, representative days, and time intervals.It also identifies line flows, phase angles, construction-start indicators, and in-service indicators.Source: Nomenclature · pp. 1–2

Data-center and communication quantities

• Data-center notation distinguishes executed workload, migrated workload, workload capacity, migratable fraction, and raw demand.The glossary also names the computing-to-grid conversion factor, workload profile, and data-transfer parameters.Source: Nomenclature · pp. 1–2

Costs and planning parameters

• Economic notation includes transmission-investment and operating costs, generator costs and emissions, carbon pricing, and penalties for curtailment or unserved workload.The glossary also defines annual investment budgets, discounting, and migration-cost components.Source: Nomenclature · p. 2

Motivation

• Data-center demand growth and geographically concentrated loads contrast with slower transmission expansion and dispersed wind and solar resources.The resulting transmission constraints can leave data-center demand unserved while renewable generation is curtailed; physical expansion remains necessary, but flexible demand can help use the existing network during the transition.Source: §1 · p. 2

Related work

• Data centers can shift delay-tolerant computing tasks, and prior work has studied their operational flexibility and effects on congestion and renewable curtailment.Most network-constrained operational studies treat network investment as fixed.Source: §1 · pp. 2–3

Research gap

• Prior planning studies have not systematically examined how spatial workload flexibility affects candidate-line choice, construction timing, and staged commissioning under communication constraints.The gap concerns growing AI-driven demand, transmission construction lead times, and migration limits from bandwidth and latency.Source: §1 · p. 3

Study approach

• The study co-optimizes workload migration and staged transmission expansion, comparing four resource cases and a fixed-expansion counterfactual.Sensitivity analysis examines when VTL is valuable and when physical reinforcement remains necessary.Source: §1 · p. 3

2. Transmission expansion planning model with data-center virtual transmission resources

• The framework jointly plans line selection and construction timing with workload reallocation subject to workload-balance, bandwidth, and latency constraints.Only physical transmission lines carry electrical energy; VTL supplies spatial demand flexibility without transferring electricity.Source: §2 · p. 3, Figure 1 · p. 4

2.1. Model assumptions

• The model uses DCPF for active-power transfer and omits line losses, reactive power, and bus-voltage constraints; generators have output limits and linear costs but no unit-commitment or ramping constraints.Renewable availability is represented by profiles, with unused output curtailed and penalized.Source: §2.1 · pp. 3–4

2.1. Model assumptions; 2.2. Sets and decision variables

• Data centers are aggregated as power-equivalent workloads with spatial migration, while joint temporal and spatial flexibility is outside scope and unserved load is permitted at a high penalty.The model’s sets, indices, variables, and parameters are summarized in the Nomenclature.Source: §2.1 · p. 4, §2.2 · p. 4

2.3.1. Multi-year load, renewable, and data-center demand growth; 2.3.2. Conventional generator output constraints; 2.3.3. Renewable utilization and curtailment constraints

• Annual growth relations scale conventional load, renewable availability, data-center workload, and executable capacity, while generators obey output bounds and renewable output is either used or curtailed.Source: §2.3.1 · p. 4, §2.3.2 · p. 5, §2.3.3 · p. 5

2.3.4. DCPF constraints for existing transmission lines

• Existing-line flows follow the DCPF relation and remain within line capacity.Source: §2.3.4 · p. 5

2.3.5. Candidate-line construction and service-status constraints

• Candidate lines become available only after their construction duration, can be built at most once, and are fixed unavailable in cases without expansion.Source: §2.3.5 · p. 5

2.3.6. Candidate-line flow constraints

• Candidate-line flows are zero before commissioning and obey DCPF and capacity constraints in service, linked to service status through a Big-M formulation.The candidate-line coefficient and Big-M constant are specified for this formulation.Source: §2.3.6 · p. 5

2.3.7. Reference-bus phase-angle constraints

• A reference bus is fixed at zero phase angle to make the DCPF solution unique, and all bus angles obey absolute limits.Source: §2.3.7 · p. 6

2.3.8. Data-center workload migration and executed-load constraints

• Executed local workload equals raw demand adjusted for unserved load and inbound or outbound migration, and must remain nonnegative and within data-center capacity.Unserved workload cannot exceed raw demand.Source: §2.3.8 · p. 6

2.3.9. Data-center migration-flexibility constraints; 2.3.10. Communication bandwidth and latency constraints

• Outbound migration is limited by the migratable fraction; communication links constrain migration rates by bandwidth and are disabled above the latency threshold.Communication energy is monetized in migration cost rather than added as electric load at a grid node.Source: §2.3.9 · p. 6, §2.3.10 · p. 6

2.3.11. Nodal power-balance constraints; 2.3.12. Annual investment budget constraints

• Nodal power balance equates generation with conventional and data-center load plus net outgoing flow, while candidate-line investment may be capped by an annual budget.VTL changes executed load across nodes and reshapes physical line flows without changing grid topology or adding a physical line.Source: §2.3.11 · p. 6, §2.3.12 · p. 7

2.4. Objective function and solution method

• The objective minimizes total discounted multi-year cost through transmission investment and representative-day operating costs.Operating costs include generation, emissions, renewable curtailment, unserved data-center load, workload migration, and communication.Source: §2.4 · p. 7
• Unit migration cost combines base workload-migration cost, communication energy cost, and data-traffic cost.Source: §2.4 · p. 7
• Physical line construction and workload migration are jointly optimized under grid-operation, construction, service-quality, and communication-network constraints.Source: §2.4 · p. 7

3.1. Test system

• The modified IEEE 14-bus system places wind and DC2 at bus 8, photovoltaic generation and DC3 at bus 3, and DC1 at load-center bus 9.Candidate lines have a two-year construction lead time and carry no power until that delay elapses.Source: §3.1 · p. 7, Figure 2 · p. 8

3.2. Data

• Twelve representative power-system days are selected from 2019–2021 seasonal profiles, with weights summing to 365 days per year.The procedure prioritizes high-renewable days, then high-load days among those remaining, and selects observed days nearest each class mean.Source: §3.2 · p. 8
• Data-center computing profiles are estimated from Alibaba GPU-cluster traces using instance CPU/GPU usage and machine and GPU power specifications.Instance resource use and lifecycle records are linked to estimate average computing power.Source: §3.2 · p. 8
• Instance energy is allocated across overlapping 15-minute intervals, and average-power series are normalized by their full-period 95th percentile.The common reference preserves relative differences among days while limiting extreme observations; the normalized profiles are grouped into low-, normal-, and peak-load representatives.Source: §3.2 · pp. 8–9
• The trace represents CPU–GPU computing power rather than total facility demand; the nodal balance converts executed workload to grid-side demand using a composite factor for auxiliary equipment and facility overheads.Source: §3.2 · p. 9

3.3. Case settings

• Four cases compare neither resource, VTL only, build-only, and coordinated VTL-aware build; a fixed-build counterfactual enables VTL operation on the build-only plan.The unserved-load penalty is high, and sensitivity analysis also reports physical unserved-load reductions because the penalty can mechanically affect cost savings.Source: §3.3 · pp. 9–10
• The four-case comparison is also run on an IEEE 118-bus network over five years, with six data centers and 20 candidate corridors.The larger-system setup includes 12 representative days, 19 conventional generators, and six renewable plants.Source: §3.3 · p. 10

4.1. Overall planning performance

• C3 lowers total discounted cost from 13.850 to 11.987 billion yuan, unserved DC load from 168.80 to 80.42 GWh, and renewable curtailment from 56.37 to 1.86 GWh versus C2.C3 builds three lines rather than four and reduces discounted investment from 0.706 to 0.553 billion yuan; migration reallocates demand toward nodes with less curtailment and corridor stress.Source: §4.1 · p. 10, Table 2 · p. 10
• In C3, the three corridors shared with C2 start construction one year earlier, while C2’s line 7–9 is omitted.Figure 3 bars span construction start to in-service year; the cited text reports the schedule shifts and omitted corridor.Source: §4.1 · p. 11, Figure 3 · p. 11

4.2. Annual trajectories

• C1’s unserved DC load rises rapidly in later years as the no-build network reaches capacity; C3 maintains much lower curtailment and unserved load by coordinating migration with commissioned lines.Communication energy remains below 0.6 GWh/year in the plotted runs, while normalized carbon emissions decline over the horizon and are lowest for C3 in later years.Source: §4.2 · p. 11, Figure 4 · p. 11

4.3. Congestion mechanism

• With the C2 construction plan fixed, VTL lowers average loading from 46.73% to 43.50%; congestion counts for C3 are not a direct measure of VTL’s operating effect.C2 and C3 differ in construction, so the fixed-plan counterfactual isolates nodal load redistribution; counts can also reflect differences in served load and built lines.Source: §4.3 · p. 12, Table 3 · p. 12

4.4. Sensitivity analysis

• Reducing line capacity from 1.2 to 0.8 times baseline raises the absolute discounted cost saving from 0.695 to 9.287 billion yuan.At twice baseline line cost, C3 builds no line and leaves 454.46 GWh unserved; conventional-load growth and greater DC flexibility also increase VTL benefits, with flexibility gains flattening when constraints bind.Source: §4.4 · pp. 12–13
• As annual renewable growth rises from 0 to 10%, C3’s absolute curtailment reduction increases from 28.28 to 151.86 GWh; carbon-price changes leave selected plans unchanged across tested points.The percentage curtailment reduction declines at the highest renewable-growth setting because the C2 denominator changes; absolute C2–C3 cost saving shifts from 1.864 to 1.860 billion yuan across the carbon-price sweep.Source: §4.4 · p. 13
• Bandwidth, data intensity, and communication cost have little effect over tested ranges, but a 10-ms latency limit removes all migration links and eliminates VTL benefit.All modeled links have 20-ms latency; limits of 20 ms or more retain all six links and restore the baseline response. Figure 5 compares three benefit metrics against C2, while Figure 6 includes discrete link availability.Source: Figure 5 · p. 12, §4.4 · pp. 13–14, Figure 6 · p. 13

4.5. Validation on the IEEE 118-bus system

• On the five-year IEEE 118-bus case, C3 reduces total discounted cost from 44.25 to 43.35 billion yuan and unserved DC load from 39.53 to 7.34 GWh versus C2.C3 builds two candidate lines rather than three; no renewable curtailment occurs in any case, and VTL does not fully replace physical reinforcement.Source: §4.5 · p. 14

• Coordinated VTL-aware planning cuts eight-year discounted cost 13.45%, unserved DC load 52.4%, and renewable curtailment 96.7%, while reducing expansion from four lines to three.Source: §5 · p. 14
• Migration is most valuable under tight transfer capacity, expensive construction, rapid conventional-load growth, difficult renewable delivery, or greater workload flexibility, but sustained DC-load growth still requires physical expansion.Source: §5 · p. 14
• Future work will coordinate temporal task deferral with spatial migration and incorporate higher-resolution workload and communication data.Source: §5 · p. 14

• Complete existing-line and candidate-line data are listed, including reactance, ratings, and undiscounted candidate-line investment costs.Source: Table A1 · p. 14, Table A2 · p. 15
• The input tables provide conventional-generator data and bus base loads with complete data-center parameters.Source: Table A3 · p. 15, Table A4 · p. 15
• The appendix also lists directed communication-link data, representative-day weights for annual scaling, and optimized candidate-line construction schedules.Source: Table A5 · p. 15, Table A6 · p. 15, Table A7 · p. 15

• The appendix lists the sensitivity parameters and tested settings.Source: Table A8 · p. 16
• Overall congestion counts for cases C0–C3 across the eight-year horizon are reported in Appendix Table A9.Source: Table A9 · p. 16
• Sensitivity tables cover line capacity and candidate-line investment cost, renewable growth and availability, and carbon price.Source: Table A10 · p. 16, Table A11 · p. 16, Table A12 · p. 16