A Guide to Reducing Carbon Emissions through Data Center Geographical Load Shifting

Overview

Recent computing needs have led technology companies to develop large-scale, highly optimised data centres, which are large loads on electric power networks with the unique flexibility to shift load both geographically and temporally. Electricity markets are operated by independent system operators (ISOs) to minimise generation cost without direct consideration of carbon emissions. Because market clearing accounts for congestion and power-flow physics, the carbon emissions associated with electricity use vary between (potentially geographically close) locations. Lambda-CO2 is derived from DC optimal power flow sensitivities.

The test is a full year on the 73-bus IEEE RTS-GMLC system (158 generators, 120 lines) with four 250 MW data centres, each allowed to shift up to 20% (50 MW) of its load, at most 3.92% of total system load. The data centres used 8,784,000 MWh, 19.6% of the 44,821,000 MWh annual load. As a benchmark, an ISO that controls the flexibility while minimising cost was also compared. Actual annual results versus the no-flexibility baseline of 13,964,000 t CO2: lambda-CO2 shifting -1.44%; LMP shifting -0.68%; average-carbon shifting -0.68%; ISO-controlled flexibility (cost-minimising) -0.65%; excess-low-carbon shifting +0.14%.

Lambda-CO2 shifting saved about 0.14 t CO2 per MW shifted and still reduced generation cost by 0.27%; LMP shifting cut cost by 1.07% but saved less carbon. Even with full system information, ISO use of data-centre flexibility under a cost-only objective cut emissions by only 91 t, 0.65%; adding a light carbon weight (alpha about 0.1) gave large emission reductions with little cost increase. Every metric's predicted savings exceeded the actual savings after re-dispatch, because the linear sensitivity assumes the binding constraints stay the same; the excess-low-carbon metric predicted a 14,000 t reduction but produced an increase.

Regional average and excess-clean-power signals ignore intra-regional transmission congestion: a data centre in a region with surplus clean generation may be unable to access it because of line limits. The conclusion is that lambda-CO2 gave the largest emission cut, about twice the ISO benchmark, while shifting toward regions with 'excess clean power' increased emissions because of transmission constraints. Limits: a single test system (IEEE RTS-GMLC) and four-site setup; it is a preprint.

Key points

1. Test system: IEEE RTS-GMLC (73 buses, 158 generators, 120 lines) with four 250 MW data centres using 8,784,000 MWh, 19.6% of the 44,821,000 MWh annual load; each could shift up to 20% (50 MW), at most 3.92% of total system load.
2. Against the no-flexibility baseline of 13,964,000 t CO2: lambda-CO2 shifting -1.44%; LMP -0.68%; average-carbon -0.68%; ISO-controlled (cost-minimising) -0.65%; excess-low-carbon +0.14%.
3. Lambda-CO2 shifting saved about 0.14 t CO2 per MW shifted and still reduced generation cost by 0.27%; LMP shifting cut cost by 1.07% but saved less carbon.
4. Even with full system information, ISO use of data-centre flexibility under a cost-only objective cut emissions by only 91 t, 0.65%; a light carbon weight (alpha about 0.1) gave large emission reductions with little cost increase.
5. Every metric's predicted savings exceeded the actual savings after re-dispatch; the excess-low-carbon metric predicted a 14,000 t reduction but produced an increase.
6. Regional average and excess-clean-power signals ignore intra-regional transmission congestion; a data centre in a region with surplus clean generation may be unable to access it because of line limits.

Using it in design

• For geographic load shifting, consider locational marginal carbon emissions (lambda-CO2) as the signal instead of LMP or regional average carbon intensity; it gave the largest reduction (-1.44%).
• Avoid shifting toward 'excess clean power' regions; transmission congestion can make it raise emissions (+0.14%).
• Verify predicted carbon savings against savings realised after re-dispatch; every metric over-predicted.
• Consider adding a light carbon weight (alpha about 0.1) to the cost objective; it gave large emission reductions with little cost increase.

Who should read it

Operators, grid planners and researchers choosing a data-centre geographic load-shifting strategy and signal.

Document contents

1. Abstract
2. I. Introduction
3. II. Load shifting methods (A. Data center-controlled load shifting, B. Shifting Metrics, C. ISO-controlled load shifting)
4. III. Test system and scenarios
5. IV. Results (A. Carbon Emissions Reduction from ISO-controlled Load, B. Comparison of Outcomes with Different Shifting Metrics, C. Comparison of Accuracy)
6. V. Discussion and conclusion
7. VII. Acknowledgements

Related standards

• GHG Protocol Scope 2 Guidance

Read the document

Publication details

Authors

Julia Lindberg; Yasmine Abdennadher; Jiaqi Chen; Bernard C. Lesieutre; Line Roald

Publisher

Proceedings of the Twelfth ACM International Conference on Future Energy Systems (ACM)

Year

2021

Document number

arXiv:2105.09120

Pages

430-436

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.1145/3447555.3466582

Licence

CC BY 4.0 — share with attribution

Licence evidence

arxiv.org

Citation