Paper notes – A systematic review of the costs and impacts of integrating variable renewables into power grids

Paper #

Heptonstall, P.J. & Gross, R.J.K. (2021) A systematic review of the costs and impacts of integrating variable renewables into power grids. Nature Energy. 6 (1), 72–83. Available from: doi: 10.1038/s41560-020-00695-4.

Abstract #

The impact of variable renewable energy (VRE) sources on an electricity system depends on technological characteristics, demand, regulatory practices and renewable resources. The costs of integrating wind or solar power into electricity networks have been debated for decades yet remain controversial and often misunderstood. Here we undertake a systematic review of the international evidence on the cost and impact of integrating wind and solar to provide policymakers with evidence to inform strategic choices about which technologies to support. We find a wide range of costs across the literature that depend largely on the price and availability of flexible system operation. Costs are small at low penetrations of VRE and can even be negative. Data are scarce at high penetrations, but show that the range widens. Nonetheless, VRE sources can be a key part of a least-cost route to decarbonization.

Notes #

• This paper is not directly about sustainable computing but covers a relevant topic – what happens to the electricity grid as more renewables are added? Wind and solar energy is variable – the output can change relatively quickly – and is not dispatchable – you can’t choose when it generates. This poses operational challenges as we add more and more to the grid.
• This is relevant to computing, and data centres in particular, because compute workloads generally need highly reliable sources of electricity. Data centres are designed with multiple systems of redundancy which include batteries and backup generators. If the grid becomes more unstable as renewables increase, this poses a problem for data centre operators.
• Renewables penetration is expected to increase over the coming years. For example, in Ireland and Denmark, they expect wind and solar to form the majority of generation capacity by 2030.
  • Currently in Ireland and Northern Ireland, the maximum energy from wind and solar is 65%, but this is going to increase to 75% by the end of 2021 and 95% by 2030. Ireland’s physical location at the edge of Europe and the weather caused by the jet stream result in forecast challenges where errors can affect a significant part of demand. Ramping margins are being used to bridge this gap (source).
  • In Denmark they expect to be at 54% renewables by 2030, but electricity generation will be 109% renewables by 2030 (which means domestic production exceeds domestic consumption, so Denmark will be a large net exporter of electricity from 2026 onwards) (source).
• This is a challenge because a large part of the demand growth on the grid in Ireland and Denmark is expected to come from data centres.
  • Ireland: Median demand scenario projects data centres making up to 27% of total grid demand by 2029, outpacing industrial users. Typical load is 40% of contracted Maximum Input Capacity, which is expected to grow as data centres build out (source).
  • Denmark: Data centres grow from 1% to 15% of total energy demand, 7 TWh, by 2030. Growth in industry and services is 80% from data centres (source).
• Some see this as an opportunity for data centres to participate in demand response but others see these grid challenges as a reason to be cautious about renewables for data centres – issues of grid reliability are not well understood by energy buyers, and claims have been made that falling spot prices hurt the business models of energy businesses and their ability to invest in reliability.
• The review by Heptonstall & Gross looks at a significant body of literature to examine the cost of increasing the penetration of renewables on the grid. These costs cover various categories such as systems integration, transmission and distribution, curtailment and reserve costs. The results show that up to around 50% penetration the costs are relatively small. However, past 50% there is limited evidence, probably because few electricity grids have reached such a high level of renewables penetration.

• Data centre demand response has been a topic of academic interest for over a decade. Data centres could participate in “ancillary service markets” as a load resource to help with the balance of supply and demand e.g. increasing/decreasing demand (and being paid to do so) (Wierman et al., 2014). Past models have focused on profit (Ghamkhari & Mohsenian-Rad, 2012; Liu et al., 2014) but is that the motivation for data centres?
• Field tests were carried out in 2012 – not particularly large participants (550 kW – 2.3 MW) – but they showed up to 25% demand savings available. Also showed similar data centres can participate with no damage to SLAs (Ghatikar et al., 2012). However, there are challenges with market design because markets assume customers are price takers whereas data centre loads are so high they are often price makers (Liu et al., 2014).
• Given that the demand response flexibility for a small (30MW) data centre was modelled to be worth between $500,000 – $5,000,000 almost a decade ago (Wierman et al., 2014), and data centres have become larger and more automated/sophisticated, why has demand response from data centres not become more popular? My thoughts are that:
  • Reliability is a higher priority than energy efficiency.
  • Timing of demand response from the grid do not correlate well with periods of high load for the DC.
  • Participating in the market for these systems is very complex and difficult to incorporate into the data centre management systems.
• Participating in demand response is difficult for colo providers (Ren & Islam, 2014) – colo pricing is based on peak power rather than actual usage – but what is the impact of hyperscale cloud where design guidelines assume failure?
• “The cloud” is an abstraction of physical resources. Architecture best practices for cloud applications always emphasise the importance of assuming failure. If cloud zones and regions are used correctly, they could allow the movement of demand based on signals, such as from spot instance prices. Hyperscale providers operate at such a scale that the cost savings from the grid could become worth it.
• Either way, more research is needed to fully understand the impact of high renewables penetration on energy systems. For data centres in particular, this means examining grid reliability and market design to encourage participation in demand response.