Using the cloud to balance the grid
Accommodate renewable sources while avoiding negative pricing
Human activities involve the use of electricity, and supply and demand vary on a daily basis. On the supply side, the increasing use of less predictable renewable energy sources like solar and wind results in increased price fluctuations and volatility. For grid operators, it complicates the job of maintaining the balance between supply and demand — crucial to prevent outages and keep voltages within acceptable limits.
As can be seen from the graph, any given country will have a base level of energy usage — around 40GW in case of Germany —and a varying demand (around 30GW). Power is preferably sourced from renewable sources, but when the supply is insufficient conventional sources are added.
In 2020, in some European countries the combination of reduced energy demand (due to COVID19) with the increased volatility of renewable energy sources led to negative energy prices ( during low demand and high supply ). The exact causes for these negative prices are complex, and have to do with the limited flexibility of the power system. The frequency of these effects can be expected to increase, as the percentage of renewable sources increases over time (as required to meet emission reduction objectives).
As illustrated in the graph for April 2020, negative prices result in higher peak demand as consumers with variable contracts get paid to use energy. The German market affects other markets in Europe; the linked article from 2020 explains how the pricing structure actively inhibits a shift towards decarbonization, as gas is much cheaper than electricity.
The bigger picture: Import/export in Europe and copper plate models
Power grids are interconnected across national borders, and international import/export of energy forms an important part of the balancing equation.
For Germany, import and export levels can go up to +/- 8 GW across all neighboring countries, with high export levels correlated with low electricity price (i.e. low local demand).
Physics versus economics and the “copper plate” approach
As powerfully illustrated below, the reality of energy trading versus physical flows of power has some interesting dynamics.
Demand side management (DSM) is complicated by the ‘copper plate approach’ followed by Distribution System Operators (DSOs); this regulation system is anchored in European law, and assumes that the transmission lines have sufficient capacity. In practice, this leads to millions of euros in congestion payments from grid operators to producers, when the grid is unable to accommodate their supplied power; these costs are ultimately paid by consumers.
Energy grid regulations and economic structures will have an enormous impact on the feasibility and ultimate success of demand management schemes as discussed here.
Coarse-grained power supply and inefficiencies
In Germany, power is provided by a number of operators, each operating a number of power plants. For example, Uniper operates 13 facilities providing 10.5 GW of total capacity.
During times of low renewable supply, each operator has to generate supplementary power using conventional plants. Given the dynamics of supply and demand, it seems unavoidable that one or more plants will be running at less than maximum capacity, at least part of the time. As power plants are designed to maximize efficiency at full capacity, running at lower capacity generates higher-than-average emissions per unit of energy.
Demand side: Data center energy usage
Data centers form a stable and growing form of energy demand.
In 2020 the annual energy consumption was about 16 TWh, or 44 GWh per day (roughly 2 GW per hour assuming 24x7 operations). This energy was used across roughly 450 sites.
Designs for new hyper-scale data centers (like the one for Meta in Zeewolde, The Netherlands) can go up to 200MW; a rack of servers can be assumed to consume about 20kWh (although higher densities up to 100kWh are possible with liquid cooling)
Edge data centers
While definitions vary, for the purpose of this article an edge data center is assumed to be a relatively small site with a few racks of servers. An example is DataQube, who offers self-sufficient pods with low Power Usage Efficiency (PUE) which implies that most of the power is used for IT workloads.
Distributed servers
In another article a distributed deployment of repurposed servers for heat reuse is described, in the form of combined heating/compute appliances in homes and office buildings. In case of hot water boilers, such heating appliances could be orchestrated to warm their water tank at times when it is convenient for the grid (e.g. sometimes at night, sometimes when its sunny or windy)
A note on Power Purchase Agreements (PPAs)
Large data center operators procure power in bulk using so-called Power Purchase Agreements (PPAs). In 2019 and 2020 180 corporate PPAs were signed across Europe, pricing structures and conditions (including settlement intervals) have a major effect on the incentives to time the use of power in support of grid load balancing.
PPAs can have various settlement intervals; for the purpose of grid balancing, only a shorter interval (e.g. hourly) can provide an economic incentive to shift workloads to off-peak periods. Some companies use “Virtual PPAs” — financial instruments without corresponding physical delivery of power — to offset the use of conventional power sources (e.g. due to lack of availability of renewable sources at the location of usage)
Similarly, some companies are working on hourly Energy Attribute Certificates (EACs) — also known as Renewable Energy Certificates (RECs) in the US and Guarantees of Origin (GOs) in Europe — to create a secondary market that promotes investment in balanced renewable energy resources.
Compute infrastructure as dynamic grid load
Servers only use power when they are switched on, and they can be switched off at relatively short notice. Provided that suitable applications and Service Level Agreements (SLAs) are established with users of compute resources, workloads could be shifted in time (locally) and/or moved to a different (international) location where renewable energy sources are available.
For the purpose of grid management, it does not matter where the servers are located; the dynamic load can be provided as a combination of hyper-scale data centers, a set of smaller edge sites, and individual appliances in households.
For select applications, multiple international power grids could cooperate to shift workloads where local grid conditions are optimal. For this article, a single power grid is assumed. Given some assumptions on peak demand and the percentage of usage that could be allocated dynamically, a sample calculation could look like this:
In the context of Germany, 5 GW of dynamic demand is about 7% of total peak demand (70 GW). More would probably be better, the point here is that IF grid operators see value in having these resources at their disposal, the national infrastructure should be prepared and designed with that in mind. For example, standards like Open Automated Demand Response (OpenADR) might be considered to coordinate the management of supply and dynamic demand.
In light of coarse grained supply considerations, it may be beneficial to add additional dynamic load during times of low renewable supply. Likewise, during times of excessive renewable energy, distributed compute resources can be activated to do something useful with that energy.
Open dynamic renewable grids: The power of ♵
A systems approach creates synergetic benefits in optimizing the supply and demand of energy. By considering the power grids, heat applications and networks and the internationally distributed compute infrastructure as one system, a global optimum reduction in emissions can be realized.
Achieving these objectives will require a shared vision and coordination across ecosystems and value chains, well beyond traditional boundaries. It may not be the easiest or most obvious path forward — but it will take non-conventional systems approaches like this, to achieve results that are different from those that came about organically in the past.