Balancing India’s electricity grid in 2030: A detailed, granular analysis under uncertainty

UPDATE – The final and updated study has been published in the Journal, Transactions of the Indian National Academy of Engineering, August 2022. Click here for the summary and downloading.

Balancing the electricity grid is a complex issue with millions of nodes, where supply and demand have to match continuously in real time.  Varying demand was traditionally met by over-engineering “firm” supply, but the rise of variable renewable energy (RE, like wind and solar) makes supply itself both variable and uncertain. India has some of the most aggressive RE targets in the world – to reach 450 GW by 2030 (from about 100 GW installed today).  India’s overwhelmingly dominant form of supply, coal, is under environmental—and economic—pressure, and its growth is slowing down to the extent that capacity might plateau in some years.

Because electricity has two aspects—energy (kilowatt-hours) and capacity (kilowatts)—and given the importance of time of day of supply and seasonality, CSEP (then Brookings India) built the first public portal for real-time 5-minute analysis of all-India demand and supply fuel-wise (carbontracker.in).[1]  Using data from this tool, we built a national-level 30-minute resolution despatch model (grid balancing) for India spanning 2021 to 2030, under a range of assumptions of fuel costs, capital costs, growth of demand, etc.  The study highlights findings and details as part of the study “Granular Time of Day Analysis of Balancing the Indian Electricity Grid in 2030”.

We tackle a range of questions such as:

1. How much RE is feasible, and are there risks of “too much” RE, which would lead to curtailment?
2. Does India need more coal-based power? Does it need to build more coal power plants?
3. What supply (fuel) options make the most sense for new capacity going forward? This isn’t just a function of their economics but also issues like system security and managing volatility or uncertainty.
4. Are batteries the answer in the short term or longer-term? How should they be thought of for planning?
5. What are the factors that matter for grid planning and policy?

We find that not only is high RE possible, it is also usually the most cost-effective solution, even if there is measurable surplus RE leading to curtailment (throwing it away). However, even projected very high RE isn’t sufficient to meet all the incremental demand for power, even with extensive storage (which is very expensive).

Not only is high RE possible, it is also usually the most cost-effective solution, even if there is measurable surplus RE leading to curtailment (throwing it away).

The analysis suggests a multi-pronged strategy for India:

1. Focus on matching demand and supply.
   This helps reduce the growth of the peak, which is much more important for planning than average demand. Peak should focus on “net demand,” which is demand minus RE (treated like negative demand).  Storage is a tool in the arsenal, but more cost-effective solutions remain combinations of control/planning (like solar pumps) and pricing signals that reflect grid surplus or scarcity.
2. Utilize existing capacity to the extent feasible.
   India has a temporary surplus of electricity, to the extent that it has concerns over stranded and non-performing assets. The flip side of this generation capacity is the ability to meet incremental demand simply by paying the marginal or fuel costs. As demand rises, within this decade the surplus will exhaust, perhaps as soon as within the next four to five years under a range of assumptions on capacity growth as well as load profiles. At the point where India cannot meet demand through existing capacity or even planned new capacity (RE as well as hydro or nuclear), it will have to further add new capacity.  This is not only expensive in general, but it is especially expensive if used with a low capacity utilization factor (CUF, also called Plant Load Factor, or PLF) to meet the peak.
3. Increase system nimbleness.
   The above points emphasize the transformation of the existing grid from a heavily centralized costs-plus system where adding capacity was always the best solution, and costs were simply passed through to consumers. In addition to time-of-day pricing, the grid needs vastly greater coordination across states and regions. The larger transformation towards a Smart Grid will increase not just granular visibility but also provide instruments and mechanisms for peak load management such as through Demand Response.
4. Change pricing signals for choosing new generation capacity.
   Most importantly, “cheap electricity” shouldn’t be measured by levelized cost of energy (LCOE) which ignores both time of day aspects as well as system-level costs. It is a reality that as the share of RE rises in a system, its marginal value declines, and its marginal cost of integration rises. India’s Central Electricity Authority (CEA) estimated integration costs in 2017 to be as high as 1.5 Rs./kWh based on then costs, while varying values of RE are highlighted in a framework that shows a ladder of competition between coal and RE, where one shouldn’t simply compare LCOEs but also factor in fixed versus marginal costs (especially given the surplus of existing coal plants), location, and time of day.

LCOE is an especially poor tool that applies only on an energy basis for individual plant costs, and is inferior to metrics that focuses on system-wide (portfolio level) costs for instantaneous matching of capacity with demand. As an example, there are a number of publications that state that not only are battery costs falling, the total costs are also low because we can blend a battery with RE. Such publications state, for example, if RE is Rs. 2/kWh, and a battery is Rs. 8/kWh (LCOE), given we only need 25% of the RE to be stored in a battery, that means we have 2 + 25%*8 = 2+2 = 4 Rs./kWh for the total costs. This is incorrect given we might need just 25% battery on an energy basis, but we need double or triple the average battery size to meet the peak demand for part of the time. This larger-sized battery isn’t wasted during off-peak hours, but its value is much lower, simply equal to the marginal cost of alternatively avoided generation. In the short run we might manage with energy-basis blending of storage with RE, but in the very short run we don’t even need storage at all. The real challenge occurs as and when we reach very high levels of RE and simultaneously need to meet incremental peak demand.

There are a number of studies looking at prospective grid balancing in India, including by NREL, TERI, Prayas, and CSTEP.  While some are much richer spatially, going down to the state or even power plant level, a key difference is that we do not use simulated data for supply and demand curves by time of day but instead use actual 2019 data as a starting base. We also cover a very wide range of uncertainty not just across capital costs, foreign exchange escalation rates, discount rates, etc. but also for how RE grows over time and in what ratio of solar:wind – we do not assume government targets as a base. More importantly, we segregate battery costs between energy and capacity, and do not simply assume a “4-hour battery,” i.e., $200/kWh = 0.25 kW output for 4 hours at $50/kW.

While this study is not a full-blown optimization model, we can find lowest cost solutions with optimal despatch by having a wide portfolio of generation choices as well as potential fuel mixes for future growth. Given we only focus on national data, we simplify the system to assume perfect transmission and coordination, which gives results as a bounding exercise.  Any grid congestion, local bottlenecks, or plant-specific outages would only make things harder than the optimal based on a national analysis. Our focus is less on any specific cost number for comparison across fuel choices but rather on trends and key factors that are robust regardless of assumptions.

A few key findings and insights include:

1. There is significant uncertainty in supply and demand as well as prices, and assumptions matter significantly. These include uncertainty in load profiles, capital costs, finance costs, share of solar:wind, etc.  We start with the expectation that the PLF of solar and wind in the future will be higher than in the past, but year-on-year variations will be very critical, more so for wind.  A forthcoming paper by Schwarz & Tongia shows wind output in regions of India can change by over 30% different across years. Even hydropower depends measurably on the monsoon and climate patterns, which are shifting due to climate change.
2. Even with a high PLF, new RE is unlikely to be sufficient to meet upcoming demand through 2030. Not only is the target of 450 GW by 2030 a stretch, even if this is met, it wouldn’t meet the incremental demand rise of, say, 5.25% after we account for system curtailment, storage losses, incremental transmission losses, etc.  Some of the gap would have to be met from existing but under-utilized capacity, perhaps even fossil fuel assets. This is a calculation for 2030, which doesn’t consider the trajectory of growth for RE. Would it be equally spread out – 350 GW of RE growth means 35 GW/year over ten years? Alternatively, would it take more time to scale up, following a more typical increasing growth (CAGR – compound annual growth rate)?
3. The current system suffices due to surplus capacity, but this surplus will end in a few years. By when it would exhaust depends on not just load growth or RE growth but also on what happens to the existing coal fleet. Will all coal plants get pollution control equipment, or will some retire before their end of capable life? This analysis doesn’t segregate completion of existing under-construction coal plants as a cheaper option given the lack of data on remaining costs. If we assume all new coal capacity to be brownfield or greenfield, coal is not cost-effective at least through 2030 due to low expected PLFs, by when alternatives will likely be even more cost-effective (or caps on carbon emissions may grow). If we assume much of the under-construction coal power plant capacity gets built, some of it may only displace or replace older plants.  Even with a net increase of 25 GW over 2019’s capacity before a plateau, that’s only about a 12% increase, which would be a small fraction of a percent of present global carbon dioxide emissions.
4. If the objective is “no new coal”, there is a need to make existing plants much more flexible. The ability to lower coal plant output is a key factor for system costs due to curtailment of RE driven not by lack of electricity demand but by limitations on lowering the output from coal power plants. Such coal plants are often needed for peak evening duty cycles when solar output is zero, and we assume the system doesn’t plan for daily start-stops of coal plants (which is very expensive).
5. Adding any new capacity regardless of type (coal, gas combined cycle, gas open cycle, combustion engines, RE+batteries, etc.) to meet upcoming unmet demand will be expensive due to its low PLF. The good news is that much of the planned RE growth doesn’t need a battery – a little curtailment is still cost-effective.
6. “Surplus RE” being curtailed doesn’t match well with batteries needed for meeting peak demand due to seasonality issues. If we do choose storage with RE for future supply, it will need relatively predictable and incremental RE, probably new solar. But if we weren’t able to growth capacity to meet the 450 GW RE target, separate new RE for storage will also be difficult unless we are able to pay a premium for such power.
7. Storage isn’t the panacea many people believe.  Beyond economic issues of LCOE we highlighted previously, right-sizing batteries is a complex issue, e.g., for use as a seasonal speaker.
8. It is cheaper to blend storage and RE with an option like limited-use biodiesel. This allows a smaller battery focused on energy requirements instead of sizing the battery for occasional peaks. While biofuels have land use implications, the land requirements for a small volume of power to come from biofuels are much smaller than proposed transportation biofuel blending requirements.