Sunday, July 7, 2019

Baseload's threatened ability to contribute to lower emissions

“Baseload” is a contentious term in energy discourse. In analysing electricity data in Ontario it occurred to me there’s a simple way to demonstrate the potential value of supply that delivers a consistent output all of the time - one that ignores all generation technology, using only hourly demand data. In this post I’ll demonstrate this methodology before discussing implications for supply mixes.

“Base” and “Load” are two fairly well-defined terms - neither of which are strictly adhered to in my methodology.

“Load” I treat as whatever data I have. I’ve collected available hourly, or half-hourly, data for 3 Canadian provinces, 5 Australian states, and 5 US systems. The data is unlikely to be equal: one example is the figure used for Alberta is “Alberta Internal Load” which includes “behind-the-fence” self-generation unlike the Ontario system operator’s “Ontario Demand”, which only reflects supply from their grid. I am not aware of what supply is included, or excluded, in data I’ve collected from the U.S. Energy Information Administration (EIA) or the Australian National Energy Market. Until the case study section of this analysis the differences can be ignored.

“Base” could be called minimum, but I think it’s helpful to eliminate outliers. The most extreme example is the great blackout in August 2003 that impacted most of Ontario, but more generally there will be some ideal nights on holiday weekends where demand is below its normal lows. In this analysis I define “Baseload” in relation to the statistical mean, which is better known as the Average (A) by those of us who determine it using the available spreadsheet, or other database, function.

While I am well-acquainted with data, I’ve only met statistics. Wikipedia explains the standard deviation, represented by the symbol “σ” (sigma), “is a measure that is used to quantify the amount of variation or dispersion of a set of data values,” and provides a very helpful graphic displaying 1st, 2nd and 3rd standard deviations on a plot of a normal distribution.

(By M. W. Toews - Own work, based (in concept) on figure by Jeremy Kemp, on 2005-02-09, CC BY 2.5,

People who are well-acquainted with statistics might be able to anticipate the results of much of my analysis, and probably could use it to determine how the distribution of electricity demands differs from a standard distribution. For instance, the average percentage of hours where I find demand is below A - σ (the statistical mean less one standard deviation), in 13 electricity system hourly data sets, is 15.45%: in the diagram above of a standard distribution it’s 15.8%. That result should not surprise a statistician, but perhaps some other metrics I’ve collected will be - and if not I will attempt to present the analysis for those that those unfamiliar with statistics.

Technique for determining baseload potential in electricity systems

I looked at 4 values for “baseload”, which could be put into equations, using the statistical mean (A) and/or multiples of the standard deviation(σ): A, A-σ, A-1.5σ, and A-2σ. One definition of outlier values is those outside of 2 standard deviations (σ) from the statistical mean (A), so that was my starting point in searching for the base value after culling outliers. Queries counted hours where the baseload would exceed demand and the total excess over a year. I was surprised there were very few hours of demand below the A-2σ baseload level, which is consistently very close to the actual minimum hourly demand for a year (outlier values on the high side are much greater than the A+2σ limit)

Experimenting with “base” levels that would meat demand without significant excess/curtailment I queried the data using baseload set to the average, and then less both one standard deviation and one-and-a-half standard deviations. There will be multiple ways to interpret the data: I will show analysis including other jurisdictions with the baseload defined as one standard deviation below the statistical mean (A-σ) as at that level over 83% of “Ontario Demand” can be met but above that level much of the potential output from incremental basload would be excess supply.

The same analysis was performed with data from 12 other systems, including 2 Canadian provinces, the 5 Australian states with data from its National Energy Market operator, and 5 U.S. systems. In all systems baseload supplied as A - σ (one standard deviation below the mean) could service at least 76% of demand, and averaged 83% of supply met on less than 1% excess supply created (max was 1.3% in BC).

There are some differences in the systems that impact the measurement, which will be indicated in the case study of Ontario later in this analysis.

The results infer that the greater the standard deviation is in relation to the statistical mean, the lower the share of total demand met at a baseload of A-σ. The relationship is clearer with baseload set at two standard deviations below the mean (A-2σ) - at that level the jurisdiction with the lowest relative standard deviation (AB) still could serve 85% of demand from the steady baseload, but the highest standard deviation jurisdiction (South Australia) could only service 54% of demand from a baseload supply two standard deviations below the mean.

Ontario: a Case study of the impact of distributed energy sources

While conclusions can be drawn from examining available hourly data for multiple systems there is much more to be learned in revealing the change in a single system over time. 

This analysis opened with a graphic illustrating the results of querying hourly “Ontario Demand” data from the system operator in Ontario (IESO). The “demand” term doesn’t reference the actual provincial consumption of electricity - only the demand from wholesale consumers and distribution companies for supply from the IESO-controlled grid (ICG). For a number of years distributed energy (Dx) grew rapidly in Ontario (the IESO should know as they contracted it). Today solar is about 2/3rds of 3300 MW of contracted distribution-connected capacity operating commercially in Ontario, with almost all of the solar entering service since 2008.

I’ve estimated hourly embedded generation back to the start of 2008, allowing me to add the hourly total to “Ontario Demand” for a more complete picture of electrical load in the province. The results become fairly constant after the recession of 2009 - indicating the widely perceived decline in provincial demand is primarily due to a failure of reporting on embedded generation.

A comparison of the statistics for the system with, and without, distributed supply, indicates the impact of the added supply in distribution networks. Omitting distributed supply from analysis lowers the statistical mean by about 800 MW in recent years (7 terawatt-hours annually), but because the standard deviation also grows the baseload, defined as A-σ, decreases by about 600MW, from about 14,000 to about 13,400.

In both the consumption analysis (with distributed supply) and the “Ontario Demand” analysis (without Dx) a baseload supply of the statistical mean less one standard deviation meets around 83% of the annual total demand/consumption, but excluding distributed supply from the calculation means the statistical mean is lower, and the variance is also found to be lower in the analysis without the, mostly solar, distributed supply. This is a positive for solar, indicating its output profile is more aligned with the provincial consumption profile than not. Note at some point this would change if solar continued to grow to the point where afternoon residual demand (demand less solar) was greater overnight than in the afternoon.

The most notable change in the consumption and “Ontario Demand” scenarios is the annual maximum demand, which drops about 1,500 MW, in recent years, when omitting distributed supply from the analysis. This indicates the roughly 2,130 MW of embedded solar is reducing system peak consumption very impressively. Remembering that one definition of outliers is data values more than two standard deviations from the mean (A +/- 2σ), the much steeper drop in annual maximum consumption than the reduction in A+2σ, indicates solar has been effective in meeting outlier consumption.

Because of the ability of embedded solar to address outlier demand highs, the need for non-baseload (A-σ) supply is reduced despite the reduction in baseload levels.

The impact of Industrial wind generation

In most systems sporadic generation from wind has been contracted on a preferential basis, and Ontario is no exception. The preference can be in mandates to reach a certain share of supply, price (either high or guaranteed), preference in granting grid access, or any combination of the above. It is common to therefore hear of residual electricity demand, referencing the load left to be met after the output from solar and wind generators.

The previous section on distributed generation indicated some benefits of solar: while a baseload supply would be smaller if excess generation is to be avoided, the peak hours are addressed and therefore not as much total capacity is required on the non-solar system.This does not appear to be the case with wind.

Removing contracted wind from the main data set, including estimates of hourly distributed generation, requires removing the estimated distributed wind generation and the estimated curtailed wind. I have done this, and the change is dramatically different than was seen when distributed solar was removed. With wind it is the minimum annual hourly load that drops the most, and the standard deviation actually increases. The annual total generation from the wind turbines drops the statistical mean, and once the higher standard deviation is removed from that the opportunity for efficient baseload drops significantly. The A-σ equation for baseload still produces a figure that can meet about 80% of the total (which is now only residual demand), and at that level it still only produces 1.2-1.4% excess supply, but 5,000 MW of wind effectively reduces the prudent level of baseload supply by 2,000 MW.

The prudent level of baseload decreases with the addition of wind, in Ontario, more rapidly than the statistical mean (A) of the residual load. That means more non-wind system capacity is required when wind is added to a system than would be necessary in a system without wind and with an optimal baseload amount.

Concluding thoughts

The good news is the analysis began with a search for a definition of baseload that would not result in large excess supply and A-σ was quickly found to be useful for doing that in a wide variety of situations. Less good was the revelation of the impact of other supply decisions on the amount of baseload supply to meet the residual demands, and the impact on overall system capacity of wind - at least in Ontario. This isn’t a point modern pro-nuclear people are supposed to bring up, but it’s long been an argument of anti-nuclear promoters of wind - enunciated well by influential German politician Siegmar Gabriel years ago:
“If someone declares publicly that nuclear power would be needed in the baseload because of fluctuating energy from wind or sun in the grid, he has either not understood how an electricity grid or a nuclear power plant operates, or he consciously lies to the public. Nuclear energy and renewable energies cannot be combined.”
I deliberately avoided connecting generation technologies to the “baseload” in the initial sections of this analysis, but a connection is unavoidable.

Currently Ontario has 12,000 MW of supply coming from nuclear generators, and its hydro generators are probably being curtailed when the fleet is producing less than 3,000 MW - for 15,000 MW of what I’d categorize as baseload supply. This analysis suggested 12,000 would be more appropriate - the 3,000 MW coincidentally being the size of the Pickering nuclear generating station scheduled to close by 2025. The analysis also suggested more supply would be needed due to wind on the grid than the baseload generators it tends to displace -which explains the launching of an incremental capacity auction as Pickering units are removed from service.

In reality generators don’t often run 24*7*365 hours a year, and most would have outages scheduled at lower demand periods of the year. Ontario’s hydro-electric facilities include both peakers that ramp up during certain hours of the day and, far more numerously, ones that provide a fairly constant output based on water conditions. I’ve graphed an annual average of the system operator’s weekly baseload forecast (minus the wind component - as I disagree with their categorization). For the current year their average baseload component of supply is over 13,000 MW.

Ontario also had a number of gas-fired generators that were contracted decades ago in a manner that paid them to deliver a constant output. Fortunately these must-take gas contracts have been expiring and are now nearly eliminated. The end of those contracts, and removal from service of Darlington’s reactor 2 for refurbishment mean that baseload, since 2017, has returned to the levels it was at prior to the return to service of 1500 MW of nuclear (Bruce units 1 and 2) late in 2012.

Ontario has been reducing baseload supply and will continue to do so, and this analysis indicates industrial wind can rightfully claim responsibility for the cuts.

I’ll conclude by noting Ontario’s emissions intensity of electricity generation in recent years has been no more than 40 kg of CO2 equivalent per MWh (according to Table A13-7 of Canada’s National Inventory Report 1990-2017). Ontario has shown the very low system emissions are possible with a high share of supply from baseload generators - as has France. Nobody has yet built a system with a high share of wind with very low emissions.

While no systems exist, studies claim they are possible. While Professors and school children promote the possibility of grids entirely powered by wind, water and sun, more serious people look to 80% renewable as achievable with today’s technology. Here’s some quotes from a National [US] Renewable Energy Laboratory [NREL] study’s Executive Summary:
  • Electricity supply and demand can be balanced in every hour of the year in each region with nearly 80% of electricity from renewable resources, including nearly 50% from variable renewable generation, according to simulations of 2050 power system operations…
  • ...a more flexible system is needed to accommodate increasing levels of renewable generation. System flexibility can be increased using a broad portfolio of supply- and demand-side options and will likely require technology advances, new operating procedures, evolved business models, and new market rules….
  • ... additional transmission infrastructure is required to deliver generation from cost-effective remote renewable resources to load centers, enable reserve sharing over greater distances, and smooth output profiles of variable resources by enabling greater geospatial diversity.
80% renewables in the NREL study includes 30% firm renewables (not wind or solar), and require new wires, markets, and technologies.

I have attempted to show that 80% trivial-emissions supply is easily modeled. Models seem to gain fans for certain technologies, while there seems no excitement for technologies currently existing and operating within functional systems.

No comments:

Post a Comment