HOME OZ-ENERGY-ANALYSIS.ORG   -   open science for the new millennium

THE STORIES   |   DATA   |   ANALYSIS   |   MODELS   |   LITERATURE   |   DISCUSSIONS

Status: round one; last active here July 2010

Wind Penetration by Installed Capacity

[fc: 30th May 2011] This was an initial exploration, and it remains a valuable introduction. The context here is limited to wind farms in SA, and the extreme demand remainder seen in Figure AWP2 reflects this. Two follow-up aspects await treatment: first, it was suggested that including further data for other SA wind farms may improve the spatial smoothing, although inspection of the currently processed data suggests this would make only a minor difference; second, down-sampling of the wind farm generation data acts to introduce artificial smoothing, and it remains that an analysis of what is lost in downsampling from five minute to hour-level data is needed more generally. This work was followed, in particular, by the first modelling work: Bucket Storage Model and Using Gas.

ABSTRACT

What are the basic limits on Wind Power penetration? We start with a very simple 'cutout' of SA (no interstate connectors, no storage), and consider increasing levels of wind power. We simply assume (for now) that demand over and above that supplied by wind (the 'remainder') is filled in 'somehow'. With only a little wind power, it is all used; at higher levels the available wind power is sometimes greater than the demand, and the excess is dumped / spilt (in our analysis). At increasing levels of installed wind power, we count up the the overall percentage of the demand that is supplied by the wind power (the penetration), and also the total amount that must be spilt in the absense of storage.

DATA and METHODS

I have taken the SA demand data for 2009, downsampling the 30min data to the hour-by-hour level, and Wind Farm output data for 5 SA wind farms (CATHROCK, LKBONNY1, MTMILLAR, STARHLWF, and WPWF), also downsampled to the hour level. These data are given in this 54 KB gziped text file.

The wind farm data constitutes 340 MW of installed (nameplate) capacity, and so this data was divided by 3.4 to obtain a nominal supply curve for 100 MW of installed wind capacity. This curve was then multiplied up as needed to produce the supply curve for increasing levels of wind penetration.

At a given level of wind penetration, the 8760 individual hours were processed, and the percentage of demand provided by the wind power was calculated. If this value was greater than 100%, it was capped to 100% and the 'spill' was recorded. The total percentage of the overall demand supplied from wind (the penetration) was then calculated.

ANALYSIS

The SA demand for 2009 averaged 1.5 GW (1,538 MW).

The wind farm data constitutes 340 MW of installed (nameplate) capacity, and for 2009 produced a mean of 104 MW (indicating a Capacity Factor of around 31%). We multiply this up to simulate higher levels of installed wind power.

Part 1: Overall Penetration

As a concrete example, consider that the SA mean demand = 1.5 GW = 30% of 5GW. What would happen if there were 5 GW of installed wind? Sometimes there would be too much power (when the wind blows), and sometimes there would not be enough (when the wind don't blow). The overall penetration of wind power at increasing levels of installed wind power, calculated as per methods above, is shown in the following plot:

Figure AWP1: the percentage of demand met for increasing levels of installed wind power (based on SA, 2009, with no storage or interstate connection).

We see that IF there were 5 GW of installed wind power, this would supply around 2/3 of demand, and about 1/3 of the power generated would be at the wrong time, and so would need to be somehow stored, or otherwise lost. Similarly, 1/3 of the power requirement would not be satisfied directly by Wind Power.

Some other comments:
- As demand rarely goes below ~1 GW, the initial linearity is unsurprising.
- Up to about 2 GW installed wind, there is minimal spill (this gives 40% penetration - in this simplified view).

Note also that for now we are simply assuming that the remainder of the required power is filled by other generators (but, see below). It will also be interesting to look at how these curves change when various storage and/or demand management capabilities are modelled.

Part 2: A look at the Demand Remainder

In the above figure 50% penetration occurs at just under 3 GW installed wind (which is double the average demand). Using this as a useful reference point we now examine the all important other half (that is, the power that must be supplied in some other way). The figure below shows the January 2009 SA demand curve (red), and the demand curve that remains after the wind power is taken away as above (shaded light-blue). Note that this new demand remainder is even more variable than the full demand.

Figure showing Remainder after Wind
Figure AWP2: SA demand for January 2009 (including a heat wave at the end), and the demand remainder in the case of wind power at 50% penetration (according to the simplified analysis above). Note that the demand axis has been scaled (1534 MW = the average = "1").

OPENING COMMENTS

The reason for looking at this demand remainder is that supplying this demand in an efficient way is key to a high-renewables grid. To this point we have simply been combining two data types with some basic number crunching, and hence this work is presented here (i.e. as an analysis). What comes next?

First, we are going to use gas turbines as the starting point for supplying this demand remainder. It is thus necessary to develop a calculator of sorts that takes as input a demand curve such as here, and provides a sensible schedule for the use of gas turbines to supply this demand. This work is being treated as method development here: Meet a Demand Curve with Gas.

Second, this analysis does not consider storage or demand management. Using these we can both capture some of the wind power that is, in this analysis, simply spilt. The goal is to substantially smooth the demand remainder curve in ways that will make it easier (cheaper) to supply with gas or other power. This modelling work starts here: Model #1 - Bucket Storage and Gas.


[Jump to bottom]

DISCUSSION: (on wind penetration analysis)

2

OzEA_AWP0002

Rapid Rabbit
Subject: Deficit vs penetration
Date: 2010-05-31 (at 21:07:32)

For comparison, could you show the inverse plot also, which plotted the deficit vs penetration (i.e. the accumulated gap between demand and supply). Perhaps it could even go on the same graph.

4

OzEA_AWP0004

Peter Lang
Subject: Chart 'Overall % of demand' versus '% penetration' of wind power
Date: 2010-06-02 (at 12:03:07)

This is an interesting chart. I have a few comments:

Having gone to such effort to produce this, I think it would be valuable to produce the equivalent chart using the 5 minute data. It may not be much different, but we do have the data and if you don't do it on the 5-munute data I am left wondering how much has this been smoothed by averaging?

I would like to see the chart formatted so it can be more easily used for the general application.

As an example, I have a particular question relating to a comment by Neil Howes on the BNC web site. This chart could help us answer the question if the chart was presented differently.

The questions I would like to be able to answer are:

How much wind energy (also shown in terms of average power) is spilt and what is the capacity factor (sent out) when:

1. penetration is a%
2. potential capacity factor (gross power that could be generated if no wind power is spilt) is b%
3. Transmission capacity to wind farms is c% of installed wind capacity
4. Back up generators cannot fall below d% of demand (e.g. 40%)
5. Ramp rate cannot exceed e% per minute

Here is an example:

What is the capacity factor of wind and how much wind energy is spilt when:

1. Penetration of wind power is 3,300MW in total system installed capacity of 14,000MW (10,700MW excluding wind power)
2. Potential gross capacity factor of wind turbines is 30%
3. Transmission capacity is 80% of wind farms installed capacity
4. Backup generators cannot fall below 40% of demand
5. Ramp rate cannot exceed 2% per minute

I would suggest changing the x-axis to "% penetration". Then have several charts (or a pivot chart facility) to display, the different series; for example, capacity factor at say 20% to 40% at 2% increments.

Rapid Rabbit's suggestion to show deficit versus penetration would be good.

6

OzEA_AWP0006

Neil Howes
Subject: variance of wind farm output
Date: 2010-06-04 (at 12:27:28)

Before doing too many calculations with data from 5 sites it would be important to look at the data available form a larger number of sites as see how this changes variability. IF we have 3-5GW of wind its going to come from 20-50 differnt farms located over a wider geographic area than the 5 locations being used. If one of the 5 locations is operating at 0 or 100% capacity this is going to make a major contribution to average, with 20 sites only a minor contribution. One suggestion would be to examine data from all 9 locations in SA and the sites in VIC along the SW coast, to simulate another 300km of SA coastline and compare with the 5 site data.

fc: yes, I agree this is worth doing, and we will do it. But I'll proceed here with the 5 for the moment. Will add new material here later today.

7

OzEA_AWP0007

Eponymous
Subject: Storage?
Date: 2010-06-04 (at 17:07:15)

Isn't there any pumped storage in SA? How do they currently deal with demand fluctuations? My understanding is that the interconnect isn't very big, and the pumped storage in the East coast is a long way away. Seems an inefficient way to run a grid, but little surprises me these days.

8

OzEA_AWP0008

Barry Brook
Subject: Storage
Date: 2010-06-04 (at 17:45:47)

Eponymous, there is no PHS in SA, although there is some pumping of water from the Murray to the Adelaide Hills dams, which is what you may be thinking of. We currently deal with wind-based fluctuations (19% average demand) using Open Cycle Gas Turbines, hence this is the first focus of the modelling. The existing interconnectors can supply a few 100 MW of power, some details here:

http://www.oz-energy-analysis.org/data/generators.php

and

http://www.oz-energy-analysis.org/data/transmission_lines.php

10

OzEA_AWP0010

John Newlands
Subject: SA wind power deficit
Date: 2010-06-04 (at 18:41:05)

SA total demand can approach 3 GW an example if I recall was mid March 2007 when temperatures approached 46C. The analysis matches total State demand against partial wind output, with actual nameplate being nearer to 900 MW I believe. It would be interesting to plot the practical following gas output with allowance for ramp times and spinning reserve, not just total demand minus potential wind contribution. While the Torrens Island station is nominally baseload I think it consists of 8 separate gas fired boilers (no jet engines) if I understand correctly. Modules can be simply switched off though I guess it would take several hours to bring them back to steam. Perhaps for practical purposes Torrens Island output cannot usually fall below a minimum (say .5 GW) regardless of potential wind output.

11

OzEA_AWP0011

Peter Lang
Subject: Modelling wind variability, firming and cost at short time increments
Date: 2010-06-05 (at 16:18:19)

This paper may be of interest:
https://wpweb2.tepper.cmu.edu/ceic/PDFS/CEIC_10_01_CWV.pdf

The abstract is here:
http://wpweb2.tepper.cmu.edu/ceic/papers/ceic-10-01.asp?&em=mbelively@aol.com

TITLE: Compensating for Wind Variability Using Co-Located Natural Gas Generation and Energy Storage
AUTHOR: Eric Hittinger, J.F. Whitacre, Jay Apt
ABSTRACT: Wind generation presents variability on every time scale, which must be accommodated by the electric grid. Limited quantities of wind power can be successfully integrated by the current generation and demand-side response mix but, as deployment of variable resources increases, the resulting variability becomes increasingly difficult and costly to mitigate. We model a co-located power generation/energy storage block which contains wind generation, a gas turbine, and fast-ramping energy storage. Conceptually, the system is designed with the goal of producing near-constant "baseload" power at a reasonable cost while still delivering a significant and environmentally meaningful fraction of that power from wind. The model is executed in 10 second time increments in order to correctly reflect the operational limitations of the natural gas turbine. A scenario analysis identifies system configurations that can generate power with 30% of energy from wind, a variability of less than 0.5% of the desired power level, and an average cost around $70/MWh. The systems described have the most utility for isolated grids, such as Hawaii or Ireland, but the study has implications for all electrical systems seeking to integrate wind energy and informs potential incentive policies.

12

OzEA_AWP0012

Rapid Rabbit
Subject: Gap filling
Date: 2010-06-05 (at 18:39:25)

Okay, the obvious implication of the above that for the first abstraction, there must be as much gas turbines installed as there is demand, since sometimes wind is apparently contributing nothing.

Looking at the January record, demand management could perhaps shave the tops of the mega-peaks (e.g. Jan 28). That still leaves about 1.5 to 2 GW of 'at call' OCGT to be installed. Is this a reasonable assertion?

Torrens Is, which is 1.2 GW, is an old oil-fired power stations with steam turbines, so would be unsuitable for this purpose. So new gas turbines will be required.

How much OCGT is currently build in SA?

fc [6pm, Sun 6/6] -- yes RR, that sounds ballpark. Not clear on current Open / Combined split in SA (we have current total as 2.8 GW + 1.6 GW proposed). Note that Jan 2009 was an extreme month.

13

OzEA_AWP0013

Peter Lang
Subject: Gap filling
Date: 2010-06-06 (at 17:22:57)

Rapid Rabbit,

I agree with your interpretation: "the obvious implication of the above that for the first abstraction, there must be as much gas turbines installed as there is demand, since sometimes wind is apparently contributing nothing."

Furthermore, notice what is happening with the ramp rates. Look at 27 and 28 January for example. As demand is dropping, the back-up power is dropping even faster than the demand. The back-up power drops to near zero. So the ramp rates required of the back-up generators must be even faster than would be required if there was no wind power. This means more inefficiency, more generators maintained in spining reserve or in cool down mode for longer. Result: higher emissions. [Francis, I recognise your model is not dealing with back-up and emissions, but the principl applies to whatever back up system you have. There is no possibility of having sufficient pumped hydro for back-up, batteries are prohibitavely expensive at the moment and no sign of a breakthrough and CAES is costly too.]

This may be of interest to gice some perspective on the issues involved in providing our electirity supply and the extra problems and costs being forced on us by the Renewable Energy Targets.

http://www.oakleygreenwood.com.au/images/Gas_Markets_DLAP_Snow_6October09.pdf

I am not sure if this is acceptable to post here, but I feel it is relevant as background to help guide the exercise being embarked on here.

14

OzEA_AWP0014

Francis
Subject: acceptable to post with introduction
Date: 2010-06-06 (at 17:36:38)

Peter, a couple of points:
1. the issue of backup is on the way -- this analysis here is just about opening the can;
2. As you know I want to keep away from emissions until we are at the costing stage. What I want here is for everyone to follow the stated etiquette for posting links:

Citing literature and other sources: appropriate and interesting citations and links within comments are welcomed, but please DO NOT cite material that you have not yourself read, digested and understood. As a general rule, please introduce any and every link or reference with a short description of the material, your judgement on its quality, and the specific reason you are including it (i.e. how it is relevant to the discussion).

It also seems to me that you are tilting towards the politics of renewables rather than focusing on the stated aim of working out what would be involved for (South) Australia to have a high-level of renewables integrated into the electricity grid.

15

OzEA_AWP0015

Stephen Gloor (Ender)
Subject: Gap Filling and Ramp Rates
Date: 2010-06-07 (at 20:41:45)

The main problem with this model is that basically you are building super large wind farms on 5 sites which is not really valid. As noted in Jacobson&Archer2008 "Supplying Baseload Power and Reducing Transmission Requirements by Interconnecting Wind Farms" the variability of wind is mitigated by dispersal and numbers of wind farms.

"The analysis indicated that the reliability of interconnected
wind systems increased with the number of farms."

Additionally the effect of a large number of wind farms is to smooth the aggregated output so reducing ramp rates of the backup power.

Electric power from offshore wind via synoptic-scale interconnection
Willett Kemptona,1, Felipe M. Pimentaa, Dana E. Verona, and Brian A. Colleb

"Each individual wind power generation site exhibits the expected power ups and downs. But when we simulate a power line connecting them, called here the Atlantic Transmission Grid, the output from the entire set of generators rarely reaches either low or full power, and power changes slowly."

In this analysis a small number of wind farms is supersised rather than large numbers of wind farms being spread out over SA into different wind regimes. By a quite massive accident we have managed to place all our major wind farms at roughly the same longtitude making them vunerable to single weather events like the large blocking highs that really kill the wind.

Perhaps we need to develop a wind model of SA much like the EWITS study did for the Eastern States of the USA. We also need to profile the wind up to the required 80m hub height like in Jacobson&Archer. In this way the upscaling of wind capacity be be accompanied by a greater dispersal of wind farms which should result in a smoother output. Certainly we can confirm the results of the USA studies and see if their results are unique or can be applied to other countries.

fc: please consider this analysis before employing language such as "really not valid". I'd prefer such language were not used at all. In any case, the point about spatial smoothing will be considered further shortly.

fc, 10pm: Please also see the new comment at the top of the wind data page.

16

OzEA_AWP0016

Stephen Gloor (Ender)
Subject: Apologies
Date: 2010-06-08 (at 13:38:55)

fc - "before employing language such as "really not valid"."

Apologies - still feeling my way here.

fc: no worries Stephen - and thanks for your contributions in general

17

OzEA_AWP0017

Stephen Gloor (Ender)
Subject: Problem with this approach
Date: 2010-06-08 (at 13:57:12)

"fc: please consider this analysis before employing language such as "really not valid". I'd prefer such language were not used at all. In any case, the point about spatial smoothing will be considered further shortly."

The only problem with not constructing an overall wind model for SA is that you cannot place virtual wind farms in different spots and increase the diversity of wind farm placement. This could also be used to examine the least cost tranmission network as Jacobson found that the connected star network was more cost efficient.

I agree with Neil Howes that the selection of only a small number of sites is used, even if you add some from Tasmania, is that one site affects the overall output far too much. We only have a limited amount of wind farm data to use so perhaps the effort to extend this data will be rewarded with a better outcome.

I take on board the comment at the top about the data that is being added. Do we know of a way to reliably convert the data into 80m height data. Is there available a set of vertical wind profiles that we can use to convert this data?

fc: I appreciate these concerns - rather than address them piecemeal, they will be addressed in the second story, which I hope to post in a couple of days.

18

OzEA_AWP0018

Francis
Subject: working depth into breadth
Date: 2010-06-08 (at 14:23:55)

Stephen, What I have not made clear here (and this will come out in the second story when I post it) is that the 'game' at the moment is to do layers of 'breadth', with no pretence whatsoever that this is anything but exploratory analysis. As we flesh out these layers (and I'm still working to complete layer one), we stand back and look at the places that are most important to refine the 'depth' on the next pass. I expect we will need a good few passes before we get down into some of these more detailed issues.

The reason for this approach is as follows:

It is very very easy to get lost down the shaft of any particular issue, chipping away at the issues and imperfections, AND never get back out. Or, put another way, it is very easy to end up studying in detail a small number of trees, but never usefully navigate through the forest.

19

OzEA_AWP0019

Barry Brook
Subject: 80 m height data conversion
Date: 2010-06-08 (at 14:44:28)

"Do we know of a way to reliably convert the data into 80m height data"

Not really, unless we have monitoring from this height, to use as a basis of developing a statistical model to link to ground data.

Stephen, you mentioned elsewhere that Archer & Jacobson 2007 modelled 80 m height data -- I don't think that was the case. My recollection was that they used ground station data. Could you find the reference for this (pg/line) in their paper, if I'm incorrect?

There are rules-of-thumb regarding the increase in average wind speed with height, and from memory I'd seen figures of something like 30% greater speeds at 80 m compared to ground level. I'll have to do some digging, but I think I got that figure from Mackay's SEWTHA book.

20

OzEA_AWP0020

Stephen Gloor (Ender)
Subject: Jacobson and Archer 2007
Date: 2010-06-08 (at 15:38:48)

Barry Brook - \"Stephen, you mentioned elsewhere that Archer & Jacobson 2007 modelled 80 m height data\"

From the conclusion:

http://www.stanford.edu/group/efmh/winds/aj07_jamc.pdf

"Wind speeds at 80 m were calculated via the least squares method, which involved a combination of 10-m wind speed observations at the sites of interest and vertical wind profiles retrieved at nearby sounding stations. Observed data from the Kennedy Space Center in Florida were
used to validate the method.\"

21

OzEA_AWP0021

Stephen Gloor (Ender)
Subject: Atmospheric Soundings
Date: 2010-06-08 (at 16:02:12)

A while ago I was thinking of getting into gliding so I did a bit of reading. I must have sort of remembered this site as when I thought about the 80m hub height problem some more this niggled at me.

This is the site glider pilots use:

http://slash.dotat.org/atmos/info.html

It has a reasonable amount of vertical soundings to use.

22

OzEA_AWP0022

Barry Brook
Subject: Re: Jacobson and Archer 2007
Date: 2010-06-08 (at 17:16:06)

Right, thanks Stephen, so I was (mostly) right - they used 10 m wind speed data and extrapolated to 80 m using a simple statistical model. That was much what I proposed to do in my preamble to that comment. That glider data could be useful for further refinement of such a model (in due course).

24

OzEA_AWP0024

Arthur
Subject: Multi-Year Capacity Credit Variations & Lit Survey
Date: 2010-08-29 (at 02:12:12)

Checkout this June 1997 NREL report by Milligan on "Wind Plant Capacity Credit Variations: A Comparison of Results Using Multi-Year Actual and Simulated Data".

NREL/CP-440-23096

I believe it confirms that data for one year can underestimate the Loss of Load Probability and provides references to appropriate modelling techniques for more refined analysis.

Found it via this 2005 survey of Capacity Credit literature, which may also be of interest:

WIND POWER HAS A CAPACITY CREDIT: A CATALOGUE OF 50+ SUPPORTING STUDIES by G Giebel

25

OzEA_AWP0025

Francis
Subject: Re: #24 Multi-Year Capacity Credit Variations & Lit Survey
Date: 2010-08-31 (at 20:46:20)

Arthur (#24), thanks for the Milligan and Giebel references; I've now had a chance to have a quick look at both.

A point of clarification; I will avoid as much as possible statistical calculations of Capacity Credit or LOLP; the idea is very strongly to empirically count up what happens in real data.

And it is certainly the case that limited data (such as a single year of wind farm generation) will give less precise information on things such as LOLP than more extensive data. I can be careless in assuming that this is second-nature in the thinking of others. The point of working with data in year chunks is, and remains, one of containing the work we do in the 'learning' and 'exploration' phases -- at the point where OzEA seeks to make any definitive calculations or claims, then is the time to look at as much usable data as possible.

[Show Full Lists]

Post Comment:

A name or alias, email and concise subject are required. Your email will not be abused.
Comments are required to be polite and on topic (commenting etiquette)

Name:*
Email:*
Website:
Subject:*
What is the longhand for 'Oz'? :*


HOME  

  TOP   |   REFRESH PAGE   |   ANALYSIS

fc - 31st May 2010