Operation of power systems - Wind Power Integration: Connection and System Operational Aspects - page 181

For this mode to be fully effective, state of the art, meteorologically based
wind forecasting tools are required, with a prediction horizon of at least 4 hours and
more acceptably 24-48 hours. Even longer time horizons can be beneficial
for managing limited hydro reservoir reserves. Significant errors in longer-term
forecasts, say 12 hours hence, can be corrected, at a cost, by additional unit
start-ups. Beyond this time horizon, forecasts can assist in maintenance scheduling.
For shorter time horizons, an overly optimistic wind power forecast could lead to a
shortfall of online generation to meet the current demand and/or seriously under-
mine the system's operating reserve. Additional load-following requirements may
be placed on the scheduled conventional plant, impacting on unit maintenance costs
and plant life expectations. Generally speaking, an under-prediction of wind
generation will not cause great concern to the TSO since scheduled plant can be
backed-off (tending towards fuel saver mode), with positive ramping rates applied
to sufficient wind farms if required.
It may not be possible to redeem a shortfall by starting up additional conven-
tional plant: drum boiler steam plant can typically be brought from start-up to full
load in 2-4 hours (for hot plant) and 6-10 hours (for cold plant), while
the equivalent cold-start figures for once-through plant (3-4 hours) and CCGTs
(5-6 hours) are noticeably reduced. OCGTs and diesel engines offer much shorter
start-up times, but at the expense of higher operating cost and CO 2 emissions.
Conventional generation also tends to have many scheduling constraints, often
embodied within connection agreements, which are linked to the cost of hot, warm
and cold starts. Severe limits can be placed on both the minimum run time of
individual units and the number of start-ups allowed during a given period, e.g. 2 hot
starts per day or up to 200 per year, and up to 50/10 warm/cold starts per year.
All of these factors can place restrictions on the operational flexibility of
the units, and consequently the generation mix will determine the required wind
forecast prediction horizon - the more flexible the units, the later unit commitment
decisions can be delayed. For example, a power system with a large capacity of
quick-start plant (such as OCGTs, hydroelectric generation, pumped storage
schemes, etc.) can more readily cope with large forecasting errors than a power
system comprised mainly of less responsive plant (such as nuclear power stations
or CHP schemes). An optimum level of scheduling aggression can be envisaged,
whereby the more accurate the wind power forecast and the higher the associated
confidence level of the predictions (both of which may be subject to weather
patterns), the more confidently the system operator can de-commit conventional
plant in readiness for wind generation. So, for example, during periods of lower
prediction confidence (typically associated with the passage of storm fronts when
there may be rapid and large changes in wind output) upper limits could be placed
on the amount of wind generation that can be accepted, and/or tighter restrictions
imposed on the rate of change of wind power.
As discussed further in Chapter 6 (Wind Forecasting), for up to 3-4 hours
ahead simple persistence forecasting methods, whereby it is assumed that wind
generation output will persist at its current level, perform reasonably well. Alter-
natively, statistical regression techniques, which follow the recent trend in wind