
62
Chapter 4
control room engineers, needs to know how much power will be needed
in the next hour. While both predictions could conceivably be made by
the same neural network based on the same historical data, there are
two reasons for not doing so.
The obvious reason is that the time scaling of the requisite data
may be different. The yearly prediction would most likely be based on
yearly or monthly data. Any finer resolution would be excess baggage.
Also,
their history would need to go back many years if large scale
patterns were to be detected. In contrast, the hourly prediction would
naturally need hourly (or finer) data, and it would a ...