Most thermal fossil fired generation is designed to be most efficient at full load. Large coal and oil units are typically 36% efficient at max ouput dropping to 32% at half load. CCGTs can be 55% efficient at maximum, but only 40% when at half load.
Max ramp rates are around 10MW/minute and other dynamic restrictions apply when operating thermal units - minimum stable generation, minimum run and shutdown times. In addition, each unit requires a miminum notice to synchronise and will consume start up heat to bring it on load. Both of these increase with the time the generator has been shut down.
The variations in the daily demand curve dictate that a number of generators start up for the plateau and peak periods of the day. Some demand rises are so fast (up to 3000MW/hhr in GB) that a number of units will be ramping simultaneously. At all times, some units are also part-loaded for response, reserve and spare duty, to cover unexpected demand or generation changes. Units have to be ordered far enough in advance that they will synchronise at the correct time
It is vital that the demand curve is accurately predicted and generation is reliably operated to avoid unneccessary part loading, allocation of excess reserve or ordering of generators that aren't actually needed in the event. Prediction, reliability and timing are the key to efficient operation.
The conventional power plant is designed to be controllable for instruction following. Thus, its output is predictable for the purpose of Generation-Demand matching. Even so, allowances have to be made to cover the risk of plant breakdown; response, reserve and spare output is carried to cover the anticipated level of generation shortfall and failure as against the instructed output.