In such circumstances, different kinds of data (specified operating temperatures, batch duration, charge weights, number of starts and stops, etc.) may need to be recorded and fed into a mathematical model to arrive at a theoretical energy requirement given the pattern of production activity.
The model need not be complex, although indeed it may be so. The only requirement, from an M&T standpoint, is that the model yields a single number representing an index of energy demand. Note that there will probably be a different model for each energy source feeding a monitored process: one for steam, one for electricity, and so on as the case may be.
The models used for this purpose often have a great deal in common with those used for fine-grained demand forecasting, except that they have to be fed with actual achieved, rather than planned, production schedules.