Exactly how much energy should your company have consumed last week? If you know the
answer, not only will you be able to spot unexpected waste, you will be able to evaluate,
objectively, the effect of any energy-saving measures. Vilnis Vesma tackles a difficult subject
head-on

Ihave written this article to accommodate
plants making a changing mix of
products with different energy
intensities, with weather influences thrown
in. And I am going to try to explain it all
without formulae or graphs.

So imagine a hypothetical commercial
bakery making just bread loaves. The
amount of gas used in the bakery each week
is partly fixed (because of standing heat loss
from the ovens and other purely timerelated
uses). On top of that, a variable
amount of gas would be needed in
proportion to production, because each
tonne of bread they make requires a certain
amount of heat, energy in effect ’embedded’
in the product. Suppose that the fixed
weekly consumption were 81,000kWh and
that on top of that an additional 190kWh
were needed for each additional tonne of
bread baked. I will explain later how those
numbers might have been arrived at, but if
you know them, hopefully it is self-evident
that for a given quantity of bread produced
in a week you could easily calculate the
expected gas consumption (here you can
test yourself:Would you agree that
119,000kWh would be needed for a week
when 200 tonnes were produced?).

Now suppose they add two new product
categories: tarts, which require 310kWh per
tonne to cook and rolls (250kWh per
tonne), while the ovens in which tarts and
rolls are produced add 42,000kWh per week
to the standing load, bringing it up to
123,000kWh per week. Finally they install a
gas-fired warm air heating system which
adds nothing to the standing load but
consumes 1200kWh per degree day (a
degree-day value is a single number,
calculated from outside air temperatures,
which represents how cold the week was).

The numbers I have quoted describe how
the bakery’s weekly gas consumption should
respond to weekly changes in product
throughputs and weather (what we call the
‘driving factors’). You can see them in Figure
1 (below), a table which the energy manager
would use for routine weekly management.

In this table the variable weekly drivingfactor
values are entered in the middle
column, while on the right is a column
which is simply the product of the driving
factor and its associated coefficient. So, for
example, 5 tonnes of tarts (embedded
energy coefficient 310kWh per tonne)
account for 1550kWh of the total.

The grand total in Figure 1 is the total
expected energy consumption and provides
a dynamic yardstick against which to assess
the week’s actual gas consumption.

Technical faults, poor maintenance or
sloppy operation will all cause actual
consumption to diverge from the ideal and
suspected excess consumption can thus be
quantified straight after the end of the week.

But what about the individual coefficients
in the first column of the table? How might
they have been arrived at? One method is a
statistical technique called multiple
regression. It is available in Excel. Given
historical weekly values for each of the
driving factors (and for the total actual
consumption) it computes values for the
coefficients which, when applied to the
historical data, would have given the least
error when comparing actual and expected
consumption. Preferably some of the
coefficients should be individually
estimated. The options for establishing
individual coefficients include the following:
Collect data more frequently (daily rather
than weekly, for instance) while the plant
happens to be making only one product;
Analyse performance at a different plant
which only makes the single product
grade you are interested in;
If one product grade is made exclusively
on a distinct process line, fit a temporary
sub-meter;
Compute the coefficient from first
principles; or even
Use an accepted industry norm
The first three options still rely on
regression analysis, but only against a single
variable, which is more reliable as it easier to
spot and eliminate rogue data.Multiple
regression can if necessary be used to mop
up any coefficients that could not be
independently evaluated.

What I have demonstrated in this article
is the use of a straightforward and hopefully
obvious ‘mathematical model’ to calculate
expected energy consumption. It is a
method which can be applied wherever two
or more measurable driving factors
influence energy consumption (cooling
demand and available daylight, for
example).

More complex models can be constructed
if circumstances dictate; but for most
situations a single driving factor will often
suffice. The main point is to have a good
estimate – better than any other method can
achieve – of how much energy you should
have used in a given period. And that is the
first step on a very rewarding journey.

Vilnis Vesma is an independent specialist
in energy monitoring and targeting. If you
would like details of his training courses on
the subject, information can