A demand meter records the highest average power a facility drew, but everything hinges on the window that average is taken over, the demand interval. Two facilities with identical load profiles can record different peaks depending on whether that window is a fixed block or a sliding one, because the same spike can be diluted by one method and captured by another. This guide goes a level deeper than the demand meter itself: it explains what a demand interval is, how fixed and sliding windows average power differently, and how SCADA-side demand prediction lets operators shed load before an interval closes so they never set a new peak.
Demand Interval in one line: A demand interval is the window of time over which a meter averages power to determine demand, commonly a set number of minutes such as fifteen. A fixed or block interval divides time into consecutive, non-overlapping windows and averages within each one, while a sliding or rolling interval continuously averages over the most recent window, moving forward in small steps. The averaging method matters because a short spike can be diluted or captured differently depending on the window type, so the same load can produce a different recorded peak under a fixed window than under a sliding one.
A fixed or block interval carves time into back-to-back windows of equal length that do not overlap, and computes one average for each. The clock effectively resets at the end of every block, so the meter reports a sequence of separate averages, one per window, and the peak demand is the largest of those block averages. The defining feature is that where a load falls relative to the block boundaries affects the result: a burst of power neatly contained within one block raises that block's average more than the same burst split across the boundary between two blocks, where each half is diluted by the quieter portion of its own block.
A sliding or rolling interval removes that dependence on boundaries by continuously averaging over the most recent window of the same length, advancing in small increments rather than resetting. At every step it looks back over the last interval's worth of time and computes that average, so the window is always moving and always the same length, just anchored to the present moment instead of to fixed marks on the clock. Because it never resets, a sustained burst is captured wherever it occurs; there is no lucky or unlucky alignment with block boundaries, since every possible placement of the window is effectively examined as it slides.
The practical difference is how forgiving each method is of load placement. Under a fixed window a facility can sometimes benefit from a spike straddling a boundary, because the split dilutes it across two blocks and no single block average rises as high. Under a sliding window that escape is closed, because a window will always exist that fully contains the spike, so the sustained draw is caught at its true intensity. Neither method is more correct in the abstract; they are simply different definitions of demand, and a facility has to know which one its tariff uses to understand and manage its recorded peaks.
Consider a heavy load that runs for a short time, less than a full interval, and imagine it happening to start just before the end of a fixed block. Under fixed-window averaging, part of the load falls in the first block and part in the next, and each portion is averaged together with the lighter draw that fills out the rest of its block. The result is that neither block average rises to reflect the true intensity of the load, because in both blocks the spike is surrounded by quieter time. The peak the meter records understates how hard the facility actually pulled, purely because of where the spike landed relative to the boundary.
Now take exactly the same load under a sliding window. As the window rolls forward, at some moment it lines up to contain the entire burst within a single interval, and at that instant the computed average reflects the burst combined with only as much surrounding time as the window length requires, without any boundary splitting it. That rolling average therefore climbs higher than either fixed block did, and it is that higher value that becomes the recorded peak. The identical physical event produces a larger demand figure under sliding averaging than under fixed, simply because the sliding method finds the worst-placed window rather than accepting whatever the fixed boundaries happened to give.
The reverse case is equally instructive. A load that runs steadily for the full length of an interval registers the same under both methods, because a sustained draw fills a whole window regardless of how the window is defined. The two averaging methods diverge only for loads that are short or unevenly placed relative to the interval, which is exactly where the choice of window matters most. Knowing this tells a facility where its exposure lies: brief coincident bursts are the events whose recorded cost depends on the averaging method, and they are therefore the events worth managing most carefully.
Because a peak, once an interval records it, cannot be undone, the useful moment to act is before the interval closes, and this is where SCADA-side demand prediction comes in. By watching the power drawn so far in the current interval, a system can project where the interval's average is heading if the present load continues, and warn when that projection is climbing toward a level that would set a new peak. A cloud SCADA platform such as Merobix can surface this developing figure to operators, turning the abstract idea of a demand interval into a live number they can respond to while the window is still open.
With a projection in hand, operators can shed or defer load before the interval closes to keep its average below the threshold that would create a costly new peak. If the prediction shows the current window trending high, non-essential or deferrable loads can be paused or staggered so that the average over the full interval settles back down before it is locked in. The value of prediction is precisely that it acts in time: it does not merely report the peak after the fact, it gives operators the warning needed to prevent the peak from being set in the first place.
The window type shapes how this prediction has to work. Under a fixed block the system knows exactly when the current window ends and can count down to that reset, aiming to bring the average down before the block closes. Under a sliding window there is no reset to aim for, so the projection has to account for the fact that the average continuously reflects the most recent interval, making sustained load the thing to control rather than timing relative to a boundary. In both cases the goal is the same, to use a live, forward-looking view of demand so the facility never quietly sets a new peak it will pay for, but understanding whether the interval is fixed or sliding is what makes that intervention effective.
A fixed or block interval divides time into consecutive, non-overlapping windows and averages power within each one, so where a load falls relative to the boundaries affects the result. A sliding or rolling interval continuously averages over the most recent window, moving forward in small steps and never resetting. Because the sliding method effectively examines every possible window placement, it captures short bursts that a fixed window might dilute across a boundary.
Because the two methods average the same load differently. A short spike that straddles a fixed block boundary is split between two blocks and diluted by the quieter time in each, understating the peak. A sliding window will always find a placement that fully contains that spike, capturing its true intensity and recording a higher peak. Loads that run steadily for a full interval register the same under both; only short or unevenly placed bursts diverge.
By watching the power drawn so far in the current interval, a SCADA system can project where the interval's average is heading and warn when it is trending toward a level that would set a new peak. Operators can then shed or defer load before the interval closes to bring the average back down. Because a recorded peak cannot be undone, this forward-looking view lets a facility prevent a costly new peak rather than merely observe it afterward.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.