Demand charges bill the single highest 30-minute peak of the month. A battery can flatten that peak — but it holds only so much energy. The reservoir law decides, minute by minute, how fast to release it: enough to shave, never so much that it empties before the real peak lands.
The meter integrates power into a 30-minute average. Your demand charge is set by the one highest such average in the whole month — a single half-hour can define the bill. Shaving that half-hour is the entire game; the rest of the month is watching for it.
This day has two peaks — an afternoon stressor and a higher, billed peak in the evening. Drag the evening peak up toward the ceiling and watch two things move together: the battery discharge that fills the gap down to the target, and the reservoir level draining below. As the peak nears the ceiling, the release tapers — the reservoir stops draining and holds a reserve for a genuine breach.
Every site is calibrated to three levels from its own history. Together they turn one battery into a strategy that behaves differently on a quiet day than on a dangerous one.
The deepest the target ever goes. On a quiet day the peak is shaved down toward the floor and no further — going lower just burns cycles for demand you were never going to be billed for.
When the running peak crosses the guard, the day is flagged as a potential spike day. The law stops draining freely and starts pacing — because the real peak may still be ahead.
The highest demand the battery commits to defend at full power. Between guard and ceiling the release tapers to zero, so the reservoir arrives full — with everything in hand for a true breach.
The release rate scales down smoothly as load climbs from the guard to the ceiling — shave = Sdyn × (1 − approach), where approach runs 0 at the guard to 1 at the ceiling. It keeps shaving the whole way up, so nothing is wasted if the feared breach never comes; yet it self-preserves energy near the ceiling, so a real breach is still met at full power. One rule, no thresholds to tune.
The rule above is the surface. In production the same reservoir runs with more on top — flip the Production layers in the diagram to see them move.
Between peaks the demand dips into a lull. The controller uses that headroom to top the reservoir back up — kept below the floor, so it can never lift the billed 30-min peak — leaving more energy for whatever comes next. Watch the reservoir recover before the evening peak.
The controller carries a forecast of the rest of the day and ring-fences the energy a later peak will need — the red floor the reservoir stays above. An early peak can spend only what's truly spare, so a bigger one later is never left undefended. λ sets how hard it rations.
Production also diverts surplus PV into the battery instead of curtailing it, derates power on temperature, and anchors a recharge to the floor — details left out here so the core idea stays legible.
Illustrative model of the EWISER EDM reservoir control law, simulated live in your browser — the target, taper, valley recharge and λ reserve run the same decision logic as the production engine. Levels and battery size are representative; each production site is calibrated from its own metered history. Drag Peak demand and toggle the Production layers to explore. Colours match the EMon platform: load · PV · battery · grid · reservoir.