Utility electricity bills have two parts: energy (kWh, priced by time of day) and demand (the highest 15-minute kW in the month). A pump scheduler that only shifts kWh out of peak hours can miss the second part. We simulated it.
Setup. EPANET's Net3 example network through WNTR, at a 15-minute hydraulic step. We simulated all 21,806 schedules of the lake pump with at most two on-runs a day and billed each on PG&E Schedule B-19 (published rates, effective March 2026, secondary voltage), with the tanks back to within 0.1 % of a day's demand.
Result. Net3's shipped schedule peaks at 373 kW because a 310 kW booster pump runs whenever tank 1 is low. Demand charges are 60 to 64 % of its monthly bill. Running the lake pump nearly all day keeps the tank full, idles the booster and cuts the peak to 63 kW: the bill falls 56 to 67 % depending on season (62.5 % over a year). The schedule that minimises energy cost and the one that minimises the full bill are the same, and stay the same when we scale the demand rates from 0 to 20 times. Demand charges make the saving bigger (39 % at no demand charge, 65 % at twenty times) but do not change the answer.
Caveats. Net3 is a teaching network with an unusual gravity river source; B-19 is mandatory above 499 kW so this is what the tariff would charge, not a tariff a utility this small is on; list prices only; one repeated day (see the follow-up below). Not operating advice.
Code, numbers and limits: research/infra-pumpdemand.
Follow-up: one hot day
The caveat "one repeated day" matters, because a demand charge is set by the highest 15-minute kW of the month. We re-simulated all 21,806 schedules with every Net3 demand scaled by 0.8 to 1.6 and billed 30-day months with one, three or ten high days. With demand just 1.5 % above Net3's shipped level the lowest feasible peak among one-run schedules jumps from 63 kW to 310 kW (the booster), and at +20 % it is 372 kW, level with the shipped schedule. A month with three days at +20 % saves 43.5 % in summer and about 19 % in winter and spring, against 56, 66 and 67 % with identical days. Each day is simulated as periodic, which overstates how soon the booster is needed; carrying tank levels between days is the next step. Numbers and limits: research/infra-pumpdays.