Electricity and water demand forecasting, hour by hour and together — see the summer peak, and what it does to your reservoirs, before it arrives
Plan for the summer peak before it arrives.
See one case, screen by screen ↓At a utility whose power stations also make drinking water, the load forecaster publishes the day-ahead hourly forecast by 10:00 for unit commitment and fuel nominations, and the water production planner forecasts water demand by zone in a spreadsheet to set the MSF units and RO trains. The dispatch centre then applies its standing rules — for example, curtail interruptible desalination load when the reserve margin at peak is under 10 % — with no check on the water side. When a new weather bulletin lands mid-morning, only the power model is re-run.
So nobody sees where the two collide. A heatwave pushes the 3 pm peak up, the reserve margin drops under the trigger, the rule cuts the RO trains that fill a reservoir, and the evening rise in water demand takes that reservoir under its minimum by midnight. Heat days, dust haze on rooftop solar, school terms and Ramadan — which moves every year with the lunar calendar — are where forecast error lives, and a monthly MAPE report averages that tail away.
Estimated split for one day with a revised weather bulletin, in hours across the forecasting, water and dispatch desks, by hand and with the solution.
Six agents forecast power and water together and check one against the other; the lead forecaster publishes and the shift manager approves.
Lumina Utilities runs power stations that also make drinking water. On Wednesday 12 August 2026, Leila Farouk has already published Thursday's day-ahead forecast. Then the weather centre raises Thursday's maximum. Here is what happens next, screen by screen, in the working solution.
Leila's 06:00 run, published at 06:41, had Thursday's peak at 9.58 GW, water at 596 MIGD and the Northgate reservoirs at 47 % at midnight — no conflict. The 09:40 bulletin from the Lumina National Weather Centre puts Thursday at 48.6 °C at 15:00, up 2.9 °C on the 05:30 bulletin, with dust haze from 09:00 to 17:00. The heatwave band is drawn across all four lanes of the chart before anything is re-run.
“Heatwave in the 09:40 bulletin · not yet in the forecast”
Leila presses Re-forecast both. The weather reader takes the bulletin and the 51-member ensemble (47.3–49.5 °C at 15:00). The calendar keeper confirms a working day on summer office hours, 07:30–14:30, with schools on holiday until Mon 24 Aug. Both forecasters then run on that one weather and calendar, and the run log fills in the side rail, step by step, with a time on each.
Power demand forecaster: “Thursday 15:00: 9.58 → 10.12 GW (P10–P90 9.89–10.35); above the 9.84 GW record”
Forecast v2 against v1, each change attributed to its driver: temperature +2.9 °C adds 0.41 GW, humidity 0.06, dust haze on 1.92 GW of rooftop solar 0.05, building heat carried from Wednesday 0.02. EV charging and the calendar were checked, no change. Beside the bars, the bulletin text the weather reader used, each line cited, and the weather by zone that both forecasts used.
“The temperature effect is strongest from 13:00 to 17:00, when air conditioning is at full load.”
At 10.12 GW the reserve margin at peak is 9.2 %, under the 10 % trigger, so rule DR-4 cuts Marram Point RO trains 5–8 — 76 MIGD and 61 MW — from 13:00 to 19:00. Water demand rises from 596 to 612 MIGD, most of it after 17:00, when people come home. Northgate, which feeds North Hills, Airport & Free Zone and Eastvale, reads 57.7 % on SCADA at 10:00 and falls to 34 % by midnight.
“Northgate falls under 40 % at 19:00 and reaches 34 % at midnight. A power-only forecast would not have shown this.”
A: keep Marram Point at full and run peaker GT-11 — Northgate 45 %, $78,400, 218 t CO₂. B: pre-fill Northgate through the north main from Sandhill at 24 MIGD, 22:00 tonight to 20:00 Thursday — Northgate 46 %, $5,700, 45 t CO₂, and the RO cut still frees 61 MW for the peak. C: ask the north zones to save 5 % from noon — Northgate 37 %, five hours under the minimum. Each option shows its Northgate path and who acts.
Option B is recommended. The balance checker proposes options only — it never changes a plant schedule or sends a dispatch instruction.
The signature box shows what she is publishing: peak 10.12 GW at 15:00 (P10–P90 9.89–10.35), water 612 MIGD, made from the 09:40 bulletin, replacing v1. She chooses the meaning — “Approved for dispatch and production planning” — and signs with her password. On publish, v2 goes to unit commitment, fuel nominations, the water control room and energy trading.
Only the lead forecaster publishes. The forecasters draft; they never publish.
Leila sends option B to Daniel with a note: “Heatwave Thursday — Northgate short by midnight on the current plan.” Daniel approves at 10:34, and the plan goes to Coral Bay and the water control room. The water control room opens valve NM-3 at 22:00; Hana Qureshi, the water production planner, is informed. Option A would have had Tomas Reyes booking gas for GT-11.
“Option B approved by Daniel Mensah at 10:34 — Northgate 46 % at midnight”
The Forecast screen now shows v2 published at 10:21 and the north main transfer drawn on the coupling strip, Wednesday 22:00 to Thursday 20:00. The storage lane stays above the 40 % line. The same strip shows the coupling the other way: Coral Bay units C1–C4 are held at 1.42 GW overnight, because their MSF units need steam.
“North main opens at 22:00 tonight; Marram Point trains 5–8 still give 61 MW to the peak.”
Every version is kept, with what it was made from, its peak and water total, its status and who signed it. Thursday's v1 shows Replaced 10:21; Tuesday's v2 already carries its actual, 9.52 GW, and its MAPE, 1.7 %. The v1-against-v2 chart shows the change by hour: the largest is +0.54 GW at 15:00.
“Every version is kept · a version is published only with the lead forecaster's signature”
The accuracy tracker scores every published version against actuals: over 90 days, a day-ahead MAPE of 2.4 % for the joint forecast against 3.6 % for the previous model, with the error highest at the afternoon peak and again at 21:00, when EV charging and evening cooling overlap. It has found the Eastvale water forecast 2.6 % low every day since 21 Jul, after 2,140 new connections, and has a retrain ready.
The accuracy tracker never retrains a model without approval — the retrain waits for Aisha Karim.
The same heatwave Thursday, seen by the five people who forecast, schedule, produce and decide — what their day looked like, and what it looks like now.
Six agents forecast power and water together and check one against the other; the lead forecaster publishes and the shift manager approves.
Reads the weather centre bulletin, the 51-member ensemble and the dust bulletin; builds hourly temperature, humidity and solar haze by zone and draws the weather bands. Compares the bulletin with a second weather model and quotes each value with its bulletin line.
Keeps Ramadan, Eid, school terms, holidays, events and office hours for 18 months, each with its source and status. Ramadan and Eid stay “expected” until the moon-sighting announcement.
Forecasts hourly electricity demand for 7 days with a P10–P90 band, net of rooftop solar and EV charging, from weather, calendar and the last 3 years of load — and attributes each change against the last version to its driver.
Forecasts hourly water demand by zone for 7 days with its band, from weather, calendar, connections and the last 3 years of flow — on the same weather run as the power forecast.
Joins both forecasts with the plant plan and rules — MSF steam must-run, RO power draw, transfer limits, rule DR-4 — and projects every reservoir hour by hour. Where a reservoir falls under its minimum or the reserve under its floor, it prices options in storage, reserve, cost and CO₂ and recommends one.
Scores every day's published versions at 01:00 — MAPE by hour, heat days and holidays tracked separately — finds drift when a zone is biased for 14 days, prepares the retrain and runs back-casts.
Publishes; the shift manager approves plan changes. The agents propose; a named person decides.
Ask about any hour of the power or water forecast — or say what to change. Every answer points to its sources.
Why does Northgate run short tonight?
Because the power peak cuts the desalination that fills it. The 09:40 bulletin puts Thursday at 48.6 °C at 15:00, which lifts the peak to 10.12 GW and the reserve margin to 9.2 %. Under 10 %, rule DR-4 cuts Marram Point RO trains 5–8 from 13:00 to 19:00 — 76 MIGD that feed Northgate. With the evening rise in demand, Northgate falls under 40 % at 19:00 and reaches 34 % at midnight.
Compare the fixes for Thursday
Three options, priced with today's fuel and energy prices. A: run peaker GT-11 instead of cutting RO — Northgate 45 % at midnight, $78,400. B: pre-fill Northgate through the north main from 22:00 tonight — 46 %, $5,700; recommended. C: ask the north zones to save 5 % — 37 %, still under the minimum.
What if Thursday reaches 50 °C?
At 50.1 °C — 1.5 °C above the bulletin and beyond the ensemble's 90th percentile of 49.5 °C — the peak would be 10.40 GW and the reserve margin 6.3 %, under the 8 % floor, so dispatch would need more capacity. Northgate would reach 33 % at midnight without a fix, or 45 % with option B.
How close were we last Ramadan?
Over Ramadan 2026 the joint forecast's day-ahead MAPE was 3.4 %, against 6.8 % for the previous model. The gain is the late-night peak after 22:00, which the old model treated as a normal evening. Ramadan 2027 is expected from 8 February, to be confirmed by moon sighting; it is already in the calendar.
13 screens from the working solution, on its sample data. Pick one to see it large.
Reservoir minimums, the dispatch rules the balance checker applies, the publish deadline and who decides are settings. Changes are logged and need the owner shown.
| Setting | Default | Choose from |
|---|---|---|
| Water storage minimums | Northgate 40 % | A % for each reservoir — Northgate 40 %, Sandhill 35 %, Harbour towers 40 % |
| Curtail interruptible desalination (rule DR-4) | Under 10 % | The reserve margin at peak below which Marram Point RO trains 5–8 (61 MW) are cut, 13:00–19:00 |
| Reserve margin floor | 8 % | Below this, dispatch must add capacity |
| Coral Bay must-run for MSF steam | 1.42 GW | Units C1–C4 minimum while MSF units C1–C4 run |
| North main transfer limit | 32 MIGD | Sandhill → Northgate |
| Day-ahead publish deadline | 10:00 | The time unit commitment and fuel nominations read it |
| Re-forecast on every new weather bulletin | On | On · Off — both forecasts together, as a draft for the lead forecaster |
| Heat band | ≥ 46 °C or +2 °C on the last run | Draws a heatwave band and widens the confidence band; the P10–P90 band is shown on every chart and sent with each version |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate: $72,700 is the peaker option's $78,400 (390 MWh at $201) less the north main option's $5,700 (110 MWh at $52) for the one day, and 173 t is 218 t less 45 t of CO₂; the split of hours is an estimate of desk practice · “typical” = published figures (Tao Hong, EnergyBiz 2015, as summarised on Wikipedia, not re-checked — about $300K a year for a 1 GW utility per point of short-term MAPE; CAISO root cause analysis of August 2020, via Western Resource Advocates; Sydney Water / Data61 pump scheduling simulation, about 15 % lower energy cost). Lumina Utilities, its people, plants, reservoirs, zones and the Lumina National Weather Centre named on this page are fictional — characters and sample data in the working solution.
It is the hourly and day-ahead forecast of how much electricity and how much water a system will need, used for unit commitment, fuel nominations and desalination scheduling. Electricity and Water Demand Forecasting makes the two forecasts together, on the same weather and calendar, so a utility whose plants make both can see where one forces the other.
The power demand forecaster forecasts hourly demand for 7 days with a P10–P90 band from weather, calendar, rooftop solar, EV charging and the last 3 years of load. Each change against the last version is attributed to its driver — in the sample case, +0.41 GW from temperature, +0.06 from humidity and +0.05 from dust haze on rooftop solar.
Because the same plants make both. A power peak can trigger a rule that cuts RO trains, and MSF units need steam that holds power units on overnight. The power–water balance checker pushes the power result through the plant rules into every reservoir's path, so in the sample case the 34 % midnight level at Northgate showed up a day ahead, where a power-only forecast would not have shown it.
The weather reader draws a heatwave band and widens the confidence band at 46 °C or 2 °C above the last run, and accounts for dust haze on rooftop solar. The calendar keeper keeps school terms, holidays and Ramadan, which stays “expected” until moon sighting. Last Ramadan the joint forecast's back-cast MAPE was 3.4 %, against 6.8 % for the previous model.
The accuracy tracker scores every published version against actuals at 01:00 each night, by hour of day, with heat days and holidays tracked separately. When a zone is biased for 14 days it prepares a retrain — in the sample, the Eastvale water forecast had been 2.6 % low every day since 21 Jul.
Yes. The forecasters draft and the lead forecaster publishes each version with an e-signature. The balance checker proposes priced options but never changes a plant schedule or sends a dispatch instruction; the dispatch shift manager approves any change to the plant plan, and the head of system planning approves retrains.
Yes. The minimum for each reservoir, the reserve margin that triggers desalination curtailment, the reserve margin floor, must-run limits, transfer limits, the publish deadline, re-forecasting on each bulletin, the heat band, the weather sources and who decides are all settings, and every change is logged.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your load and water history, weather feeds, calendar, plant plans and dispatch rules — and it goes live once every quality gate has passed. We will walk you through it first on a back-cast of your own last heatwave.
We'll run Electricity and Water Demand Forecasting on a back-cast of your own last heatwave.