NZ Grid Simulator

Step 1, Load historical data

On launch the simulator fetches two datasets for the selected year:

• Half-hourly generation and demand, 17,520 rows for the selected year, with separate columns for North Island / South Island metered generation by fuel type, grid demand, and HVDC flows.
• Installed capacity, the actual installed capacity (MW) per fuel per island for that year, used to compute scaling factors when you change the slider values.

Each row covers one 30-minute trading period. Timestamps are NZ Standard Time (UTC+12). The model simulates all 17,520 periods sequentially and carries state (reservoir levels, battery SoC) forward between them.

Step 2, Scale variable renewables to target capacity

Wind and solar output in each period is multiplied by a capacity scale factor:

For example, if the base year had 2.1 GW of NI wind and you set the slider to 4.2 GW, every NI wind half-hour value is doubled. This preserves the real capacity factor and temporal variability of wind and solar, it does not assume a different shape, only a proportionally larger fleet.

Solar yield correction, if you set a target yield (kWh/kWp/year), the scale factor is also adjusted so annual solar output matches the yield, compensating for differences in panel orientation and technology between the base year fleet and a new installation.

Step 3, Model residential solar and battery

Behind-the-meter (residential) solar and battery are modelled separately from grid-scale assets. For each half-hour:

1. Residential solar generation is estimated by scaling utility-scale solar output by the ratio of home panel capacity to grid capacity, multiplied by the number of homes with solar (set by the % slider × total dwellings per island, NI 1.57 M, SI 0.55 M).
2. Self-consumption, residential solar first meets the household's own load (NI 38% / SI 25% of island demand by default, based on MBIE data; the SI share is lower because Tiwai Point aluminium smelter inflates industrial demand). Surplus is exported to the grid.
3. Home battery, if enabled, surplus residential solar charges the home battery before export. The battery discharges during peak periods: in summer all day (24 h); in winter mornings and daytime only (06:00–21:00); in shoulder months from 05:00 onwards. The net effect of home batteries is that less residential solar reaches the grid but grid demand in the evening is reduced.

The residential solar that reaches the grid is included in national and island generation totals as "Residential solar & storage".

Step 4, Dispatch each half-hour (merit order)

North Island and South Island are dispatched independently, then linked via the Cook Strait HVDC cable. Within each island, generation is allocated in strict merit order:

1. Must-run generation, geothermal (100% of metered output) and run-of-river hydro (NI 70%, SI 20% of metered hydro) are dispatched first at full output and cannot be curtailed.
2. Variable renewables, scaled wind and solar (including residential net export) are added next. Output follows the historical shape and cannot be switched off, but surplus is curtailed if demand is fully met.
3. Flexible hydro dispatch, remaining hydro (the reservoir portion) is held back initially. A virtual reservoir tracks accumulated inflow minus discharge. If the NI island has a deficit after wind/solar, flexible hydro is released up to the available reservoir level. SI hydro is managed similarly but with a much larger reservoir (Waitaki/Manapouri); SI reservoir losses (evaporation and spill) are proportionally higher. Reservoir is capped at configured maximum; overflow is spilled (curtailed).
4. HVDC surplus transfer (first pass), if one island has a surplus after renewables and flexible hydro, it sends power to the other island up to the HVDC capacity limit (default 750 MW southward, 1,050 MW northward). Low-carbon surplus is transferred before thermal is committed on the receiving island.
5. Cross-island flexible hydro (HVDC), if either island has idle flexible hydro left from this period's flow, it covers the other island's remaining deficit over spare HVDC headroom. This is the cheapest cross-island option because it spends no stored water.
6. Reservoir hydro, local then cross-island (HVDC), stored water is then released to meet any deficit still outstanding. Crucially, hydro may now be pushed above the historical output for that half-hour, up to the fleet's demonstrated turbine capacity, drawing the reservoir down. If one island's lakes still have spare turbine capacity, that water serves the other island's deficit over remaining HVDC headroom, so a full South Island reservoir can displace North Island gas rather than being stranded behind the link. In a dry year, depleted storage falls back to thermal as before.
7. Grid-scale battery, the NI utility battery charges from any remaining NI surplus (up to power rating and remaining capacity), then discharges to cover any remaining NI deficit. A separate SI battery can be configured. Battery round-trip efficiency is 92%.
8. Thermal dispatchables, NI thermal plant is committed in order: Wood/biomass → Gas → Diesel → Coal → Other. Each type is dispatched up to its installed capacity until the remaining deficit is met. Thermal plant can serve NI demand directly, or if SI has unmet demand it can export via the remaining HVDC headroom. There is no SI thermal in the model (Huntly is in NI; Tiwai-era SI thermal is not modelled).
9. Unserved energy, any demand not met after all plant is dispatched is recorded as unserved. This represents grid insecurity rather than modelled load shedding.

Dry-year smart dispatch (optional). The standard merit order is greedy, it spends flexible hydro and battery energy on whichever deficit comes first, so by the worst windless evening they can be depleted and gas fills the peak. When you press “Run vs Dry Year” with Smart dry-year dispatch enabled, the model instead runs a day-ahead, peak-aware dispatch: for each day it builds the North Island net-demand profile (demand minus must-run, wind, solar and run-of-river) and “water-fills” it against that day's available flexible-hydro and battery budget to find a flat gas floor. Stored hydro and the grid battery are then held back to shave only the peak above that floor, the overnight battery pre-charge runs deeper (40% C-rate to 7 am), and gas runs flat at the lower floor. This lowers the peak gas MW wherever there is battery power to shave with (it cannot beat physics, at the single worst instant the peak is still set by installed turbine and battery power), and it conserves reservoir water so the lakes last. The toggle lets you compare smart dispatch against today's run-as-it-comes operation on the same fleet and weather.

Step 5, Carry state forward

Two pieces of state persist between half-hour periods:

• Hydro reservoirs (NI and SI), each period, the reservoir fills from inflow (metered actual hydro output that was not dispatched as run-of-river), and drains when flexible hydro is dispatched. Reservoir losses (evaporation/seepage) of 0.5 MWh/period NI and 4.5 MWh/period SI are subtracted. The reservoir starts at 25% of capacity (NI 500 GWh, SI 4,500 GWh). Spill occurs when the reservoir overflows above its maximum.
• Battery state of charge (SoC), the NI utility battery (default 1,000 MW / 8,000 MWh) starts at 50% SoC. Each period the SoC changes by ±(charge or discharge) in MWh. SoC cannot go below 0 or above the configured capacity.

Step 6, Calculate operability metrics

After dispatch, the model calculates three grid security metrics for each period:

• Frequency Instantaneous Reserve (FIR, 0–6 s), available from the headroom of online synchronous plant (hydro, geo, gas, coal, wood). Each generator type contributes its headroom scaled by a response fraction. FIR is compared against the contingency requirement: the larger of the HVDC import volume or the largest single-unit risk (400 MW NI, 400 MW SI).
• Sustained Reserve (SIR, 6–60 s), similar to FIR but with larger headroom contribution per unit as plant ramps up further. The contingency is the same as FIR.
• System inertia (GVA·s), synchronous generators contribute inertia proportional to their output and inertia constant H (Hydro H=5.5, Geo H=5.0, Gas H=6.5, Coal H=6.0, Wood H=5.0). Grid-forming batteries provide equivalent synthetic inertia (H=5.0) based on their rated power. Wind and solar are inverter-connected and contribute no inertia. Low inertia increases the rate-of-change-of-frequency (RoCoF) after a fault.

Step 7, Aggregate and display results

Results are accumulated at monthly and annual resolution and displayed in four sections:

• Annual statistics, total demand (TWh), renewable share (%), carbon intensity (gCO₂/kWh), system LCOE (NZD/MWh), average marginal cost (NZD/MWh), curtailment (GWh), unserved energy (%), and fossil share (%). Carbon factors and LCOE are applied per fuel type.
• Monthly generation tables, NI and SI generation by fuel type (GWh/month), with peak output and installed capacity shown.
• Half-hourly timeseries charts, scrollable 7-day or 30-day views of dispatch by fuel, HVDC flows, frequency reserves, system inertia, battery SoC and curtailment. The date window can be stepped forward/backward using the Prev/Next controls.
• Residential energy section, separate breakdown of behind-the-meter solar generation, home battery charge/discharge, and the net load after residential assets.

Key assumptions and limitations

• Single historical weather year, wind, solar and hydro output use one year of real data. Results depend on how wet, windy or sunny that year was. Use the Year selector to compare different weather years.
• Demand is unchanged, the model does not simulate demand growth, EV fleet charging, or industrial electrification. Future scenarios should be interpreted as "how the grid would perform in this weather year if these assets existed, at today's demand." EV batteries are a known future addition: as a load, unmanaged EV charging would increase evening peak demand; as a flexibility resource, vehicle-to-grid (V2G) discharge could reduce it. Both effects are omitted here.
• Transmission is simplified, the model treats each island as a single node. Intra-island congestion (e.g. upper North Island constraints) is not modelled. HVDC is modelled as a reserve-constrained link: actual transfer each period is limited by both the configured capacity cap and the sending island's available spinning reserve (a cable trip must be coverable by local reserves instantly). Intra-island thermal or transmission constraints that further limit practical HVDC loading are not modelled.
• Geothermal is fixed, geothermal output is taken from the historical metered record and is not scalable (it is capacity-constrained at actual plant levels). A future version may allow new geothermal capacity to be specified.
• Hydro reservoir is simplified, the model uses a single reservoir per island. The South Island store represents the Waitaki/Lake Pukaki system, which dominates NZ hydro storage; the North Island store represents the Waikato chain, which has far smaller storage. Individual catchments, water rights constraints, and spill rules are not modelled. Inflow equals metered actual hydro output; there is no separate hydrology model.
• Dry-year data coverage, the dry-year analysis applies a configurable inflow cut (default 20%) to the selected weather year. Historically significant drought years (e.g. early 2021, 2017) are not yet available as separate data inputs; a 20% cut on 2024 data is the current worst-case baseline.
• Transmission losses, gross generation exceeds metered demand by typically 3–5% due to resistive losses in the transmission network. The simulator accounts for this by deriving a per-period, per-island loss factor from historical data (gross generation ÷ metered demand, capped at 10%) and scaling the generation dispatch target up accordingly. Domestic rooftop solar and home batteries are exempt, they generate and consume power at the home, bypassing the transmission network entirely.
• No market price modelling, the LCOE and marginal cost figures use fixed assumed cost parameters (capital/fuel cost per fuel type) and do not reflect the actual spot market.
• Carbon factors, lifecycle emission factors are used (gCO₂/kWh): Hydro 24, Geo 46, Wind 12, Solar 45, Wood 230, Gas 490, Coal 820, Diesel 800. These include upstream and construction emissions.

Method

For each weather year we ran the model at that year's actual capacities and read back its full-year dispatch, then set it alongside the metered generation and demand for the same 17,520 half-hourly periods. Because the base year's wind, solar and hydro shapes are the real ones, an accurate model should closely reproduce the historical outturn, any gap points to a dispatch or accounting error rather than a difference in weather.

Four corrections the validation produced

• Wood pinned to metered output. It is fuel-limited co-generation, not dispatchable, this removed a ~2,000 GWh artifact.
• HVDC losses added. Power now arrives ~4% de-rated across Cook Strait instead of losslessly.
• Loss factors corrected for HVDC. Using each island's full energy balance (~8.6% NI, ~6.4% SI) closed a ~1 TWh generation gap.
• Water value on hydro. Storage below 45% is conserved rather than drained, so winter thermal runs as it does in reality.

Result, annual totals, 2022 to 2025

Difference between modelled and metered generation at historic capacities (percent):

Every generation source matches the metered record to a fraction of a percent. The only material divergence is the small emergent fossil residual, the point of the tool. ¹No grid solar existed in 2022/23 (metered = 0; sim = 0). Coal and gas are treated as interchangeable; the model runs the fossil residual as gas while the real system split it between the two during periods of gas supply constraint.

Result, half-hourly shape, 2024

Annual totals can hide timing errors that cancel out, so we also compared every half-hour. Correlation between modelled and metered dispatch across all 17,568 periods:

The quantities the model genuinely decides, the fossil residual and the interisland flows, follow the real record period by period, including the daily shape of the evening peak. Hydro and wind are near-perfect partly by construction, since they are anchored to the metered inputs. There is no systematic time-of-day or seasonal divergence.