# 2035 supply and vehicle-flexibility screening

Edition: 18 September 2026. Author: WattInfo.

The purpose is to quantify the electricity still to procure after an illustrative first supply package, and the incremental contribution of vehicle flexibility. Results are deterministic scenarios, not probabilities or bankable capacity commitments.

## Proposed supply and demand

Demand reuses the independent sector model at 2035 (central adoption path), with weather-dependent hourly shapes calibrated to the complete 2024 calendar. The same national demand and driving requirement are used for each flexibility comparison. The stock sensitivity changes only the cohort available for flexibility; it is not a separate reconciled transport-demand forecast. Heating allocation retains the earlier model's end-state intensity ratios for splitting combined space/water components; this approximation and historical calendar shapes need further validation.

Supply retains Gösgen at 1.010 GW net and Leibstadt at 1.233 GW net. Beznau is absent after its announced retirements; new nuclear has no 2035 delivery credit. This bridge sits alongside, not instead of, the proposed new-nuclear development programme. Each remaining unit has 30 assumed maintenance days in spring/summer and 98% availability otherwise. Stress runs remove Leibstadt for the first 90 calendar days. Neither the outage nor its timing has an assigned probability.

The proposed 26 GWp PV fleet uses the existing eight-station radiation proxy, calibrated to 1,100 kWh/kWp in 2024 and clipped at nameplate. The increase from the BFE 2025 estimate of 9.4986 GWp implies 1.65014 GWp/year over ten years. This is a build proposal, not a contracted pipeline. No inverter, snow, geographical build plan or distribution-grid validation is added. Historical river output and derived reservoir inflows correspond to each historical weather year. Monthly averages smooth actual hourly hydro variability. Other thermal and wind generation retains the 2024 monthly observed totals, multiplied by a 98% assumed delivery factor and spread uniformly through each month. Retaining this generation in 2035 requires continued viable operation; these sources are not omitted and then presented as a new gap. No unapproved hydro-project energy is added.

Hydro reuses four finite pumped stores and a separate pooled natural-inflow store from the earlier feasibility edition. Natural capacity is 8,510.546275 GWh; finite pumped stores total 211.453725 GWh. Existing modelling simplifications in lake allocation, gross/net conversion and natural turbine power remain. Only the pumped stores can absorb grid electricity. River output, inflow and stores do not create extra annual energy. All stores have cyclic initial/final conditions; the starting fill is optimized with perfect foresight and can be optimistic. Contracts, water rights, local constraints, reservoir cascades and profitability remain outside this national model.

Most runs permit at most 2 GW of imports; one permits none. That availability is an assumption, not guaranteed emergency access. Export capacity is capped at 8 GW. There is no new stationary battery or fuelled reserve in the first package. The remaining energy and peak are requirements for additional supply; neither equals a specified plant nameplate. Available power, fuel, outages, legal operation and emissions must be established before counting backup.

## Fleet turnover and vehicle services

The BFE/EBP May 2023 study projects 2.1 million battery-electric passenger cars in 2035. Its assumption of more than 50% plug-in new-car registrations in 2025 was ahead of the subsequent BFS outcome (22.7% BEV plus 11.2% PHEV). This does not determine the 2035 result but warrants a slower stock sensitivity.

The simple sales-paced case starts from 249,832 BEVs within 4,839,465 passenger cars at 30 September 2025. It holds total cars and annual registrations at 232,602, and assumes a linear increase in the BEV new-sales share from 22.7% in 2025 to 100% in 2035. Annual stock = previous BEV stock × (1 − 232,602 / 4,839,465) + 232,602 × that year's assumed BEV share. The equal retirement rate for young electric and older combustion fleets is a simplification, not an empirical survival model. Stock measured in September and registrations over the calendar year are approximate starting boundaries. The 100% assumption is not a statement of current law. Results are a sensitivity, not a forecast or confidence limit.

All service cases use the same cohort: 50% compatible cars × 50% equipped among compatible cars × 40% willing participants among equipped cars. Every fraction is an assumption. Connection at equipped homes is assumed for 70% of that cohort during 18:00–07:59 local time, and 20% during 08:00–17:59. The clock windows are separate from astronomical day/night. No empirical national connection schedule is claimed. A fixed aggregate pool does not reproduce individual cars leaving with different charge states; its result is an optimistic availability envelope.

Per vehicle: 60 kWh usable pack; protected 20 kWh minimum and 50 kWh maximum operating state; 12,000 km/year at 0.18 battery kWh/km; 90% charging and 90% export conversion efficiency; 7 kW charge limit. The 30 kWh working window is not all the car's battery. Driving removes the same annual battery energy in every service case, equally allocated to 07:00 and 17:00 travel withdrawals. The pool must meet travel and minimum charge constraints throughout the year, with no free end-of-year energy. This proves aggregate energy conservation, not individual trip readiness.

Fixed charging uses the earlier evening-heavy shape. Smart charging reschedules that allowance, with no export. V2H adds discharge limited to an assumed 2 kW local household load per connected car. V2G permits 7 kW per connected car. The same cars and energy cannot be counted in both pools. V2H only offsets assumed local load; actual household profiles and electrical approval remain untested. V2B shares the same building-supply principle but has no separately added national pool. V2L serves appliances and adds no national grid resource here. Islanded operation during a blackout requires suitable installation and is not established by these scenarios.

For flexible cases, the participating cars' original charging allowance is removed from the fixed passenger-car load. Explicit car charging is then added, driving withdrawn at the battery, and V2H/V2G discharge subtracted once. Annual charging × 0.90 = driving + exported electricity / 0.90. A constant national loss uplift converts household charging/avoided load to the model bus. This symmetric aggregate conversion is a simplification; local grid losses and constraints need a network model.

## Day/night and optimization

Daylight uses NOAA's solar-position equations at the midpoint of each UTC hour, a stated reference point of 47° N, 8° E and zenith ≤90.833°. Winter means October–March. Cloudy daylight remains day. A country-wide PV profile and a single reference point can place small dawn/dusk PV quantities in a nominal night bin; the solar profile is not forcibly zeroed. Annual weather labels are hydrological years: e.g. 2012 means October 2011–September 2012, not future weather predictions.

The linear programme first minimizes unserved annual energy, then minimizes the peak shortfall while preserving that optimum, then minimizes imports and needless cycling at those bounds. Numerical allowances are 10^-7 GWh/GW; conservation checks permit 2×10^-5. Objective weights in the last stage are priorities, not market prices. Hourly dispatch is one feasible optimum; alternative schedules can share the same annual and peak results. Day/night shortages and individual-hour patterns are therefore illustrative allocations, not unique forecasts. Full-year perfect foresight, pooled hydro and absence of network constraints make results optimistic. No least-cost portfolio, owner profit, loss-of-load probability or 2050 sovereignty conclusion follows.

## Reproduction

Use Python 3.12 with numpy 2.5.3, pandas 3.0.5, scipy 1.18.1 and openpyxl 3.1.5 in a virtual environment. From the extracted bundle root run `python -m unittest portfolio_model.test_model`, `python -m portfolio_model.run`, and `python -m portfolio_model.validate`. Existing case results are reused; to recompute independently, use a separate copy of the bundle without its generated `data/portfolio-2035-2026-09-18/cases` directory. Preserve the original evidence copy. The input manifest records hashes and provenance pointers. Original third-party PDFs are available through source links, not republished in the bundle.