dreamlab-ai UK Ground-Mounted Solar Planner
An end-to-end planner for UK utility-scale ground-mount solar, composing our qgis and blender skills with public solar data (PVGIS), UK-tuned forecasting (Open Climate Fix quartz), and OpenDSS grid modelling. It turns a site (coordinates + area) into a defensible capacity, yield, layout, and grid-feasibility picture.
The pipeline (run in order; each stage feeds the next)
| # | Stage | Tool / skill | Status here |
|---|---|---|---|
| 1 | Site suitability — slope/aspect, ALC land grade, flood zone, grid proximity | qgis skill + references/grid-and-planning.md |
works (qgis sidecar) |
| 2 | Tilt + spacing + capacity + yield — the engineering core | solar-optimize (+ pvgis-fetch) |
works now (stdlib + live PVGIS) |
| 3 | 3D layout + shadow review — array on terrain, inter-row winter shadows | tools/blender_solar_array.py (blender skill) |
works (blender / gui-tools sidecar) |
| 4 | Generation forecast — 0–48h UK-tuned generation curve | tools/sidecar_run.sh forecast … (quartz) |
needs sidecar rebuild¹ |
| 5 | DNO grid feasibility — voltage rise, thermal, hosting capacity | OpenDSS MCP (register on demand) | needs sidecar rebuild¹ |
¹ Stages 4–5 use Python deps nix can't cleanly package (quartz pulls ML weights at runtime;
opendssdirect vendors a compiled engine), so they're baked into the FHS gui-tools sidecar.
Bring them online with ./agentbox.sh gui-tools rebuild then ./agentbox.sh gui-tools up.
Stages 1–3 need no rebuild.
Stage 2 — the engineering core (start here for any sizing question)
solar-optimize --lat 52.2 --lon -1.5 --area-ha 20 --tilt land
Computes, from first principles (see references/siting-engineering.md for the cited
formulas): the winter-solstice no-shade row pitch and GCR, the tilt (either the
PVGIS per-module optimum ~38–43°, or the land-optimised ~25–35° that maximises kWh/hectare),
the module count and DC capacity, the annual yield (live PVGIS PVcalc), and the
land-use ratio sense-checked against the DESNZ 2024 fleet band (1.88–2.7 ha/MW). It emits
warnings when a result falls outside real-world bounds (e.g. GCR too dense) rather than
silently returning an optimistic number.
pvgis-fetch is the standalone PVGIS client (annual/monthly yield + PVGIS's own
optimal tilt/azimuth) — stdlib, no deps.
Stage 3 — 3D layout + shadow review (leverages the blender skill)
# headless GPU-capable, or run in the gui-tools sidecar:
blender --background --factory-startup --python tools/blender_solar_array.py -- \
'{"rows":8,"modules_per_row":24,"tilt_deg":30,"row_pitch":9.9,"sun_altitude_deg":15,"render":true}'
Builds the racked array on a ground plane, reports GCR/footprint, and (with a sun at the
winter-solstice noon altitude, 90 − lat − 23.44) casts the inter-row shadows that set
the row pitch. Feed row_pitch/tilt from Stage 2. See the blender skill for GPU/sidecar
execution and rendering.
Stages 4–5 — forecast + grid (sidecar)
tools/sidecar_run.sh forecast 52.2 -1.5 14420 30 180 # lat lon kWp tilt orientation → 0–48h curve
Grid feasibility uses the OpenDSS MCP server (baked in the sidecar); register it on demand for
native tools, or script it via tools/sidecar_run.sh python …. references/grid-and-planning.md
covers UK DNO connection (G99, LV/11kV/33kV by capacity), what a first-pass OpenDSS model needs,
planning (ALC "Best & Most Versatile" avoidance, NSIP/DCO >50 MW), and SEG export economics.
References
references/siting-engineering.md— tilt/GCR/PVGIS formulas, UK yield bands, GIS inputs (all cited).references/grid-and-planning.md— DNO connection, OpenDSS pre-check, UK planning + ALC, SEG.
Attribution
PVGIS (EU JRC, free public API). quartz-solar-forecast (Open Climate Fix, MIT). opendss-mcp-server (MIT) wrapping EPRI OpenDSS. All engineering figures are cited to primary sources in the references; the optimiser and layout tools are original implementations of standard, non-proprietary methods.