This milestone tests the core research question with transparent, reproducible analysis rather than heavyweight ML. Methods must match the data we have (PVWatts outputs, tariff data, and user load/cost assumptions) and clearly communicate uncertainty.
- Objective: Recommend feasible PV + battery size bands for U.S. households using public data, while surfacing savings, payback, and backup-hour ranges.
- Inputs: PVWatts v8 production (
ac_annual,ac_monthly,capacity_factor), URDB/NREL utility tariffs, user-entered load and system cost assumptions (seescripts/config.py), optional SREC mappings. - Methods: Scenario tables across PV kW × battery kWh, bar/line charts for annual savings and payback, sensitivity on tariffs and system costs, and backup-hour estimates for critical loads.
- Uncertainty handling: Present ranges (best/base/conservative) to reflect weather variability, tariff coverage gaps, and simplified load profiles; state caveats explicitly.
- Upstream dataset:
2_data_collection/data/processed/solar_analysis_dataset.csv(8 locations). Schema and pipeline are documented in2_data_collection/data/README.md; upstream scripts pull Nominatim geocodes, PVWatts v8 production, Utility Rates v3 tariffs, and static SREC/ITC/cost tables. - Upstream artifacts: Raw API pulls live in
2_data_collection/data/raw/(geocode_results.json,pvwatts_results.json,utility_rates_results.json). Visuals from M2 (irradiance, rates, SREC maps) are in2_data_collection/data/visualizations/; load profile mock is2_data_collection/loadprofile.png. - Fields consumed here:
ac_annual_kwh,ac_monthly*,capacity_factor,electricity_rate_residential,utility_name,srec_price_per_mwh,system_cost_net,annual_savings,simple_payback_years,annual_srec_revenue,data_quality_score. - Known limitations (must be repeated in results): PVWatts uses TMY (±10–15% variance), URDB has ~15% utility gaps (state-average fallback), no shading or azimuth/tilt optimization (tilt = latitude, south-facing), 14% fixed losses, SREC prices are static snapshots.
- Refresh guidance: If you change locations, rerun the M2 pipeline before analysis (
fetch_geocode.py → fetch_pvwatts.py → fetch_rates.py → clean_merge_dataset.py → generate_visualizations.py). Keep raw pulls in2_data_collection/data/raw/and outputs inprocessed/.
- Status: Final figures are not yet documented in this folder. The Flask prototype (
scripts/) renders best/base/conservative scenarios interactively, but the static non-technical summary and technical appendix still need to be exported here. - Required actions to complete M3:
- Run the analysis notebook or script to generate savings/payback charts for representative ZIPs (e.g., CA, TX, FL) and export visuals.
- Write the non-technical summary that explains results, confidence levels, and caveats for homeowners.
- Document the technical methods, assumptions, and alternative approaches considered.
- Add group/individual retrospectives and note survey completion.
- Environment:
python -m venv .venv && source .venv/bin/activate && pip install -r scripts/requirements.txt. - Data pull: Prefer reusing the processed CSV from M2. If regenerating, run the M2 pipeline in
2_data_collection/data/(see that README) withNREL_API_KEYset in.envto refresh geocode/PVWatts/URDB data and SREC mappings. - Execution: Place notebooks or scripts for analysis under
3_data_analysis/and export figures to this folder (or to4_Communicating_Results/if reused for M4). Record command order and expected inputs/outputs alongside the artifacts. - Inputs/outputs: Keep any new raw pulls in
data/raw/(git-ignored) and processed outputs indata/processed/; include a short README if new files are added. If you modify the schema, update both this file and2_data_collection/data/README.md.
Illustrative PV + battery options from the prototype: compares backup duration, usable capacity, costs, payback, and long-run profit for different battery sizes at a fixed PV size.