Research question
How can monthly billing evidence support sanctioned-load review when actual maximum-demand measurements are unavailable?
Approach
Train XGBoost Q70/Q80/Q90 models on observed next-month kWh. In a separate evidence layer, use observed E/H as average power, a physical lower bound on maximum demand, with consecutive-month persistence.
Contribution
A no-pseudo-target workflow linking observed energy forecasting with conservative evidence for review. No assumed load factor or synthetic demand label is used.
Evidence & limitations
Energy quantiles are not maximum-demand predictions. The lower-bound test can miss exceedance. Any load change requires actual-demand or field verification and the applicable approval process.
Next step
Review the manuscript and verification protocol with the utility before external submission.
Methodology stages
01 / Observed monthly kWh
02 / Quantile forecasting
03 / E/H lower bound
04 / Verification & review