# Dutch Energy Supplier Load Forecasting Challenge — Data

来源：https://www.kaggle.com/competitions/dutch-energy-supplier-load-forecasting-challenge/data

采集时间：2026-10-20T05:47:45Z

**Granularity: 15-minute intervals.

**Timezone:** UTC (no DST shifts).

**Target: **Net aggregated load = Consumption is negative and generation is positive.

**Units:** energy per 15-min interval (e.g., kWh per interval).

**Files:**

- train.csv & test.csv

- timestamp_utc 

- net_load_kwh (numeric, may be negative)

Allowed: Open-Meteo historical/reanalysis weather only. Align by UTC and resample/aggregate to 15-minute steps as needed (e.g., forward-fill, interpolation, or engineered lags).

Leakage rule with latency: Respect a 3-day data latency: at time t you may only use information available ≤ t−72h.

With 15-minute data, that’s a 288-step cutoff (72 h × 4 steps/h).