Short Term Load Forecast with LCT Net

This repository implements a full pipeline for short‑term load forecasting (STLF) experiments using a model in the LCT‑Net family — a CNN + Local Context Transformer/attention architecture (CNN_LCT_Att). It bundles utilities for:

• Robust CSV ingestion & data quality (QC) checks
• Weather merging & time‑zone alignment (Meteostat)
• Feature enrichment (cyclic encodings, holiday flags, one‑hots)
• Rolling cross‑validation folds & calendar‑aligned splits
• Windowing & PyTorch DataLoaders (with context)
• Model definition, training loop (EMA, early stopping, checkpointing)
• Evaluation & calibration (raw / offset/affine)
• Plotting helpers for pre/post analysis

Built to support an academic paper, this repo provides the experiment code and a reproducible pipeline.