LSTM-NNW

Energy time-series forecasting experiments for Scalytics Helios.

This repository contains a small LSTM-based smart-meter forecasting prototype. The default demo uses the checked-in smart-meter CSV shards under LST-E/smartmeterdata, while the same CLI can also read customer-owned CSV exports from local paths, S3, or other fsspec-compatible storage.

Quick Start

Install dependencies:

python3 -m pip install -r requirements.txt

Run a preprocessing smoke test against the bundled demo data:

python3 LST-E/LSTEnergy.py --prepare-only --max-rows 5000

Train the default power-total forecaster:

python3 LST-E/LSTEnergy.py

Run the Helios planning API:

uvicorn helios_api:app --reload --port 8000

Run tests:

PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 python3 -m pytest -q

Documentation

• Overview
• LSTM demo
• Data sources and preparation
• Helios and GPU energy forecasting
• API and Grafana integration

Current Status

The current script supports local CSV files/globs, S3 CSV URIs through fsspec/s3fs, schema mapping flags, deduplication, optional clipping/resampling, and auditable preparation summaries before model training. The API exposes the first Helios pipeline slice: telemetry to forecast to optimizer to policy checks to scheduler/EMS/BMS integration boundaries.