LF Energy Summit Europe 2026

Home Energy Management Systems (HEMS) increasingly operate in non-stationary conditions: changing tariffs, evolving household behavior, new DER assets, and fragmented device ecosystems. Most current deployments still rely on static rules and periodically retrained models, which degrade under drift and are hard to maintain at scale.

This session presents a self-evolving HEMS architecture that closes the loop across three core tasks:
(1) classification for safety/reliability monitoring,
(2) regression for PV/load forecasting, and
(3) optimization that converts natural-language user goals into executable schedules.

An LLM-centered orchestration layer coordinates rule engines, ML models, user/context memory, and control APIs to support transparent interaction, adaptive decision-making, and continuous improvement. We share field-oriented case studies showing improved robustness in detection, better forecasting performance, and practical end-to-end scheduling execution.

The talk focuses on reusable architectural patterns and deployment lessons relevant to open, software-defined energy systems.