Reproduce “An Agentic AI Pipeline for Appliance-Level Energy Anomaly Detection and LLM-Driven Recommendations” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Dihia Falouz, Aida Douaibia, Amine Bechar, Youssef Elmir, Abbes Amira, Adel Oulefki

An end-to-end appliance monitoring system combines SSA-LSTM forecasting, appliance-specific LSTM-VAE anomaly detection, retrieval-augmented LLM diagnosis, and rule-based safety controls. On seven office appliances, the forecaster reached 1.32% WAPE and R²=0.9976 overall, while DeepSeek-v4-pro scored 90.4/100 and passed all 16 diagnostic scenarios; the local Qwen2.5 7B model also passed every scenario with 85.4/100.

Key insights

  • Forecasting quality differs sharply by appliance: predictable devices such as refrigerators and printers are easier to model, while intermittent devices such as microwaves and water dispensers require appliance-specific thresholds and reliability estimates.
  • Dynamic retrieval preserved the diagnostic scores of static retrieval while reducing the context supplied per event from six sources to roughly three to six.
  • A modular design allows numerical detectors to provide evidence while the LLM handles interpretation, making it possible to use a local model and to inspect or constrain individual decision stages.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.