HyLL: Hybrid LSTM and Large Language Model-Based Recommendation System for Energy Management and Consumption Optimization in Smart Homes
Amin Aslanzadeh Moghanjooghi, Mahdi Mazloomi, Mohammad Asgari, Mohammad Sadegh Mehrabikia · 2025
The reasoning capabilities and broad general knowledge of Large Language Models (LLMs) make them a natural choice for interpreting user requests and managing energy in smart home contexts. However, LLMs lack specific knowledge about individual homes and the ability to predict future energy consumption, limiting their effectiveness in energy optimization and management. This paper introduces HyLL, a hybrid framework that integrates Long Short-Term Memory (LSTM) networks for energy consumption prediction with an LLM, such as GPT-4o, enhanced through in-context learning for real-time optimization and energy management. The LSTM models predict daily energy usage for ten household appliances using historical data and weather conditions, ensuring accuracy through iterative updates. The LLM utilizes 24-hour energy predictions to guide users in shifting their energy consumption to off-peak hours and reducing usage during peak demand periods. By combining predictive modeling with AI-driven decision-making, HyLL improves energy efficiency, alleviates grid stress, and reduces consumer costs, underscoring the critical role of AI in sustainable smart home energy management.