Reinforcement Learning-Based Energy Management of Smart Home with Rooftop Solar Photovoltaic System, Energy Storage...

Area of Science:

• Smart Grid Technology
• Artificial Intelligence in Energy Management
• Renewable Energy Systems

Background:

• Home energy management systems (HEMS) often rely on model-based optimization.
• Existing methods may not fully capture the dynamic nature of energy consumption and user preferences.
• Integrating renewable energy sources like solar photovoltaic (PV) systems presents unique management challenges.

Purpose of the Study:

• To develop a data-driven HEMS using reinforcement learning (RL) for optimal energy consumption.
• To enhance RL efficiency with artificial neural network (ANN) predictions for accurate appliance behavior modeling.
• To reduce household electricity costs while ensuring user comfort and appliance operational constraints.

Main Methods:

• Application of a model-free Q-learning algorithm for scheduling smart home appliances (air conditioner, washing machine) and energy storage system (ESS) operations.
• Utilizing an ANN to predict indoor temperature, aiding the Q-learning agent in learning the AC's energy-temperature relationship.
• Simulations conducted on a single-home model with solar PV, ESS, and time-of-use (TOU) electricity pricing.

Main Results:

• The proposed Q-learning HEMS, enhanced by ANN temperature prediction, effectively reduced electricity bills.
• Demonstrated a 14% relative reduction in electricity bills compared to existing optimization approaches.
• Maintained user-defined comfort levels (e.g., indoor temperature) and respected appliance operational characteristics.

Conclusions:

• A model-free, data-driven HEMS using Q-learning and ANNs offers a superior alternative to traditional model-based methods.
• This integrated approach optimizes energy consumption from solar PV and ESS, leading to significant cost savings.
• The system successfully balances energy efficiency, economic benefits, and user comfort in smart homes.