Dynamic appliance scheduling and energy management in smart homes using adaptive reinforcement learning techniques

Area of Science:

• Smart Home Energy Management
• Artificial Intelligence
• Optimization Algorithms

Background:

• Traditional home energy management systems struggle with dynamic user preferences and costs.
• Existing reinforcement learning methods often lack advanced optimization integration.

Purpose of the Study:

• To develop a novel Demand Response (DR) method for smart homes.
• To improve energy consumption, cost management, and user preference adaptation.
• To enhance energy efficiency through intelligent appliance scheduling.

Main Methods:

• Integration of Self-Adaptive Puma Optimizer Algorithm (SAPOA) with Multi-Objective Deep Q-Network (MO-DQN).
• SAPOA adaptively maximizes multiple objectives; MO-DQN enhances decision-making through interaction learning.
• Utilizing past energy usage patterns for preference adaptation and appliance scheduling optimization.

Main Results:

• Dramatically reduced Peak-to-Average Ratio (PAR) from 3.4286 to 1.9765 (without RES) and 1.0339 (with RES).
• Demonstrated superior performance compared to MORL-POA, SAPOA, and POA methods.
• Efficiently managed uncertainty, improving overall system performance and flexibility.

Conclusions:

• The proposed SAPOA-MO-DQN approach offers a flexible and high-performing solution for smart home energy management.
• Effectively optimizes appliance scheduling and energy usage while adapting to user preferences.
• Significantly lowers energy costs and improves grid stability through reduced peak loads.