winonecheng/optimization-of-electricity-demand-reduction-with-comfort-conscious-learning

Home appliance controller learn from resident negative feedback

Optimization-of-Electricity-Demand-Reduction-with-Comfort-Conscious-Learning

Oral defense slides

Dataset

Smart* Data Set for Sustainability - UMass Smart* Dataset - 2017 release

Datetime encoding

Encoding cyclical continuous features - 24-hour time

Learning Algorithm

• VPG
• PPO

Reward Evaluation

• 需量反應負載管理措施
• 臨時性減少用電措施－限電回饋型

Script

• training.ipynb: take action from PPO and define environment with applianace and resident feedback
• metrics.ipynb: calculate four evaluation metrics (power, reward, comfort, dr_rate)