Core Concepts
Proposing differentiable decision trees for explainable RL policies in home energy management.
Abstract
Demand-side flexibility in the residential sector is crucial for energy transition. Data-driven control frameworks, like RL, show promise but lack explainability. A novel method using differentiable decision trees is proposed to address this gap. The approach outperforms baseline rule-based policies by 20-25% while providing simple, explainable control policies. The study focuses on a battery-based home energy management system to reduce costs and compares explainable policies with standard RL policies.
Stats
Residential sector accounts for 25% of final energy consumption globally.
Proposed approach outperforms baseline rule-based policies by 20-25%.
Quotes
"We propose a novel framework for explainable RL that uses differentiable decision trees and policy distillation."
"Our proposed approach can learn satisfactory control policies that outperform an RBC included with standard batteries."