Scholarlink

<p>Hybrid Renewable Microgrids (HRMGs) face challenges in balancing energy supply from renewable sources with fluctuating demand and storage limitations. This paper proposes a hybrid approach combining XGBoost for accurate renewable energy forecasting and Deep Reinforcement Learning (DRL) for adaptive energy distribution control. The XGBoost model predicts short-term energy generation and demand, while the DRL controller dynamically adjusts energy flow to optimize operational efficiency. Experimental results demonstrate that under the adopted simulation environment and dataset, the proposed XGBoost-PPO framework achieved 97.4% operational cost reduction, 98.3% energy efficiency improvement, and 92.5% renewable energy utilization. These results are supported by comparative, ablation, convergence, and statistical analyses; however, further validation on multiple real-world microgrid datasets is required before broader generalization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing Energy Distribution in Hybrid Microgrids Using XGBoost and Deep Reinforcement Learning

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • S. Alagammal,
  • R. Bhavani,
  • S. Arockiaraj,
  • A. Bhuvanesh

摘要

Hybrid Renewable Microgrids (HRMGs) face challenges in balancing energy supply from renewable sources with fluctuating demand and storage limitations. This paper proposes a hybrid approach combining XGBoost for accurate renewable energy forecasting and Deep Reinforcement Learning (DRL) for adaptive energy distribution control. The XGBoost model predicts short-term energy generation and demand, while the DRL controller dynamically adjusts energy flow to optimize operational efficiency. Experimental results demonstrate that under the adopted simulation environment and dataset, the proposed XGBoost-PPO framework achieved 97.4% operational cost reduction, 98.3% energy efficiency improvement, and 92.5% renewable energy utilization. These results are supported by comparative, ablation, convergence, and statistical analyses; however, further validation on multiple real-world microgrid datasets is required before broader generalization.