Scholarlink

<p>In recent years, the increasing complexity and uncertainty of large-scale energy systems have posed significant challenges for adaptive demand-side optimization under dynamic and non-stationary operating conditions. Traditional demand-side management (DSM) approaches often struggle to respond effectively to real-time variations in load and generation profiles. To address this limitation, this paper proposes a hybrid artificial intelligence framework that integrates deep learning, reinforcement learning, and physics-informed neural networks for energy demand forecasting, adaptive load control, and grid stability enhancement, respectively. A hybrid Proximal Policy Optimization–Particle Swarm Optimization (PPO–PSO) algorithm is employed to combine policy-based learning with population-based optimization, enabling improved convergence, adaptive decision-making, and robust performance under uncertainty. The proposed framework generates adaptive load adjustment signals using real-world energy datasets and facilitates consumer-level participation in demand response programs. Validation is conducted across multiple real-world-inspired scenarios, including residential energy optimization, commercial building scheduling, and industrial load management. Experimental results demonstrate substantial improvements in peak load reduction, energy cost savings, forecasting accuracy, and overall grid stability when compared with baseline DSM methods and standalone learning approaches. The proposed Hybrid PPO-PSO model demonstrates significant improvements in demand-side management, achieving up to 27.1% peak load reduction, 18.9% energy cost savings, and 40–50% faster convergence, with statistically significant performance gains across multiple real-world scenarios. The study indicates that hybrid reinforcement learning-based optimization provides a scalable, flexible, and effective solution for intelligent energy demand management, supporting the development of next-generation smart energy systems that are sustainable, reliable, and low-carbon.</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

A hybrid PPO–PSO reinforcement learning framework for adaptive demand-side energy management

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月18日

查看全文
  • P. Kiran,
  • Aparna S. Menon

摘要

In recent years, the increasing complexity and uncertainty of large-scale energy systems have posed significant challenges for adaptive demand-side optimization under dynamic and non-stationary operating conditions. Traditional demand-side management (DSM) approaches often struggle to respond effectively to real-time variations in load and generation profiles. To address this limitation, this paper proposes a hybrid artificial intelligence framework that integrates deep learning, reinforcement learning, and physics-informed neural networks for energy demand forecasting, adaptive load control, and grid stability enhancement, respectively. A hybrid Proximal Policy Optimization–Particle Swarm Optimization (PPO–PSO) algorithm is employed to combine policy-based learning with population-based optimization, enabling improved convergence, adaptive decision-making, and robust performance under uncertainty. The proposed framework generates adaptive load adjustment signals using real-world energy datasets and facilitates consumer-level participation in demand response programs. Validation is conducted across multiple real-world-inspired scenarios, including residential energy optimization, commercial building scheduling, and industrial load management. Experimental results demonstrate substantial improvements in peak load reduction, energy cost savings, forecasting accuracy, and overall grid stability when compared with baseline DSM methods and standalone learning approaches. The proposed Hybrid PPO-PSO model demonstrates significant improvements in demand-side management, achieving up to 27.1% peak load reduction, 18.9% energy cost savings, and 40–50% faster convergence, with statistically significant performance gains across multiple real-world scenarios. The study indicates that hybrid reinforcement learning-based optimization provides a scalable, flexible, and effective solution for intelligent energy demand management, supporting the development of next-generation smart energy systems that are sustainable, reliable, and low-carbon.