Scholarlink

Achieving a circular construction industry requires reducing waste, reusing materials, and recycling resources (the 3Rs). However, traditional methods for implementing these goals mostly depend on static strategies that do not adapt to changes in demand or resource availability. This paper proposes an approach that integrates Digital Twins (DTs) and Multi-Agent Reinforcement Learning (MARL) to optimize the 3Rs dynamically and collaboratively. The DTs continuously simulate construction processes, enabling real-time monitoring of waste generation, costs, and energy consumption. Meanwhile, multiple agents learn actions, via deep reinforcement learning, that jointly minimize material waste while balancing cost and carbon footprint targets. The synergy between MARL and DT is demonstrated through a simulated scenario in which each agent specializes in different interventions (e.g., recycling, scheduling, logistics). Results show that this integrated approach outperforms baseline strategies, no intervention (No-Op) and random actions, significantly reducing average waste and improving recycling rates. These results highlight the potential of intelligent, data-driven frameworks to advance sustainability in the construction industry, paving the way for large-scale adoption of circular economy principles.

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

Simulation-Based Optimization of the 3Rs in Circular Construction Using Integrated Digital Twins and Multi-agent Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Eduardo Guzmán,
  • Sandra Tobón,
  • Beatriz Andrés,
  • Marta Torres

摘要

Achieving a circular construction industry requires reducing waste, reusing materials, and recycling resources (the 3Rs). However, traditional methods for implementing these goals mostly depend on static strategies that do not adapt to changes in demand or resource availability. This paper proposes an approach that integrates Digital Twins (DTs) and Multi-Agent Reinforcement Learning (MARL) to optimize the 3Rs dynamically and collaboratively. The DTs continuously simulate construction processes, enabling real-time monitoring of waste generation, costs, and energy consumption. Meanwhile, multiple agents learn actions, via deep reinforcement learning, that jointly minimize material waste while balancing cost and carbon footprint targets. The synergy between MARL and DT is demonstrated through a simulated scenario in which each agent specializes in different interventions (e.g., recycling, scheduling, logistics). Results show that this integrated approach outperforms baseline strategies, no intervention (No-Op) and random actions, significantly reducing average waste and improving recycling rates. These results highlight the potential of intelligent, data-driven frameworks to advance sustainability in the construction industry, paving the way for large-scale adoption of circular economy principles.