基于深度强化学习的多能流楼宇低碳调度方法
A low-carbon scheduling method for multi-energy flow buildings based on deep rein-forcement learning
胥栋 1李逸超 1李赟 1徐刚 1杜佳玮1
作者信息
- 1. 国网上海市电力公司浦东供电公司,上海 200122
- 折叠
摘要
建筑减排已成为中国达到"双碳"目标的重要途径,智慧楼宇作为多能流网络耦合的综合能源主体,面临碳排放量较多、多能流网络耦合程度高、负荷用能行为动态特性明显等问题.针对这一问题,提出基于深度强化学习的多能流楼宇低碳调度方法.首先,根据智慧楼宇的实际碳排放量,建立了一种奖惩阶梯型碳排放权交易机制.其次,面向碳市场和多能流耦合网络,以最小化运行成本为目标函数,建立多能流低碳楼宇调度模型,并将该调度问题转换为马尔可夫决策过程.然后,利用Rainbow算法进行优化调度问题的求解.最后,通过仿真分析验证了优化调度模型的可行性及有效性.
Abstract
Building emissions reduction has become a crucial pathway for China to achieve its'dual-carbon'goals.As an integrated energy entity coupled with multi-energy flow networks,smart buildings face challenges such as high carbon emissions,a high degree of coupling in multi-energy flow networks,and distinct dynamic characteris-tics in load energy consumption behavior.In response to these challenges,a low-carbon scheduling method for multi-energy flow buildings based on deep reinforcement learning(deep RL)is proposed.Firstly,a reward and punish-ment ladder-type carbon emissions trading mechanism is established based on the actual carbon emissions of smart buildings.Secondly,targeting the carbon market and multi-energy flow coupling networks,a low-carbon scheduling model for multi-energy flow buildings is developed,aiming to minimize operating costs as the objective function,and the scheduling is transformed into a Markov decision process(MDP).Subsequently,the Rainbow algorithm is employed to solve the optimal scheduling.Finally,the feasibility and effectiveness of the optimal scheduling model are verified through simulation analysis.
关键词
"双碳"目标/多能流/低碳调度/深度强化学习Key words
'dual-carbon'goals/multi-energy flow/low-carbon scheduling/deep RL引用本文复制引用
基金项目
国网上海市电力公司浦东供电公司营销项目(640921220001)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据