基于线性规划和强化学习模型的零能耗住宅运行优化对比研究
Comparison Analysis on Optimization of Zero Energy House Based on Linear Programing and Reinforcement Learning
许文亚 1李岩学 1徐阳 1王子璇1
作者信息
- 1. 青岛理工大学 滨海人居环境学术创新中心,山东 青岛 266000
- 折叠
摘要
零能耗住宅"源"、"荷"、"储"空间上距离接近,更易通过实时管控促进可再生能源的就地消纳,并降低用户用能成本.以零能耗住宅能源系统为研究对象,将基于规则的优化控制作为比较基准,分别基于线性规划、Q-learning和DQN建立以用能成本最小化为目标的优化模型.引入自耗率、支出成本等作为评价参数对比分析3 种优化算法对光伏就地消纳和经济效益的影响.基于2 栋零能耗住宅屋顶光伏发电和电力负荷,在MATLAB仿真环境中进行建模并求解.结果表明,实时电价条件下,一定容量电池对提升光伏就地消纳的能力有限,但能够使用户获得更多的经济收益.线性规划模型取得经济效益最高,强化学习模型可有效避免系统复杂的物理建模过程,同时智能体可以保证较好的系统经济性调配效果.
Abstract
In zero energy house,"renewable energy","load"and"storage"are close in space.Through real-time management and optimization,it is easier to realize the local consumption of renewable energy and reduce the energy cost of users.The zero energy house microgrid is selected as the research objective.The optimization model based on established rule is used as a benchmark.The optimization model aims at minimizing energy cost based on linear programing,Q-learning and DQN algorithms,respectively.Secondly,this work introduces the self-consumption ratio and relative expenditure cost indicators to analyze the impact of the three optimization algorithms on photovoltaic local consumption and economic benefits.The simulation model is built in the MATLAB environment based on PV generation and electricity load of two zero energy houses.The results show that the capacity of a certain capacity battery to improve PV local consumption is limited,but it can enable users to obtain more economic benefits in responding to real-time electricity price.Simulation results shown the linear programing gets the largest cost saving.The reinforcement learning model effectively avoids the huge efforts of energy system modeling.Meanwhile,the agent can ensure an effective economic dispatch scenario.
关键词
优化运行/基于规则/线性规划/Q学习/深度Q网络/经济效益/就地消纳/零能耗住宅Key words
optimal operation/rule-based/linear programming/Q-learning/deep Q network(DQN)/economic benefits/local consumption/zero energy house引用本文复制引用
基金项目
山东省自然科学基金资助项目(ZR2021QE084)
国家重点研发计划资助项目(2018YFE0106100)
出版年
2023
NETL
NSTL
维普
万方数据