源荷不确定冷热电联供微网能量调度的建模与学习优化

Modeling and learning-based optimization of the energy dispatch for a combined cooling, heat and power microgrid system with uncertain sources and loads

中文关键词  冷热电联供微网  能量调度  马尔科夫过程  强化学习

英文关键词  combined cooling, heat and power microgrid system  energy dispatch  Markov process  reinforcement learning

中文摘要

      针对含光伏, 微型燃气轮机组等分布式能源的冷热电联供微网系统, 研究源荷双侧不确定情况下多类型能 量调度动态优化问题. 首先, 针对光伏出力和异类负荷的随机不确定性, 将光伏和负荷的变化描述为连续马尔科夫 过程; 然后以决策时刻, 负荷需求以及分布式能源出力的离散值为状态分量, 以微型燃气轮机组启停行动和储能充 放行动为动作分量, 在分时电价模式下, 以降低包括购电成本, 燃料代价, 启停代价等在内的日运行成本为调度优化 目标, 将源荷不确定冷热电联供微网系统调度动态优化问题描述为马尔科夫决策过程模型, 并引入强化学习方法对 该问题进行策略求解. 最后通过算例仿真对不同策略进行了比较, 验证了优化方法的有效性.

英文摘要

      The dynamic dispatch optimization of the combined cooling, heat and power (CCHP) microgrid system with uncertain renewable sources and demands is focused in this paper. Firstly, the variations of photovoltaic and loads are described as continuous Markov processes considering their random properties. Then, define state vector of the system which consists of decision epoch, multiple load demands level, and outputs level of distributed energy sources (DESs), and the action vector which consists of the actions of micro turbines (MT) and storages. The time-of-use electricity price mode is applied in the system to minimize operating cost including electricity purchasing cost, fuel cost and starting-stopping cost. The dynamic optimal dispatch problem for CCHP microgrid system is described as a discrete Markov decision process (MDP), and a reinforcement learning method is adopted to obtain the optimal or suboptimal policy. Different policies are compared in simulation part and it shows that optimal policy can achieve a better performance to reduce the daily operating cost of the system. At last, simulation experiments including the comparison of different policies are performed to validate the effectiveness of the method.