论文部分内容阅读
对于处在碳配额市场条件下以乙醇胺(MEA)进行碳捕集的燃煤电厂,本文应用了基于强化学习的Sarsa时间差分算法为其自行搜寻一种统一的竞标和运行策略。电厂的决策者的目的被定义为最大化电厂寿命下的贴现累计利润。其中,我们引入以下两个限制条件:一是碳捕集的高能耗和电力生产之间的权衡;二是碳排放交易市场中竞得的碳配额数量与电力生产导致的实际碳排放量的近似相等。本文给出了三个案例方便研究。第一个案例中,我们展示了Sarsa算法将收敛到一个确定且优化的竞标和运行策略。第二个案例中,相互独立设计的运行和竞标策略与统一设计的运行和竞标策略相互比较,以表明加入了随时间变化、市场导向的碳捕集水平后,Sarsa算法将有助于电厂决策者获得更高的贴现累计利润。第三个案例则引入了处在同一碳配额市场的另一电厂作为原电厂的竞争对手。两家电厂设置了相同的发电和二氧化碳捕集设备,但新电厂采用不同的策略获得利润。比较两家电厂的贴现累计利润,结果表明:采用Sarsa学习算法、找到统一的竞标和运行策略的原电厂会更具竞争力。
For a coal-fired power plant that uses carbon tetrachloride (MEA) for carbon capture under the carbon market conditions, this paper uses Sarsa time-difference algorithm based on reinforcement learning to search for a unified bidding and operation strategy. The purpose of the power plant’s decision maker is defined as maximizing the discounted cumulative profit over the life of the plant. Among them, we introduce the following two constraints: First, the trade-off between high energy consumption and electricity production of carbon capture; second, the number of carbon credits competed in the carbon emissions trading market is similar to the actual carbon emissions resulting from electricity production equal. This article gives three convenient case studies. In the first case, we show that the Sarsa algorithm converges to a well-defined and optimized bidding and running strategy. In the second case, the run-and-run strategy for independently designed operations and bidding strategies are compared to the one for uniform design to show that Sarsa’s algorithm will contribute to power plant decision-making after the addition of market-oriented carbon capture over time Get higher discounted cumulative profits. The third case introduced another power plant in the same carbon quota market as a competitor of the original power plant. The two plants have the same power generation and CO2 capture facilities installed, but new plants use different strategies to generate profits. Comparing the discounted cumulative profits of the two power plants, the results show that a power plant using the Sarsa learning algorithm to find a uniform bidding and operating strategy would be more competitive.