arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2401.01133eess.SYcs.SY
收藏

不确定性下聚光太阳能热发电的随机-混合整数线性规划调度优化模型

A Stochastic-MILP dispatch optimization model for Concentrated Solar Thermal under uncertainty

Navid Mohammadzadeh, Huy Truong-Ba, Michael E. Cholette, Theodore A. Steinberg, Giampaolo Manzolini

AI总结:

针对CST电站调度中的不确定性问题,提出SMILP模型最大化预期利润,并开发三种启发式求解方法;在115 MW案例中,SMILP在有限历史数据下利润高出基准6%-9%,而大数据下Heuristic-2在保持相似利润的同时大幅节省计算时间。

摘要:

聚光太阳能热发电(CST)为大规模太阳能利用提供了有前景的解决方案,因为热能存储(TES)使得发电不受日常太阳波动的影响,并可转移至高电价时段发电。开发调度规划工具是必要的,以应对太阳辐射和电价预测相关的不确定性以及有限的存储容量。本研究提出了随机混合整数线性规划(SMILP),以在特定场景空间内最大化预期利润。SMILP场景空间由储罐中潜在累积太阳能的不同经验累积分布函数百分位数生成,并使用样本平均近似(SAA)方法估计预期利润。SMILP表现出稳健的性能,然而其计算时间构成了挑战。因此,开发了三种启发式求解方法,它们在不同的历史天气概况上运行一组确定性优化,以生成候选调度计划(DPs)。随后选择在所有概况上平均性能最佳的候选DP。新方法被应用于南澳大利亚一座115 MW CST电站的案例研究。当历史数据库具有有限的历史天气概况集合时,将DP应用于新天气条件,SMILP比最接近的基准实现了6%至9%更高的利润。拥有大量历史天气数据时,SMILP和Heuristic-2的性能变得几乎相同,因为SMILP只能利用有限数量的轨迹进行优化,否则在计算上变得不可行。在这种情况下,Heuristic-2成为一种实用的替代方案,因为它在合理的时间内提供了相似的平均利润(节省了约7小时的计算时间)。

英文摘要:

Concentrated Solar Thermal (CST) offers a promising solution for large-scale solar energy utilization as Thermal Energy Storage (TES) enables electricity generation independently of daily solar fluctuations, shifting to high-priced electricity intervals. The development of dispatch planning tools is mandatory to account for uncertainties associated with solar irradiation and electricity price forecasts as well as limited storage capacity. This study proposes the Stochastic Mixed Integer Linear Program (SMILP) to maximize expected profit within a specified scenario space. The SMILP scenario space is generated by different Empirical Cumulative Distribution Function percentiles of the potential solar energy to accumulate in storage and the expected profit is estimated using the Sample Average Approximation (SAA) method. SMILP exhibits robust performance, however, its computational time poses a challenge. Thus, three heuristic solutions are developed which run a set of deterministic optimizations on different historical weather profiles to generate candidate dispatching plans (DPs). The candidate DP with the best average performance on all profiles is then selected. The new methods were applied to a case study for a 115 MW CST plant in South Australia. When the historical database has a limited set of historical weather profiles, the SMILP achieves 6% to 9% higher profit than the closest benchmark when the DP is applied to novel weather conditions. With a large historical weather data, the performance of SMILP and Heuristic-2 becomes nearly identical because the SMILP can only utilize a limited number of trajectories for optimization without becoming computationally infeasible. In this case, Heuristic-2 emerges a practical alternative, since it provides similar average profit in a reasonable time (saving about 7 hours in computing time).

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑