当前位置: X-MOL 学术IEEE Trans. Sustain. Energy › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Deep Learning to Optimize: Security-Constrained Unit Commitment With Uncertain Wind Power Generation and BESSs
IEEE Transactions on Sustainable Energy ( IF 8.8 ) Pub Date : 2021-08-27 , DOI: 10.1109/tste.2021.3107848
Tong Wu , Ying Jun Zhang , Shuoyao Wang

This paper proposes a new model of scenario-based security-constrained unit commitment (SCUC) with BESSs. By formulating such a model as a mixed-integer programming (MIP) problem, we can obtain the optimal control strategy of units and BESSs to reduce the operating cost. To solve this MIP with the proposed model, we propose a new learning-based approach to tackle the SCUC problem. The proposed convolutional neural network (CNN)-based SCUC algorithm (CNN-SCUC) has two main stages. First, CNN-SCUC trains a CNN to obtain solutions to the binary variables corresponding to unit commitment decisions. Then, the continuous variables corresponding to unit power outputs are solved by a small-scale convex optimization problem. In contrast to existing work, CNN-SCUC eliminates the need of explicitly considering the scenario-based security constraints in the optimization problem, and thus greatly reduces the computational complexity. The average gap to the optimal solution is as small as 0.0267%. The algorithm is also scalable in the sense that the computational time is reduced from about 1236.32 seconds to 0.8379 seconds in a 10-unit and 200-scenario system. Besides, the computation time remains almost constant when the number of scenarios increases. Case studies show that compared with the traditional scenario-based SCUC model, more than 4.70% operating cost reduction is achieved by incorporating BESSs in the system.

中文翻译:

深度学习优化:风力发电和 BESS 不确定的安全约束单位承诺

本文提出了一种具有 BESS 的基于场景的安全约束单元承诺 (SCUC) 的新模型。通过制定混合整数规划(MIP)问题等模型,我们可以获得机组和BESS的最优控制策略,以降低运行成本。为了使用所提出的模型解决此 MIP,我们提出了一种新的基于学习的方法来解决 SCUC 问题。提出的基于卷积神经网络 (CNN) 的 SCUC 算法 (CNN-SCUC) 有两个主要阶段。首先,CNN-SCUC 训练 CNN 以获得对应于单元承诺决策的二元变量的解。然后,通过一个小规模的凸优化问题求解对应于单位功率输出的连续变量。与现有的工作相比,CNN-SCUC 消除了在优化问题中明确考虑基于场景的安全约束的需要,从而大大降低了计算复杂度。与最佳解决方案的平均差距小至 0.0267%。该算法在计算时间从大约 1236.32 秒减少到 0.8379 秒的意义上也是可扩展的,在 10 个单元和 200 个场景的系统中。此外,当场景数量增加时,计算时间几乎保持不变。案例研究表明,与传统的基于场景的 SCUC 模型相比,在系统中加入 BESS 实现了 4.70% 以上的运营成本降低。该算法在计算时间从大约 1236.32 秒减少到 0.8379 秒的意义上也是可扩展的,在 10 个单元和 200 个场景的系统中。此外,当场景数量增加时,计算时间几乎保持不变。案例研究表明,与传统的基于场景的 SCUC 模型相比,在系统中加入 BESS 实现了 4.70% 以上的运营成本降低。该算法在计算时间从大约 1236.32 秒减少到 0.8379 秒的意义上也是可扩展的,在 10 个单元和 200 个场景的系统中。此外,当场景数量增加时,计算时间几乎保持不变。案例研究表明,与传统的基于场景的 SCUC 模型相比,在系统中加入 BESS 实现了 4.70% 以上的运营成本降低。
更新日期:2021-08-27
down
wechat
bug