Skip to main content
QUICK REVIEW

[论文解读] Human-in-the-loop MGA to generate energy system design options matching stakeholder needs

Francesco Lombardi, Stefan Pfenninger|arXiv (Cornell University)|Jul 19, 2024
Real-time simulation and control systems被引用 4
一句话总结

本文提出了一种新颖的人机协同(HITL)建模以生成替代方案(MGA)框架,通过迭代优化MGA搜索空间,将利益相关者偏好整合到能源系统设计优化中。通过与初步设计空间的初次互动获取偏好,并将其解码为强化目标,该方法生成了一个‘经人类训练’的设计空间,显著提升了与利益相关者优先事项的一致性,并增强了共识潜力。在葡萄牙能源系统的受控实验中,具有共识相关性的设计方案占比从1%提升至18%。

ABSTRACT

The common use of cost minimisation to support energy system design decisions hides from view many economically comparable design options that stakeholders may prefer. Modelling to generate alternatives (MGA) is increasingly popular as a way to go beyond least-cost designs, providing stakeholders with diverse portfolios to appraise. However, generating all the feasible designs is not computationally viable; modellers must choose what design features to generate diversity around, despite not knowing which trade-offs matter the most in practice. Therefore, MGA alone cannot ensure the generation of design options that match stakeholder needs. To address this shortcoming, we propose a human-in-the-loop (HITL) approach that automatically integrates stakeholder preferences into MGA. We elicit preferences by letting stakeholders interact with a tentative MGA design space. Hence, we decode those preferences to feed them back to the MGA algorithm and perform a guided search. This search produces a human-trained design space with more designs that mirror the elicited preferences. A synthetic experiment for the Portuguese energy system shows that HITL-MGA may facilitate consensus formation, promising to accelerate technically and socially feasible energy transition decisions.

研究动机与目标

  • 解决传统成本优化在能源系统建模中的局限性,即忽视了具有实际优势的可行非最优设计方案。
  • 克服在大规模、高分辨率模型中生成所有可行能源系统设计方案的计算不可行性挑战。
  • 将利益相关者知识整合到MGA工作流中,引导搜索多样性聚焦于现实决策中最相关的特征。
  • 开发并验证一种方法,使建模者能够通过自动化、偏好驱动的设计空间优化,共同生成技术上和社交上可行的能源转型路径。
  • 展示HITL-MGA在提升共识形成方面的有效性,通过增加与多个潜在冲突的利益相关者偏好相一致的设计方案占比。

提出的方法

  • 通过与初步生成的、临时的MGA设计空间进行初次互动,获取利益相关者偏好。
  • 通过识别得分最高的设计方案中具有统计显著性的技术特征,将利益相关者偏好解码为定量强化目标。
  • 使用SPORES MGA算法的改进版本执行引导式搜索,根据解码后的偏好调整强化系数(b和c),以优先考虑期望的特征。
  • 采用基于阈值的特征检测方法,识别在统计上具有显著性、因而与利益相关者偏好相关的技术特征。
  • 使用更新后的强化目标重新运行MGA算法,生成反映利益相关者输入的新一代‘人类训练’设计空间。
  • 通过多准则决策分析(MCDA)评估人类训练设计空间的性能,以衡量其与利益相关者偏好的契合度及共识潜力。
Figure 1: Conceptual representation of the proposed HITL-MGA workflow and of the controlled experiment we carry out in this study to test the workflow. The theoretical framework we propose envisions that we collect the high-level system design preferences arising from a first iteration of an MGA des
Figure 1: Conceptual representation of the proposed HITL-MGA workflow and of the controlled experiment we carry out in this study to test the workflow. The theoretical framework we propose envisions that we collect the high-level system design preferences arising from a first iteration of an MGA des

实验结果

研究问题

  • RQ1利益相关者偏好能否在自动化MGA工作流中被有效获取并编码,以指导能源系统设计方案的生成?
  • RQ2通过人机协同方法整合利益相关者偏好,与传统MGA相比,能在多大程度上提升生成设计方案与利益相关者需求的一致性?
  • RQ3HITL-MGA方法如何影响具有冲突优先事项的利益相关者之间的共识形成潜力?
  • RQ4在MGA框架内解码利益相关者偏好并将其转化为强化目标的过程中,存在哪些计算与方法论上的权衡?
  • RQ5HITL-MGA方法能否推广至真实世界的利益相关者参与,而不仅限于受控的合成实验?

主要发现

  • HITL-MGA方法将位于最高MCDA得分25%范围内的设计方案占比——表明具有高共识潜力——从传统MGA设计空间的1%提升至人类训练设计空间的18%。
  • 该方法成功改善了11项高层次利益相关者偏好中的10项,包括减少陆上风电的区域集中度以及限制特定技术的总体部署量。
  • 唯一未改善的偏好是关于低总体基础设施部署,这与模型层面的冲突有关,原因在于基于强化的解码方法倾向于支持部署规模的提升。
  • 敏感性分析表明,特征解码中阈值的选择会影响强化目标的分布;更严格的阈值会使目标分布于更多特征,可能稀释对特定偏好的影响。
  • 人类训练设计空间使原本在原始MGA空间中完全缺失的妥协型解决方案(如有限的陆上风电部署且区域集中度低)得以被发现。
  • 本研究证明,HITL-MGA能够有效与利益相关者共同生成模型结果,显著提升能源转型路径的相关性与社会可行性。
Figure 2: Stylised representation of the SPORES algorithm, in its standard ( a ) and HITL ( b ) formulations. Starting from the cost-optimal solution, the algorithm iteratively looks for additional feasible and only marginally costlier system designs in parallel batches. The standard formulation ( a
Figure 2: Stylised representation of the SPORES algorithm, in its standard ( a ) and HITL ( b ) formulations. Starting from the cost-optimal solution, the algorithm iteratively looks for additional feasible and only marginally costlier system designs in parallel batches. The standard formulation ( a

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。