Department of Computer Science, University of California, Davis, CA, USA.
Department of Civil Engineering & Environmental Engineering, Rice University, Houston, TX, USA.
Risk Anal. 2020 Jan;40(1):134-152. doi: 10.1111/risa.12923. Epub 2017 Oct 30.
Recovery of interdependent infrastructure networks in the presence of catastrophic failure is crucial to the economy and welfare of society. Recently, centralized methods have been developed to address optimal resource allocation in postdisaster recovery scenarios of interdependent infrastructure systems that minimize total cost. In real-world systems, however, multiple independent, possibly noncooperative, utility network controllers are responsible for making recovery decisions, resulting in suboptimal decentralized processes. With the goal of minimizing recovery cost, a best-case decentralized model allows controllers to develop a full recovery plan and negotiate until all parties are satisfied (an equilibrium is reached). Such a model is computationally intensive for planning and negotiating, and time is a crucial resource in postdisaster recovery scenarios. Furthermore, in this work, we prove this best-case decentralized negotiation process could continue indefinitely under certain conditions. Accounting for network controllers' urgency in repairing their system, we propose an ad hoc sequential game-theoretic model of interdependent infrastructure network recovery represented as a discrete time noncooperative game between network controllers that is guaranteed to converge to an equilibrium. We further reduce the computation time needed to find a solution by applying a best-response heuristic and prove bounds on ε-Nash equilibrium, where ε depends on problem inputs. We compare best-case and ad hoc models on an empirical interdependent infrastructure network in the presence of simulated earthquakes to demonstrate the extent of the tradeoff between optimality and computational efficiency. Our method provides a foundation for modeling sociotechnical systems in a way that mirrors restoration processes in practice.
在灾难性故障发生的情况下，恢复相互依存的基础设施网络对于经济和社会福利至关重要。最近，已经开发出集中式方法来解决相互依存的基础设施系统灾后恢复场景中的最优资源分配问题，这些方法可以最小化总成本。然而，在现实系统中，多个独立的、可能是非合作的公用事业网络控制器负责做出恢复决策，导致次优的分散过程。为了最小化恢复成本，最佳分散模型允许控制器制定完整的恢复计划并进行谈判，直到所有各方都满意（达到均衡）。对于规划和谈判来说，这样的模型计算量很大，而且时间是灾后恢复场景中的关键资源。此外，在这项工作中，我们证明在某些条件下，这种最佳分散谈判过程可以无限期地继续。考虑到网络控制器修复其系统的紧迫性，我们提出了一种特定的序贯博弈理论模型，用于表示网络控制器之间的离散时间非合作博弈，该模型保证可以收敛到均衡。我们通过应用最佳响应启发式算法进一步减少了找到解决方案所需的计算时间，并证明了ε-Nash 均衡的界，其中ε取决于问题输入。我们在存在模拟地震的情况下，将最佳情况和特定模型在实证相互依存的基础设施网络上进行了比较，以展示最优性和计算效率之间的权衡程度。我们的方法为建模社会技术系统提供了一个基础，这种方法反映了实践中的恢复过程。
