Session: 02-03-03: Optimization

Paper Number: 145487

145487 - Disruption management of interdependent power networks using a data-driven co-design approach for enhanced system resilience 

The increase in complexity of a system is likely to escalate its probability of suffering from an external disturbance. Critical infrastructure systems such as power grids are no exception, where the lack of consideration of failure can lead to catastrophic results. Therefore, it is important to design and maintain these systems to have high resilience against disruptions. A typical system undergoes three stages of operation: the normal stage, the degraded stage after disruption, and the recovery stage, which leads back to the normal operation. One approach to enhancing the system resilience is to create a design that can absorb more damage from the failure, i.e., it has less performance degradation from an attack. Having an efficient strategy of restoration is another approach to improving resilience. In this research, the focus is to merge these two approaches within a single optimization framework. The integrated framework can benefit from its interactions by considering the post-disruption management scheme in the early design steps, which can enhance its effectiveness toward full recovery.

For the optimization framework, an appropriate performance measure needs to be employed. The main objective of a power grid is to provide customers with the demanded energy. Hence, the amount of lost demand is a good measure to represent the performance of a grid. However, a design or recovery should always consider the cost, cost from the construction during design, or operational cost during the recovery. This prevents the network design from being extravagant, e.g., having redundant generators and transmission lines, as well as using excessive amounts of resources during restoration. The difficulty arises when calculating the performance for each design. Normally, the performance can be evaluated after a power flow analysis, where the optimal generation and transmission are calculated. However, this process needs to be executed multiple times for each possible case of disruption to evaluate the performance of a network. This is to prevent the effect of disruptive events being biased towards a particular scenario. Depending on the size of the network, this can lead to an impractical to nearly impossible amount of computational resources. Hence, an accurate data-driven approximation can be useful. For this purpose, a sophisticated means of surrogate modeling for graph structure data is investigated along with a suitable measure that demonstrates the performance of a power system network.

This study focuses on developing an optimization done in a nested framework, where the main loop determines the design of the power system and the nested loop determines its recovery profile about a set of predetermined disruption scenarios. The performance during the optimization is estimated using a surrogate trained by a large number of synthesized power grids, where the performances of each sample are pre-determined. An IEEE benchmark is used to compare the resulting design and prove the efficacy of the framework.

Presenting Author: In-Bum Chung University of Illinois Urbana-Champaign

Presenting Author Biography: In-Bum Chung is currently a PhD student at University of Illinois at Urbana-Champaign.

Authors:

In-Bum Chung University of Illinois Urbana-Champaign
Yi Luo University of Illinois Urbana-Champaign
Pingfeng Wang University of Illinois Urbana-Champaign