Understanding the Impact of Flash Floods on Electrical Distribution Systems Using Monte Carlo Techniques and Reinforcement Learning

Flash floods are recognized as a major threat to electrical distribution systems. To enhance the resilience of power supply systems to these natural disasters, it is crucial to evaluate risk assessment for flash floods. Typically, researchers use two-dimensional (2D) surface flow models to evaluate flood risks on power systems. However, these models can provide a description of surface flow propagation but fail to predict flood inundation points crucial in flood modeling. Moreover, these models require high computational cost, making them unsuitable for real-time analysis. This study presents a probabilistic flood model that is easy to develop and can handle high uncertainty related to urban flash floods. Using Monte Carlo techniques, the proposed model predicts flood inundation points in a grid-based environment, taking into account rainfall intensity, soil moisture, and surface roughness. Reinforcement learning is used to track the floodwater flow paths from these inundation points to identify substation main distribution stations at risk of being flooded. The proposed model was applied to a real 23-bus IEEE system and validated in a real urban area. This research aims to help decision-makers and utility operators improve the resilience of electricity supply systems against urban flash floods, as well as address the challenges of data scarcity and time constraints.

Flash floods cause extended and extensive power outages and pose significant challenges to electrical distribution systems, including public safety, landslide risks, submerged substations, and access to flooded areas. In Malaysia, electricity substations, a critical and expensive component of the electrical infrastructure, have been identified as the most flood-prone assets, requiring a clear understanding, identification, and quantification of urban flood risks against these substations. To mitigate these risks, the construction of flood levees at suitable heights at substation entrance doors and the installation of pumps in drainage sumps are common structural measures. However, many uncertainties surround these measures, as the relationship between their characteristics and effectiveness is unclear, limiting the choice of optimal structural measures. Furthermore, lack of historical data on flash flood events, due to practical difficulties in collecting data during emergency situations, complicates researchers' ability to model and predict these events.

This study suggests a novel framework for flood risk assessment of electrical substations, which is based on flood inundation maps and a Weibull distribution of failure probability of electrical substations in Malaysia. Using the proposed approach, the risk of electrical substations being affected by flash floods can be effectively assessed and visualized on a Geographic Information System (GIS) platform. The study focuses on the impact of flash floods on the electrical substations in urban areas in Malaysia and identifies areas with high risk of flooding using spatial analysis.

The proposed approach can be applied to the IEEE 33-bus system using a practical approach presented in Sec. 2.2.1 to prepare the test system. As shown in Fig. 1, a grid with equal spacing, comprising 6 x 24 = 144 cells, placed on the test system to capture the spatial impact of flash floods.

Based on the damage scores, electrical substations were classified into two classes: at-risk and not-at-risk. Substations in high-impact zones are expected to be at risk of flash flooding and marked as submerged substations. On the other hand, substations in flood-free zones, where no flood events occur, are considered not at risk of flash flooding. Visualized from the above Figure that substations 32 and 33 are located in zone z87 identified as a potential flash flood zone with an average flood depth of 0.72m. By applying the proposed failure probability model, the probability of failure for substations 32 and 33 is predicted to be 1, indicating a 100% certainty that these substations will be affected.

This study aimed to address the limitations of the previous flood risk assessment approaches by incorporating new techniques and tools. The proposed approach combines a probabilistic flood model with reinforcement learning to assess the risk of electrical substations being affected by flash floods. The proposed model was applied to a real-world case study and validated in a flood-prone area. The results show that the proposed approach is effective in identifying areas with high risk of flash flooding and assessing the likelihood of electrical substations being affected.