ABSTRACT. With an energy import dependency approaching 98%, Taiwan faces a critical imperative to enhance its energy autonomy and diversify its energy portfolio. Governmental policies have consequently focused on promoting both energy development and conservation. Central to these strategies are Virtual Power Plants (VPPs), which aggregate Distributed Energy Resources (DERs) to optimize supply-demand balancing. The foundational infrastructure for VPPs is the smart meter, which enables real-time data acquisition for granular energy monitoring and analysis by both consumers and utility administrators. The widespread deployment of smart meters has generated vast datasets reflecting household consumption patterns, which are invaluable for applications such as load forecasting and anomaly detection. Nevertheless, the real-time identification of critical electrical hazards, including overloads and short circuits, from these high-volume data streams presents a significant computational challenge. To address this gap, this study proposes a novel anomaly prediction system that integrates Large Language Models (LLMs), leveraging their sophisticated capabilities in complex pattern recognition to improve the efficacy of electrical safety monitoring. The proposed methodology involves transforming historical numerical time-series data from smart meters into a textual sequence format suitable for LLM training. A detection framework was designed to analyze real-time consumption behaviors using the trained model. Experimental validation demonstrates that the proposed LLM-based system achieves high accuracy in predicting electrical anomalies, substantiating its significant potential for practical application. This system can function as a proactive early warning tool, enhancing the safety of both grid infrastructure and residential electricity use. By mitigating the risk of electrical disasters and associated property losses, this research contributes to the advancement of smart grid technologies and supports Taiwan’s strategic objective of achieving net-zero emissions by 2050.

ABSTRACT. Building fires cause serious injuries, deaths, and major property losses every year, making them a constant threat to safety. Many casualties occur because people choose familiar but longer escape routes instead of the nearest safe exit. Current evacuation aids remain limited to static 2D floor plans, which provide little help in complex indoor spaces and give almost no guidance to people unfamiliar with the building layout. Advances in Augmented Reality (AR) now offer a way to close this gap by projecting clear, step-by-step virtual instructions directly onto the physical environment, making evacuation guidance more intuitive and interactive. At the same time, Building Information Modeling (BIM) has proven highly effective for storing and managing detailed building and safety data. Integrating AR with BIM creates the potential for personalized, real-time evacuation support that goes far beyond the limits of conventional tools. This study presents an AR-based fire evacuation navigation system that uses BIM information to strengthen situational awareness and decision-making during emergencies. The system builds AR spatial models through pre scanning, tracks user positions in real time, and calculates the shortest escape routes. It also links BIM fire safety equipment to a database so users can view critical information instantly. A case study in a university teaching building verified the system’s feasibility and showed that it reduced evacuation time while improving overall fire safety preparedness.

ABSTRACT. The comprehensive issue on the Keparakan Indonesia riverside is the dilemma faced by stakeholders regarding the best solution for the housing. On one hand, there is a perception that the top-down solution is to demolish the periphery housing to create broad access for volcanic flood mitigation while maintaining the interior thermal comfort resulting from solar heat gain. On the other hand, the citizens aspire to keep their previous housing to maintain their craftsmanship home industry, despite the risk of ecological disasters. This dilemma poses a potential clash between stakeholders if not solved properly. The proposed Artificial Intelligence (AI) definition on socio-ecological system resilience aims to address this issue by considering the physical, social and cultural aspects, leveraging the opportunity to hold the development and ensure the safety of the community. Experiments using AI tools such as LLM and computer vision for riverside housing; flood-safe; disaster resilience; tropical sustainability. the research aims to utilize AI tools to analyse riverside area both as structural buffer for disaster mitigation and to enhance the livelihood, building the economic development of the craftsmanship industry. A neighbourhood model then generated based on the result of content analysis applying guidelines of pitched roof; pro-riverside building orientation; vegetations shade; flood-buffer pedestrian lane and resulting on 28% reduction of solar heat gain. The result promotes energy efficiency by reducing cooling load in the housings interior while providing volcanic flood mitigation and socio-ecological resilience due to craftmanship enhancement. Furthermore, the study lists the limitations and future research.

Keywords: Artificial Intelligence; Housing morphology; energy efficiency; riverside housing.