目前台灣能源依賴進口的比例高達98%，因此提升能源的自主性和多元化變得極為重要。多年來，政府積極推動「開源節流」的政策，其中虛擬電廠(Virtual Power Plant, VPP)就是其中一項核心技術，虛擬電廠通過整合分散式能源來改善用電供應與需求。而智慧電表系統作為VPP的核心基礎設施，不僅能夠即時收集和傳輸用戶用電數據，還可以為用戶和用電管理者提供全面的能源監控和利於分析的數據庫。 此外，隨著智慧電表普及，大量用電數據反映家庭用電動態變化，為預測用電狀況與異常檢測提供了基礎。然而，要如何從如此龐大的數據中提前識別過載、短路等潛在的電氣危險，目前仍是個挑戰。 因此，本研究提出了一種解決方案，結合大型語言模型(Large Language Models, LLM)技術，利用其數據處理和模式辨識能力，來預測即將發生的電氣異常事故。研究中使用智慧電表的歷史用電數據進行模型訓練，將數值數據編碼為文本形式，並設計異常檢測系統，用於分析即時用電使用數據中的潛在危險模式。實驗結果顯示，本系統能夠有效率的預測電氣異常事故。 此研究對於能源管理和電氣安全領域具有重要貢獻，提供了一種可行的早期預警機制。透過結合智慧電表數據與LLM，提升了電網安全性，降低事故發生的風險；家庭用戶則可及時採取預防措施，減少財產損失與人身危害。此外，此方法亦可作為未來智慧電網技術的重要組成部分，推動能源安全與永續發展目標的實現。 ;Taiwan currently relies on imports for 98% of its energy, highlighting the critical need to enhance energy autonomy. Over the years, the government has promoted the "open source and conservation" policy, with the Virtual Power Plant (VPP) emerging as a core technology. VPPs improve the balance between electricity supply and demand by integrating distributed energy resources. As the foundational infrastructure of VPPs, smart meter systems enable real-time collection and transmission of users’ electricity consumption data but also provide energy monitoring and a rich data repository for both users and energy managers. With the widespread adoption of smart meters, large-scale electricity consumption data reflects patterns of household usage and serves as a basis for consumption forecasting and anomaly detection. However, identifying potential electrical hazards such as overloads and short circuits from massive datasets remains a challenge. This study proposes a solution integrating Large Language Models (LLMs), leveraging their capabilities in data processing and pattern recognition to predict electrical anomalies. The model is trained on historical smart meter data, with numerical information encoded into textual form. An anomaly detection system is designed to identify hazard patterns from real-time consumption data. This research contributes significantly to the fields of energy management and electrical safety by offering a viable early warning mechanism. Through the integration of smart meter data and LLMs, grid safety is enhanced, the risk of accidents is reduced, and household users are empowered to take timely preventive actions to mitigate property loss and personal harm. The proposed approach has the potential to become a key component of future smart grid technologies, promoting energy security and the realization of sustainable development goals.
