基於長短期記憶學習模型之智慧用電設備危險及老化偵測

AIoT Equipment Hazard and Aging Detection based on Long Short-Term Learning Models

用電設備是工廠、辦公及日常必需品，一旦疏於注意，將造成火災或危險事故，然而，市售的智慧插座僅能透過功率閥值確認用電設備是否有過負載，無法即時檢測危險及老化異常，例如:馬達故障／阻塞、元件或電路短路／破損／老化／生鏽、插頭鬆脫／接觸不佳、插座過熱／濕氣過重等，若使用者疏於注意，將有可能造成嚴重的火災事故。因此，我們研發實作一套智慧用電設備檢測系統，透過量測用電設備的電力資訊，包含:電壓值、電流值、功率值、功率因素，再結合AI機器學習之異常辨識技術，動態檢測用電設備的異常狀況，並實作設計以下四項功能元件: 1.安裝於插座之硬體裝置，可即時檢測用電設備運作期間之異常數值，2.具備即時異常監控與提醒，異常紀錄統計，用電設備討論平台等功能之手機應用程式3.基於長短期記憶學習模型辨識各種用電設備異常狀況。4.建立跨平台用電設備分析網頁平台，供管理員方便管理大範圍的所有用電設備。經實驗測試結果，本系統的AI機器學習判讀準確率高達93.98%~99.9%。

The electrical appliances used in factories, offices, and daily necessities are essential for our daily lives. Negligence in monitoring these appliances can lead to fire hazards or dangerous accidents. However, available smart plugs can only confirm whether electrical devices are overloaded through power threshold values. They cannot detect potential dangers or aging abnormalities in real-time, such as motor malfunctions/blockages, component or circuit short circuits/damage/aging/rust, loose connections or poor contacts of plugs, or overheating/moisture in sockets. If users neglect these issues, serious fire accidents may occur. Therefore, we have developed and implemented a smart electrical device detection system. It measures the electrical information of electrical devices, including voltage, current, power, and power factor. Combined with AI machine learning anomaly detection technology, it dynamically detects abnormalities in electrical appliances. We have designed and implemented the following four functional components: 1. The plug of an electrical device inserted into the hardware device can detect abnormal values during the operation of the electrical device. 2. A mobile application with features such as real-time abnormal monitoring and alerts, abnormal record statistics, and discussion platform for electrical devices. 3. Identify various abnormalities of electrical appliances based on deep learning neural network technology. 4. Establishing a cross-platform electrical device analysis web platform for administrators to conveniently manage all electrical devices over a wide range. According to the experimental test results, the AI machine learning interpretation accuracy of this system reaches as high as 93.98% to 99.9%.