工業技術研究院

技術簡介

本技術利用風機 Scada 多參數資料，透過 CNN＋LSTM 建立多分類故障診斷與劣化預測模型，能提前辨識風機異常並降低非計畫性停機。使用者僅需載入數據即可完成模型訓練與預測，已於實際風場驗證可有效縮短排查時間並降低停機損失。

Abstract

This technology utilizes multi-parameter wind turbine SCADA data and applies a CNN ＋ LSTM deep learning framework to build a multi-class fault diagnosis and degradation prediction model. It can identify anomalies at an early stage and reduce unplanned downtime. Users only need to upload data to complete model training and prediction. The system has been validated in real wind farms, demonstrating its ability to shorten troubleshooting time and reduce downtime-related losses.

技術規格

系統以 Scada 參數作為輸入，採深度學習模型自動萃取特徵並輸出故障分類與劣化曲線。具備資料上傳、模型訓練與視覺化介面，可部署於本地或雲端，滿足風場精準預警需求。

Technical Specification

The system uses SCADA parameters as inputs and applies deep learning models to automatically extract features and output fault classifications and degradation trends. It provides data upload, model training, and visualization interfaces, and can be deployed either on-premises or in the cloud, meeting the precise early-warning needs of wind farm operations.

技術特色

透過模組可發現傳統方法無法辨識的細微異常，且支援各場域以自身數據重新訓練，提高維運自主性並具高度擴充性。

應用範圍

電廠機組、大型馬達、泵浦、工業設備等

接受技術者具備基礎建議（設備）

1.Windows7以上64位元作業系統 2.mosquitto-2.0.11-install-windows-x64 3.如需 GPU 加速，請額外安裝 CUDA Toolkit 11.x 及 cuDNN 8.x

接受技術者具備基礎建議（專業）

需整合 AI 與資料科學、風機工程、資料工程與軟體開發等人才

聯絡資訊