在電力系統中監測信號會受到雜訊的干擾,使得資訊末端設備所獲取之資料充滿不確定性。在資料校正研究上以往所進行的資料前置處理大多採用模糊基因演算法或共同進化模糊演算法來達成。本研究嘗試以廣義迴歸類神經網路來建立一個線上、即時的資料校正前置處理方法。文中除以模糊基因演算法及共同進化模糊演算法的實驗數據進行比較驗證外,並與倒傳遞類神經網路、徑向基底函數類神經網路進行比較。由誤差校正及曲線平滑的變化情形分析結果發現,本文所建立廣義迴歸類神經網路的資料校正前處理方法,其誤差校正的精確度高、學習速度快,適合線上即時進行資料的校正。實驗結果顯示本文所提出之方法更能有效解決監測時數據的模糊及不確定性。
In SCADA-based power systems, the data gathered from remote terminal units (RTUs) contains uncertain values due to noise interference. Data preprocessing using soft computing approaches, such as fuzzy-GA and co-evolutionary fuzzy, has been used as effective means for reducing the effect caused by noise interference in past research. In this thesis, a generalized regression neural networks (GRNNs) is employed to screen the raw data as a data preprocessing step before running data analysis. Compared with back propagation neural networks (BPNNs) and radial basis function neural networks (RBFNNs), the proposed GRNN has better results in precise fitting curves, data accuracy and high speed training process, which makes it feasible for online application. Study cases have been provided in the study for verifying the proposed methodology. Experimental results show that the proposed approach can deal with the noisy signal and solve the data uncertainty problem effectively.