基於符號化時間序列資料探勘架構之電力消耗負載分析

A Symbolic Time-series Data Mining Framework for Analyzing Load Profiles of Electricity Consumption

能源在永續發展工業已被視為重要的管理資產，因此，如何減少能源消耗並有效率地追蹤及管理能源為重要的挑戰。本研究基於電力負載追蹤電力消耗狀況提出符號化時間序列電力資料探勘架構，首先，研究應用分段聚合近似法（piecewise aggregate approximation, PAA）進行時間序列降維處理，接著採用符號聚合近似演算法（symbolic aggregate approximation, SAX）將降維後序列進行符號化，並改良SAX演算法的時間序列下限制（lower-bounding）距離衡量計算公式。研究以鋼鐵鍛造公司的大型退火爐為例進行方法驗證，實驗結果顯示採用PAA法較傳統的固定端點取法較能預測機器狀況；另一實驗結果顯示改良SAX之下限制距離公式能更準確地計算負載曲線之間的相似度。本研究所提出之架構與方法將有助於工廠進行後續正異常電力樣式預測。

Electricity is critical for industrial and economic advancement, as well as a driving force for sustainable development. In turn, reducing energy consumption for sustainability and both tracking and managing energy efficiently have become critical challenges. In this research, we analyzed electricity consumption from the perspective of load profiling, which charts variations in electrical load during a specified period in order to track energy consumption of an annealing furnace in a co-operating steel forging plant. We made a preliminary proposal to use a symbolic time-series data mining framework for electricity consumption analysis. First, we adopted a piecewise aggregate approximation (PAA) approach to perform dimension reduction. Then, we refined the distance measure of the symbolic aggregate approximation (SAX) algorithm. SAX is a symbolic representation of time-series for dimensionality reduction and indexing with a lower-bounding distance measure. Our experimental results showed that the dimension reduction method known as PAA can better detect the state of the annealing furnace compared to the fixed feature point (FFP) method. In addition, the refined lower-bounding distance measure proved to be better than the traditional measure for calculating the similarity between energy load profiles. The results can help the plant conduct further normal and abnormal electricity pattern detection.