Adaptive Fuzzy Clustering Based Anomaly Data Detection For Energy System In Steel Industry

Due to the complex structure, the multiple energy media, and a large number of production equipments of energy system in a steel plant, the acquired industrial data is very significant to the safety and reliability of the energy system. However, owing to the instability and vulnerability of supervisory control and data acquisition (SCADA), the data anomaly usually exists in practice, which usually generate great influences on the effectiveness of the established model and largely impact the decision making process of the energy scheduling workers. In current practice, the identification of the data anomaly relies mainly on the personal experience of the scheduling workers, and such task is rather intricate and time-consuming.In this study, considering the data feature of energy system in steel industry, the energy data anomalies are summarized as1) the trend anomaly for the pseudo-periodic data and2) the deviants for the generic data. As for the trend anomaly, a dynamic time warping (DTW) based method that transforms the similarity of the data sequences with unequal length into the Euclidean distance metric is proposed combining with an adaptive fuzzy C means (AFCM). As for the deviants detection, a k-nearest neighbor AFCM algorithm (KNN-AFCM) is designed here for the local anomaly detection of the relatively stable data. In this algorithm the neighborhood effect of points and the sample weight coefficient are introduced into the objective function.To verify the effectiveness of the proposed method, the real-world energy data are chosen from a steel plant to conduct the validation experiments. First, the proposed method in this paper is compared with the other two methods, and then the impacts of some parameters are discussed. As for the deviants detection, some experiments are conducted to verify the effect and necessity of the sample weight coefficient. The experimental results indicate that the proposed method exhibits a higher detection precision than the others for data anomaly in steel industry.