A New Outlier Detection Method for Anomaly Detection in IoT-Enabled Distribution Networks

Real-time monitoring systems are expected to detect rare observations and anomalies in order to guarantee the stability and safety in IoT-Enabled smart environments. Unsupervised learning methods are mainly utilized for real-time anomaly detection in non-stationary environments. How-ever, the conventional unsupervised methods can only detect anomalies that deviate small portion of data from the majority of normal data. For this problem, a new unsupervised spatial anomaly detection algorithm called MOS is proposed in this paper to be applied in a hierarchical fog computing architecture for more accurate identifying anomalies in large scale distribution networks. Three datasets containing several distinct unexpected events were used for evaluating the proposed method. The efficiency of the proposed detection method was compared to three well-known unsupervised detection methods namely ENOF, DBSCSN and Isolation Forest. The experimental results showed substantial perfor-mance of the MOS algorithm compared to other methods by reaching up to 90% detection accuracy.