Outlier detection based on improved SOM and its application in power system

This paper presents a comprehensive outlier detection algorithm based on the Self-Organizing Map (SOM) neural network algorithm, K-means and a first proposed One-Dimensional Density-Based Clustering of Application with Noise (ODDBCAN) algorithm. The ODDBCAN algorithm is a simplification and improvement of DBSCAN. It is designed to detect obvious noise and guarantee the validity of the following processes. A two-stage approach combining SOM and K-means is introduced in order to reduce the computational cost. Therefore the comprehensive algorithm has high accuracy and considerable computational efficiency. It can be applied to data cleansing and knowledge discovery. The algorithm is universal and an example of electric energy data is taken to prove its applicability to power system.