Recurrent neural network with density-based clustering for group pattern detection in energy systems

Abstract

This research explores a new direction in power system technology and develops a new framework for pattern group discovery from large power system data. The efficient combination between the recurrent neural network and the density-based clustering enables to find the group patterns in the power system. The power system data is first collected in multiple time series data and trained by the recurrent neural network to find simple patterns. The simple patterns are then studied, and analyzed with the density-based clustering algorithm to identify the group of patterns. The solution was analyzed in two case studies (pattern discovery and outlier detection) specifically for power systems. The results show the advantages of the proposed framework and a clear superiority compared to state-of-the-art approaches, where the average correlation in group pattern detection is 90% and in group outlier detection more than 80% of both true-positive and true-negative rates.