Online false data detection and lost packet forecasting system using time series neural networks for IEC 61850 sampled measured values

Abstract

Migrating to a smart grid requires a paradigm shift in the implementation of power system applications. With the advent of IEC 61850, contemporary Substation Automation Systems (SAS) are utilizing electronic instrument transformers and merging units to transmit current and voltage measurements over Ethernet as Sampled Measured Values (SMV). However, if a substation's network resources are not properly managed, the high transmission rate of SMV would make them prone to packet loss. Also, the strict 4ms time constraint imposed on SMVs makes encrypting these messages nearly impossible. As such, this paper presents an online device level fake data detection system for detecting fake SMV messages without violating the 4ms time constraint set forth by IEC 61850. In order to ensure a reliable SAS operation, this paper also presents a coupled neural network — time series method for forecasting lost SMV packets. The proposed algorithm was implemented in a system composed of merging units and intelligent electronic devices developed for this purpose. Real-time experimental results of the proposed algorithms over a real IEC 61850 network showed excellent results in terms of detecting fake messages and increasing the robustness of protection schemes by accurately forecasting dropped SMV packets.