Bidirectional LSTM Based Partial Discharge Pattern Analysis for Fault Detection in Medium Voltage Overhead Lines with Covered Conductors

摘要

The continuous increase in computational power has given a new dynamic to the neural network-based solutions. Partial discharge (PD) signal in Medium Volt (MV) overhead lines has very similar characteristics to background noise which makes it very difficult to differentiate between a fault and healthy operations. If a fault doesn't have immediate consequences like a power outage, it remains undetected and can cause partial discharge. This will eventually lead to damage of line and power outage. This paper presents a bidirectional long short term memory (BLSTM) for detection of a fault in MV Overhead line with covered conductors (CC). A real power line dataset acquired by ENET Centre designed meter at VSB is used, which is the largest known available dataset in open source. The data is measured by placing very high sampling frequency meter (40 MHz) at different locations which makes the Data very diverse in terms of noise spectrum and quality of PD and increases the difficulties of classification. The obtained experimental results show that the proposed BLSTM method can learn the chaotic PD pattern with competitive performance.