Machine learning approach for recognition of partial discharge patterns for Gas insulted switchgear (GIS)

Gas insulted switchgear (GIS) is one of the key equipment in power system, which plays a crucial role in the safety and stability operation of the whole power grid. Most of the operation damages of GIS are caused by electrical insulation failure, and a key indicator of such insulation faults is the partial discharge (PD).
However, differentiating defect types through PD pattern recognition is one of the most difficult challenges and has restricted the large-scale industrial application of PD based condition monitoring. To overcome this challenge, different machine learning approaches of pattern recognition have been applied like decision tree, SVM etc. But these traditional machine learning methods have met bottlenecks in their development which restrict further improvements in accuracy.

In recent years, the deep neural network (DNN) which has stacked layers of neuronal units that learn the hierarchical representation of the data has demonstrated dramatic success in speech feature extraction, image classification and natural language processing, scaling from small to large dataset.

This project will investigate the feasibility of deep learning model to correctly identify different sources of defects within the GIS.

• Literature survey on partial discharge monitoring, partial discharge diagnosis, machine learning techniques on PD signal pattern recognition.
• DL Model development and theoretical formulation
• Practical application and demonstration of methodology
• Model evaluation and testing on industry testbed.

Other Comments: 

Essential Skills and Knowledge: Signal processing, machine learning and data analysis background. Coding experience (e.g. MATLAB)

Supervisor name: 
Professor David Flynn
Supervisor and Deputy email addresses: 
d.flynn@hw.ac.uk
Project location: 
Remote working
Deputy name: 
Mr Wenshuo Tang