Historically fault location techniques have predominantly been applied to any incidents on the network that generates fault current, both intermittently occurring faults or when the fault is permanent. This applies equally to the low voltage network and the high voltage network. The primary driver has been to protect the network’s infrastructure from serious damage that can result in hazardous and unsafe conditions as a result of the continued flow of fault current. This is achieved by protection systems that interrupt fault current in as short a time as possible to limit any damage and to minimise the duration of interruptions to customers’ supplies. This is an inherently reactive approach.
In the interests of improving customer service it is beneficial to try to avoid such unplanned interruptions by anticipating when a circuit’s performance is rapidly degrading.
Ideally we can then anticipate when a fault will occur and arrange for a live line work or a planned interruption to repair a circuit before it becomes either intermittent or permanent. This will provide both an improvement to customer service and, as a result of replacing reactive with proactive interventions should also improve operational planning and efficiency.
Objectives
The project will:
- Identify suitable existing data sets and data analysis algorithms and techniques which could be used to provide fault anticipation functionality using operational and other datasets available within Northern Powergrid and/or other DNOs or external sources. This may include those related to previous LCNF and current NIA projects, e.g. Customer Led Network Revolution and Smart Data).
- Audit the data and monitoring systems deployed and under development at Northern Powergrid in order to support the requirements analysis and specification activities for fault anticipation. This will also provide knowledge and understanding of practical ways to access data in real-time for fault anticipation.
- Make recommendations for specifications for and approaches to the capture of suitable data for fault anticipation and interpretation for any network.
- Research and develop holisitc, multivariable data analysis algorithms that can interpret signals and their interaction and identify complex degradation modes in advance of failures, in order to predict faults and enable network intervention before outages can impact customers.
- Prototype a fault anticipation decision support system for operational engineers based on the algortihms and techniques identified above.
- Report on the findings and learning from the project to other DNOs and interested parties..
Learnings
Outcomes
This project was utilising the resources available from the joint Strathclyde and Imperial College Engineering and Physical Sciences Research Council (ESPRC) Centre for Doctoral Training in Future Power Networks and Smart Grids. Following initial consultation with Strathclyde University it was agreed that analysis of existing and readily available recorded electrical network parameters, preferably recorded in real-time at a high sampling rate, would be a good starting point. It was therefore agreed that:-
· Northern Powergrid would determine what existing electrical network data sources were readily available for initial analysis and
· Strathclyde University would survey available analysis techniques.
The conclusion is that currently available standard datasets will not support the initial premise of the project. However further work may be able to address the issues identified and the techniques trialled could make a contribution to faulkt management and the prediction of network faults with further development of the techniques and data quality.
A range of analytical techniques, including machine learning, have been applied to existing network data, data developed in the Customer Led Network Revolution project, ENW's C2C project data and automatic pole recloser data from SPEN. Datasets are complex and can be incomplete and difficult to interpret. Machine learning looks to have some considerable potential in this area given further development.
Lessons Learnt
The main limitation to this project has been the lack of suitable data sources to analyse, both in quality and completeness. Although the trip coil analysis work was provided with good data sources existing and readily available data sources from the electrical network were limited. For any future work in this area to predict pre-fault activity on the electrical network the means to record real-time data from the network at a suitable sampling rate needs to be developed and deployed prior to consideration of carrying out any analysis of the data. In order to limit the amount of data generated this should only be generated by exception when an out of the ordinary event is detected on the network. It should be noted that Northern Powergrid’s Foresight project has now developed this capability for the low voltage network although its data was not available to this project.
For future projects of this type it is recommended that the design of the datasets to be used by the machine learning algorithms is carefully considered from the onset as the completeness and quality of these is critical to being able to successfully implement machine learning processes successfully.