CircuitSee is project to implement an approach to provide continuous fault-level monitoring based on using machine-learning technology which will allow us to install less equipment on our sites. It could save customers more than £1 million.
If too much power goes through equipment in our substations, they can reach what we call a ‘fault level’ and stop working efficiently. To make sure this doesn’t happen, we routinely and often monitor our substations with special software to make sure they don’t reach more than reach any more 95% of the equipment’s rating limits.
At present we use of a software tool called PowerFactory to calculate those fault levels. The number produced by the software helps us ‘check in’ on substations and take corrective action if we see that fault levels are at 95% or more.
We install live monitoring through PowerFactory where we know fault level is getting close to between 95% and 100% of our switchgear rating, and will take action to ensure reliability and safety if levels rise too high.
Installing fault level monitoring, however, can be costly and time-intensive. CircuitSee is testing a new approach to the improve accuracy and efficiency – and reduce the cost – of our current fault level monitoring methods.
The new method proposed in the project is to implement continuous on-line fault level monitoring with three key components.
The new software solution provider is Reactive Technologies (RTL). The solution builds up a picture of the fault level across an entire radial network beneath our substations using machine learning techniques. Due to the active measurement approach, the high frequency of measurements allows for high accuracy and for fault level to be correlated between different substations as well. This means we can deploy the software much wider than previously. The technology has been developed in a laboratory setting so far and is still subject to a patent application. The next stage of development is to trial for real on our network.
The approach used in this project is the first one to try to correlate passive measurement with active measurements. Generally speaking, passive take more time than active monitoring. This time depends on the number of switching activities in the network. Active measurements overcome this but are more expensive and bulky. By using machine learning algorithms to correlate the two methods, we’re aiming to have the best of both worlds.
The approach used in this project has not been previously been done by a DNO is Great Britain. As part of the project we will validate first the theoretical standing of the solution against Powerfactory. In phase 2 we will implement the solution on a study area to learn more about the costs of implementation and the accuracy of result in a real world application.