NIA SSEN 0065 Storm AI

Expected Benefits

The main benefit to customers will be in relation to increased accuracy in the Estimated Time of Restoration (ETR). Benefits to Distribution Network Operators (DNOs) will be due to greater efficiency during storms and weather events, resulting in an estimated 2% saving in Guaranteed Standards and Compensation Payments. These are estimated at £2.1m for the next five years based on assumed similar weather patterns.

Progress

The project looked into the use of Machine Learning techniques – notably those generally grouped under the umbrella of computer vision to classify customer-submitted images of storm-damaged network assets. Two main avenues of classification were considered: paid-for machine-learning-as-a-service (MLaaS) solutions, such as Google’s Vertex AI and Amazon’s AWS Rekognition, and an in-house solution constructed within the Python programming language. Through the build of the Python package, various other methods were trialled, and the first iteration of the classifier used a custom-built, 10-layer convolutional neural network (CNN) model.

The initial performance of the model on the small set of labelled data available to the model shows promise, with all target sets being predicted with accuracy scores in excess of random chance.

Storm AI is planned to be integrated as a feature of SSEN’s Power Track tool, a customer facing outage notification and reporting tool that allows consumers to see reported outages. This will result in a Business as Usual (BaU) deployment of Storm AI.