Cloud-Based Machine Learning Models to Determine Potential Misconnection and Shock Hazard on Smart Meter Installations

Installations of advanced metering infrastructure (AMS) continue to grow among power utilities as the benefits of smart meter communications with grid operators become increasingly evident. However, determining the quality of a smart meter installation is often difficult without physical site presence and local electrical testing. Misconnections could result in lost utility revenue, while earth faults could create shock hazards, and TN-grid mis-connections could potentially result in fatalities.

A European power utility faced a regulatory mandate to install AMS across its customer base. In the process of installing these smart meters, it realized it lacked processes for discovering the quality of meter installations. Arundo worked with this utility to deploy machine learning models through Arundo Composer and Arundo Fabric to monitor the quality of its installations based on pre- and post-AMS consumption, GIS, and electrical data. In addition, using Composer, the company was able to easily deploy 3rd party consumption analysis predictions built with TensorFlow.

RESULTS

The Arundo software identified that 1% of initial wave of new smart meters were not installed correctly—a number that was substantially higher than expected. This included a number of single and double-earth faults. In addition, Arundo identified several severe TN-grid misconnections. These represented a significant shock hazard that could have resulted in potential fatalities. Arundo’s software identified fault locations at specific properties based on appliance usage patterns, which could potentially improve the management and efficiency of field engineer interventions. Across the total base of new AMS installation, these findings represented up to $8.5 million in potential lost value from installation errors, in addition to the significant safety hazards that the software discovered.