Case Study: Predictive Maintenance for Remote Photovoltaic Microgrids

Nabil Humphrey, CTO, PrioriAnalytica Paul Stallan, MD/Founder, Apex Energy


This presentation will share some of the challenges involved in maintaining critical, remotely deployed large-scale PV-battery systems. Apex Energy has developed, and currently maintains a portfolio of cost-effective renewable energy deployments for oil and gas infrastructure in remote South Australia and is extending its offering to remote communities for essential services including electricity and water. Critical unscheduled maintenance requirements have been demonstrated to lead to significant lost production, and the basic commercially available monitoring solutions provide insufficient lead-time to remedy the situation before lost production has occurred.
We will provide a review of the literature surrounding the class imbalance problem for machine learning based condition monitoring and discuss a selection of both feature-level and algorithm-level techniques that has been applied to this problem. Many of the techniques discussed will be applicable to approaches to condition monitoring outside of the realm of machine learning. Additionally, we will provide an overview of the hybrid physical and ML model that has proven successful in providing transparency, explainability, and accountability to the deployed predictive maintenance system.


1. An understanding of the challenge involved in conducting predictive maintenance for highly remote installations.
2. A selection of techniques to solve the class imbalance problem.
3. An approach to hybrid physical and ML predictive maintenance.