Maintenance Detection Systems
The Problem
Grid-connected solar panels are an important part of the fight against climate change, and a near-exponential increase in deployments has seen them go from niche application to mainstream power generation technology.
The capital cost of solar panels is repaid by the electricity they generate during operation, so maintaining peak efficiency and minimising downtime is key for making the economics work.
Can Machine Learning be used to identify panels in need of cleaning or maintenance?
The Approach
Plant performance and ambient condition data were obtained from two solar plants in India, and the datasets were extensively cleaned and feature-engineered to allow relationships between different panels to be identified.
Related strategies for identifying panels in need of maintenance (a panel performs differently to the peer group) and for identifying panels in need of cleaning (panels collectively perform lower than predicted based on history) were developed, with machine learning used to predict panel output power under a variety of conditions.
The Solution
Despite some limitations in the dataset, we were able to build a series of models to predict the power output of panel arrays under different conditions.
These models were able to successfully identify when panels were in need of maintenance, and when the array as a collective was in need of cleaning.
You can view the full project, including source code, models and reports, on our GitHub page.
