This project was realized for the Machine Learning course, held at TU Delft, A.Y. 2022/2023
Team members
In this report is described the framework we followed to develop an ML model that uses Photovoltaic system power output to detect malfunctions. Using binary classification, we have separated faulty system conditions from healthy system conditions. With multiclass classification, it is possible to determine the type of fault that is causing a system to malfunction. Additionally, the different meteorological conditions like incident solar irradiance, sun position and ambient temperature can be utilized to learn complex relations between these features. Another aspect which plays a role is system age, as different fault types are more likely to occur at specific moments of the system’s lifetime. We approached the tasks by implementing many models of ascending complexity in order to find the most accurate one and then tested it to determine its goodness.
The finalreport.pdf is the summary of all the work done.