This python script is a CNN classification machine learning project. The data contains 16 days of data of a grid-tie photovoltaic plant's operation with both faulty and normal operation. The paper achieved 99.88% accuracy with overfitting likely, and 98.31% accuracy with overfitting preventing methods including L2 regularizationa and neuron dropout, improving on the previous record for the dataset,97.64% accuracy
The faults include:
Short-Circuit (Short Circuit between 2 modules of a String) Degradation (There is a resistance between 2 modules of a String) Open Circuit (One String disconnected from the power inverter) Shadowing (Shadow in one or more modules).
The paper and results i am trying to reproduce can be found here: https://www.mdpi.com/1424-8220/22/21/8515#B29-sensors-22-08515). This is the research that achieved 97.64% accuracy.
The data source can be found: https://github.com/clayton-h-costa/pv_fault_dataset
The data source origin is explained by this research group: https://www.mdpi.com/1424-8220/20/17/4688)