Title | Machine Learning to Predict Module Performance from Cell and Module Parameters |
Author(s) | Hannes Wagner-Mohnsen, Sven Wasmer, Bernhard Klöter, Pietro P. Altermatt, Marco Ernst |
Keywords | Machine Learning, Simulation, PERC, Module, SHAP |
Topic | Thin Films and New Concepts |
Subtopic | New Modeling and Characterisation Techniques |
Event | EU PVSEC 2023 |
Session | 2AO.3.1 |
Pages manuscript | 020074-001 - 020074-004 |
ISBN | 3-936338-88-4 |
DOI | 10.4229/EUPVSEC2023/2AO.3.1 |
Algorithms of artificial intelligence offer new ways to optimize solar modules in mass-production. We apply state-of-the-art algorithms combined with sophisticated device physics to identify the dominant parameters that affect solar module performance. We analyze solar cell parameters, three different binning strategies, and module parameters. All this data is fed into a machine learning model to reveal the underlying correlations between these parameters and the scattering of the modules power output. In our first results presented here, two module specific parameters (ribbon shading and Rs) lead to the strongest scattering of module power, while solar cell parameters have a smaller influence. This suggests that the cells were effectively binned, but we still observe an overall impact from the binning strategy on these impacting factors. The results of our framework depend on the architecture of the solar cells and modules, the ranges of inputs for relevant cell and module parameters fed into the model, and the binning strategy. Therefore, our approach is fully flexible and can be applied to different production lines and scenarios.