MPI Cluster with Raspberry Pi for Distributed Training of LightGBM Models and Shading Mapping in Photovoltaic Plants

• Vagner Souza IFCE
• Wendell Rodrigues IFCE
• Rejane Sá IFCE

DOI: https://doi.org/10.5753/eradco.2024.4373

Abstract

This paper proposes a distributed system using a Raspberry Pi cluster, each equipped with irradiance sensors and connected to inverters in a photovoltaic plant. The system processes data locally using MPI to train LightGBM models for real-time energy prediction and anomaly detection. Additionally, a virtual shading map is generated from irradiance data. This approach minimizes cloud processing, reducing latency and costs, while ensuring quick responses to environmental changes. The system is integrated with a CI/CD pipeline for continuous MLOps operations. Preliminary results demonstrate the solution’s scalability and potential to enhance photovoltaic plant reliability.

References

de Souza, V. H., Rodrigues, W., and Lima, M. (2023). Smartgrid improvements on photovoltaic systems by analyzing energy and weather measures. In Proceedings of the Advance 2023.

Ignacio, A. and Martins, V. (2002). Mpi: uma ferramenta para implementação paralela. Pesquisa Operacional, 22.

Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., and Liu, T.-Y. (2017). Lightgbm: A highly efficient gradient boosting decision tree. In Neural Information Processing Systems.

Oda, T., Yamada, M., Obukata, R., Barolli, L., Woungang, I., and Takizawa, M. (2016). Experimental results of a raspberry pi based wireless mesh network testbed considering tcp and los scenario. In 2016 10th International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS), pages 175–179.

Yu, F., Zhang, P., Xiao, W., and Choudhury, P. (2011). Communication systems for grid integration of renewable energy resources. Network, IEEE, 25:22 – 29.