PV Power Forecasting using Distributed Machine Learning for Smart Energy Grid

Muhammad Ammar ZAFAR

Diplomová práce

(An extension to Federated Learning)

Abstract:

This thesis investigates the application of federated learning and tree-based models, such as LightGBM and Catboost, in Photovoltaic (PV) power forecasting. Addressing challenges in accuracy, uncertainty, and scalability, the study designs a robust federated learning architecture tailored for tree-based forecasting models. The novel aggregation strategy efficiently combines updates from multiple nodes, enhancing forecast accuracy.

Abstract:

This thesis investigates the application of federated learning and tree-based models, such as LightGBM and Catboost, in Photovoltaic (PV) power forecasting. Addressing challenges in accuracy, uncertainty, and scalability, the study designs a robust federated learning architecture tailored for tree-based forecasting models. The novel aggregation strategy efficiently combines updates from multiple nodes, enhancing forecast accuracy.

Keywords

Machine Learning (ML), Photovoltaic(PV), A Friendly Federated Learning Framework (FLWR), Mean Prediction Interval Range (MPIR), and Mean Quantile Loss (MQL), Random Forests (RF), LightGBM, CatBoost

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 8. 2. 2024

Obhajoba závěrečné práce

• Vedoucí: prof. Dr. Andreas Kassler

JIHOČESKÁ UNIVERZITA V ČESKÝCH BUDĚJOVICÍCH

Přírodovědecká fakulta

Magisterský studijní program / obor:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science