Please use this identifier to cite or link to this item: https://bura.brunel.ac.uk/handle/2438/34214
Title: Hybrid Deep Learning Framework for Long-Term Forecasting of PV Generation
Authors: Babic, Ljubomir
Aunedi, Marko
Keywords: deep learning;photovoltaics;forecasting;convolutional neural networks;long-short-term memory
Issue Date: 2-Sep-2025
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Citation: Babic, L. and Aunedi, M. (2025) 'Hybrid Deep Learning Framework for Long-Term Forecasting of PV Generation', 60th International Universities Power Engineering Conference (UPEC), London, UK, 2–5 September. pp. 1–6. doi: 10.1109/upec65436.2025.11279840.
Abstract: Accurate forecasting of photovoltaic (PV) generation is critical for renewable energy integration and maintaining system reliability. This paper demonstrates the application of a hybrid deep learning framework based on Convolutional Neural Network (CNN) and Long-Short-Term Memory (LSTM) for forecasting PV generation in the UK. The proposed model uses a 45-year dataset (1980-2024) of simulated hourly PV output values, allowing it to capture local temporal patterns through convolutional layers and long-term dependencies via bidirectional LSTM layers. Cyclical temporal features (hour of the day, day of the week, and month of the year) are feature engineered to encode periodicity. Experimental results show that the predictions made by a hybrid CNN+LSTM are very close to the actual PV generation values, with a Mean Absolute Error (MAE) of 0.0061, Root Mean Squared Error (RMSE) of 0.0121, and Mean Absolute Percentage Error (MAPE) of 4.86% on the test set.
URI: https://bura.brunel.ac.uk/handle/2438/34214
DOI: https://doi.org/10.1109/upec65436.2025.11279840
ISBN: 9798331565206
9798331565213
Appears in Collections:Department of Electronic and Electrical Engineering Research Papers

Files in This Item:
File Description SizeFormat 
FullText.pdfCopyright ‘For the purpose of open access, the author has applied a ‘Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising.’1.75 MBAdobe PDFView/Open


This item is licensed under a Creative Commons License Creative Commons