Transfer Learning for Short-Term-Load-Forecasting: Convolutional Neural Network - Long Short-Term Memory Forecasting Approach

Avram, Dan Andrei (2024) Transfer Learning for Short-Term-Load-Forecasting: Convolutional Neural Network - Long Short-Term Memory Forecasting Approach. Bachelor's Thesis, Computing Science.

Preview	Text AVRAM_Bachelor_s_Thesis_University_of_Groningen.pdf Download (437kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (131kB)

Abstract

In contemporary energy systems, accurate short-term energy load forecasting is crucial from both economical and environmental standpoints, and Deep Learning has demonstrated its efficacy as a powerful tool for this purpose. However, in order for Deep Learning models to achieve optimal performance, it is typically necessary for them to undergo extensive training, using a substantial amount of data. Nevertheless, this is not possible to achieve for newly constructed buildings due to their very limited amount of historical data. The utilization of Transfer Learning (TL) is able to counteract this issue by applying the data of the target domain onto the learning of a model which has already been trained on external, more comprehensive data. While previous studies have demonstrated the high performance of hybrid models such as the Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) for energy load forecasting, and the benefits of TL in enhancing the performance of conventional LSTM models, this research work addresses the existing literature gap of the analysis of TL being performed on a CNN-LSTM model. Focusing on energy load forecasting for schools in New York, the findings of this study provide further evidence for the efficacy of TL on an LSTM model. However, CNN-LSTM yields inconsistent outcomes under the effects of TL. Ultimately, this latter model achieves its optimal performance in our setting not by means of TL, but rather through improved data pre-processing

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Dustegor, D. and Tello Guerrero, M.A.
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	29 Feb 2024 12:38
Last Modified:	29 Feb 2024 12:38
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/31993

Actions (login required)

View Item