Electric load forecasting using deep neural networks

Ahmed, Zain (2025) Electric load forecasting using deep neural networks. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (3MB)

Abstract

Short-Term Load Forecasting (STLF) is a critical and complex task that plays a vital role in the efficient management of electricity generation, transmission, and distribution. Recent research has made strides in this field through the application of advanced deep learning techniques to enhance the accuracy and reliability of load predictions. The first study introduces a novel deep neural network tailored for STLF at Memorial University of Newfoundland (MUN). This model integrates electric load data with meteorological information and features a 1D Convolutional Neural Network followed by an Encoder-Decoder Network with an attention mechanism, showing superior performance compared to traditional Gated Recurrent Unit (GRU) and Long Short Term Memory (LSTM) models. The study also focuses on optimizing the input horizon using the algorithm. The second study focuses on Multi-Energy Systems (MES) and presents a Multi-Task Learning-based approach for load forecasting. It features a cutting-edge deep learning architecture designed to forecast multiple loads simultaneously. Applied to the University of Austin Tempe Campus, this approach employs a Deep Temporal Convolutional Neural Network (D-TCNet) to effectively capture spatial and temporal correlations in the data, resulting in improved forecasting accuracy across different energy types and seasons. The third study compares various Recurrent Neural Network (RNN)-based time-series forecasting algorithms, including LSTM, GRU, Bi-directional GRU, and Bi-directional LSTM, on electric load data from MUN. The Bi-directional GRU model emerged as the top performer, achieving the highest R2 score and the lowest Mean Squared Error (MSE) and Mean Absolute Error (MAE) for day-ahead predictions. Collectively, these studies demonstrate the power of deep learning in enhancing the precision and effectiveness of short-term load forecasting, offering promising avenues for optimizing energy system operations.

Actions (login required)

View Item