Forecasting Day-Ahead Electricity Demand in Australia Using a CNN-LSTM Model with an Attention Mechanism

Accurate energy demand forecasting is vital for optimizing resource allocation and energy efficiency. Despite advancements in various prediction models, existing approaches often struggle to capture the complex, nonlinear relationships between temperature variations and electricity consumption. To address this issue, this paper introduces a novel hybrid deep learning model that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks with an attention mechanism designed to forecast day-ahead electricity demand in Australia. This research aims to enhance the accuracy of electricity demand predictions by effectively modeling the impact of heating degree days (HDDs) and cooling degree days (CDDs) on energy usage. The HDDs and CDDs capture extreme weather conditions. They are critical for understanding spikes in energy consumption for heating and cooling needs. The proposed model leverages the strengths of CNNs in extracting spatial features in HDDs and CDDs, LSTMs in capturing temporal dependencies, and the attention mechanism in focusing on the most relevant aspects of the data. This study compares the CNN-LSTM-Attention model with traditional methods, including Deep Neural Networks, and demonstrates superior performance. The results show a significant reduction in both Mean Absolute Error and Mean Absolute Percentage Error, confirming the model’s effectiveness. The primary contribution of this paper lies in the novel integration of CDD and HDD data within the CNN-LSTM framework, which has not been extensively explored in prior studies. This approach offers a robust solution for energy management, particularly in climates with significant temperature fluctuations.