Capturing complex electricity load patterns: A hybrid deep learning approach with proposed external-convolution attention

Abstract

Short-term electricity load forecasting is a critical factor in optimizing power systems, minimizing operating costs, and securing reliable energy resources. There are various approaches for short-term electricity load forecasting, but handling complex dependencies and sudden changes in load data remains challenging. This study introduces a hybrid deep learning model to improve load forecasting accuracy. The model combines the strengths of various deep learning architectures such as Convolutional Neural Network, Temporal Convolutional Network, and Bidirectional Long Short-Term Memory with a proposed attention mechanism. This approach helps to extract temporal relations and learn long-term patterns. Furthermore, the proposed External-Convolution Attention technique effectively captures global and temporal patterns within the input sequences. Three data sets are used to conduct experiments and validate the proposed model. The proposed model is compared against several machine learning and deep learning models across five evaluation metrics. The findings show the strength of the proposed model by outperforming other models. Our load forecasting method achieves improvements ranging from 2% to 21% across different evaluation metrics. The study also evaluates the effects of datasets, features, and prediction horizons. The presented hybrid deep learning model and a novel attention mechanism improve load forecasting accuracy, contributing to the advancement of artificial intelligence in energy optimization techniques. Our model demonstrates superior performance through extensive experimentation and diverse scenarios by identifying complex load patterns and adjusting to various datasets. This indicates its practical applicability in engineering for optimizing power systems and minimizing operational costs.