Advances in Deep Learning Techniques for Short-term Energy Load Forecasting Applications: A Review

Today, the majority of the leading power companies place a significant emphasis on forecasting the electricity load in the balance of power and administration. Meanwhile, since electricity is an integral component of every person's contemporary life, energy load forecasting is necessary to afford the energy demand required. The expansion of the electrical infrastructure is a key factor in increasing sustainable economic growth, and the planning and control of the utility power system rely on accurate load forecasting. Due to uncertainty in energy utilization, forecasting is turning into a complex task, and it makes an impact on applications that include energy scheduling and management, price forecasting, etc. The statistical methods involving time series for regression analysis and machine learning techniques have been used in energy load forecasting extensively over the last few decades to precisely predict future energy demands. However, they have some drawbacks with limited model flexibility, generalization, and overfitting. Deep learning addresses the issues of handling unstructured and unlabeled data, automatic feature learning, non-linear model flexibility, the ability to handle high-dimensional data, and simultaneous computation using GPUs efficiently. This paper investigates factors influencing energy load forecasting, then discusses the most commonly used deep learning approaches in energy load forecasting, as well as evaluation metrics to evaluate the performance of the model, followed by bio-inspired algorithms to optimize the model, and other advanced technologies for energy load forecasting. This study discusses the research findings, challenges, and opportunities in energy load forecasting.