Short-Term Load Forecasting in Smart Grids Using Hybrid Deep Learning

Load forecasting in Smart Grids (SG) is a major module of current energy management systems, that play a vital role in optimizing resource allocation, improving grid stability, and assisting the combination of renewable energy sources (RES). It contains the predictive of electricity consumption forms over certain time intervals. Load Forecasting remains a stimulating task as load data has exhibited changing patterns because of factors such as weather change and shifts in energy usage behaviour. The beginning of advanced data analytics and machine learning (ML) approaches; particularly deep learning (DL) has mostly enhanced load forecasting accuracy. Deep neural networks (DNNs) namely Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) have achieved popularity for their capability to capture difficult temporal dependencies in load data. This study designs a Short-Load Forecasting scheme using a Hybrid Deep Learning and Beluga Whale Optimization (LFS-HDLBWO) approach. The major intention of the LFS-HDLBWO technique is to predict the load in the SG environment. To accomplish this, the LFS-HDLBWO technique initially uses a Z-score normalization approach for scaling the input dataset. Besides, the LFS-HDLBWO technique makes use of convolutional bidirectional long short-term memory with an autoencoder (CBLSTM-AE) model for load prediction purposes. Finally, the BWO algorithm could be used for optimal hyperparameter selection of the CBLSTM-AE algorithm, which helps to enhance the overall prediction results. A wide-ranging experimental analysis was made to illustrate the better predictive results of the LFS-HDLBWO method. The obtained value demonstrates the outstanding performance of the LFS-HDLBWO system over other existing DL algorithms with a minimum average error rate of 3.43 and 2.26 under FE and Dayton grid datasets, respectively.