Advancements in Deep Learning for Electrical Consumption Forecasting : A Review of 2019 - 2023 Studies for LSTM

Abstract

Deep learning (DL) is a machine learning methodology that leverages learning algorithms based on artificial neural networks(ANN) to enable computers to perceive intricate structures and patterns in data and effectively utilize them (Patchaiammal P.and Sundar, 2022). In essence, DL constitutes a set of algorithms capable of addressing complex problems by emulating the structural attributes of the human brain (Forootan et al., 2022). DL offers numerous advantages, such as the extraction of high-level characteristics from data, the ability to process data irrespective of its labeled or unlabeled nature, and the adaptability to be trained for diverse objectives. DL algorithms commonly demonstrate remarkable predictive capabilities in time series data. DL comprises three primary types of neural network architectures, namely ANN, Convolutional Neural Networks(CNN), and RNN (Patchaiammal P.and Sundar, 2022). Noteworthy deep learning algorithms in the literature include LSTM, RNN, GRU, CNN and Feed forward Neural Networks. In the following sections of the study, information about RNN, LSTM, GRU is given. Afterwards, the literature study is given. Before the conclusion part, general information about performance methods is given.