Bidirectional LSTM Network-Based Short-Term Load Forecasting Method in Smart Grids

Abstract

Load forecasting, including classic time series analysis and more contemporary machine learning techniques, has emerged as one of the most prominent research domains. The primary emphasis of research in this field lies in predicting aggregated power usage. However, the significance of demand side energy management, encompassing individual load forecasts, is increasingly gaining prominence. This work proposes load forecasting models that rely on deep neural networks (DNNs). These models are applied to a demand-side load database for analysis. The forecasting accuracy of DNN-based load forecasting models is assessed by comparing them with Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Bidirectional Long-Short-Term Memory (B-LSTM) models. The B-LSTM is a new recurrent artificial neural network recommended as the forecasting unit due to its ability to process information from both the past and present hidden layer using memory loops. Performance of the algorithm is check based on the mean absolute error, mean absolute percentage error, and root mean square error.