Smart Grid Energy Management Using RNN-LSTM: A Deep Learning-Based Approach

Authors:
Publisher website
Google Scholar
Abstract
With the rapid increase in the energy demands from different sectors across the globe, there is lot of pressure on the power grid to maintain a balance between the demand and supply. In this context, smart grid (SG) may play a vital role as it provides the bidirectional energy flow between utilities and end users. Contrary to the traditional power grid, it has advanced switching and sensing devices (for example, sensors and actuators) for load balancing and peak shaving. In SG systems, various smart devices and electrical appliances which are placed in the smart buildings regularly generate data related to energy usage, occupancy patterns, or movements of the end users. By applying an efficient data pre-processing and data analytics technique, this data can be analyzed to extract important energy patterns which can be used in demand response management, load forecasting, and peak shaving. But, one of the main challenges in SG systems is to have an integrated approach to pre-process and analyze the data with minimum error rates and higher accuracy. To tackle the aforementioned challenges, an unified scheme based upon the deep learning and recurrent neural networks (RNN) is proposed in this paper. The data collected from smart homes is pre-processed and decomposed using high-order singular value decomposition (HOSVD) and then long short-term memory (LSTM) model is applied on it. As the data collected from SG is time series-based data so LSTM based regression model gives minimum root mean square (RMSE) and mean absolute percentage error (MAPE) values as compared to the other techniques reported in the literature. A case study of 112 smart homes with hourly basis data is considered for evaluation of the proposed scheme in which energy patterns are predicted with least RMSE and MAPE. The results obtained clearly show that the proposed scheme has superior performance in comparison to the other existing schemes

This publication has 13 references indexed in Scilit: