Energy Consumption Forecasting Based on Long Short-term Memory Neural Network with Realistic Smart Meter Data

Abstract

Energy consumption forecasting has been more and more important in saving energy and mitigating the hazard-ous environmental impact, which plays a vital role in facilitating decision-making and planning in smart grids. This paper advocates the long short-term memory (LSTM) neural network for time series energy consumption forecasting based on realistic smart meter data. The proposed approach is validated using the open-source, public smart meters data in the United Kingdom under extensive case studies. The results demonstrate that the proposed approach based on LSTM achieves high accuracy in predicting energy consumption due to the capability of modeling long temporal sequences. The impact of different weather conditions on forecasting performance is also analyzed.