Multistep Time Series Forecasting of Energy Consumption Based on Stacked Deep LSTM Network Architecture

The analysis of energy time series involves examining historical energy data and potentially considering external factors to make predictions. Various tasks fall within the broader field of energy time-series analysis and forecasting, with popular ones including forecasting electric load demand, predicting personalized energy consumption, and forecasting renewable energy generation. Given the impressive capabilities of Deep Learning (DL) models in numerous vision-related tasks, they have been applied to time-series forecasting. This paper explores whether and in what ways recently developed DL-based architectures, like Long Short-Term Memory (LSTM), can enhance performance in energy time-series forecasting tasks. To mitigate overfitting, the proposed model incorporates dropout layers to enhance its generalization capability.