Combining auto-regression with exogenous variables in sequence-to-sequence recurrent neural networks for short-term load forecasting

In this paper we propose a sequence-to-sequence machine learning architecture for time-series forecasting based on recurrent neural networks. This architecture can be used as a general purpose forecasting method and is evaluated for the application of short-term electric load forecasting in this paper. The proposed sequence-to-sequence architecture ¹ combines elements of auto-regressive forecasting techniques with multivariate regression by including exogenous variables for each forecasted time step as well as previous values when inferring forecasts. We assess the proposed architecture on a load data set provided by the Global Energy Forecasting Competition. The conclusion is that it outperforms other machine learning forecasting techniques as well as time-series analysis methods. ¹ The implementation is available at: https://github.com/HenWil13/ieeeINDIN18