Robust short-term load forecasting using a new modeling approach

In this paper, a new modeling approach for short-term load forecasting is proposed. The electrical consumption time series in France is represented as a multivariate combination of Seasonal Autoregressive Integrated Moving Average (SARIMA) models with output additive Gaussian white noises. A fast-executing robust method to estimate these models without being influenced by the adverse effects of special days, known as outliers in statistics, is also illustrated. Finally, a comparative analysis shows the effectiveness of the proposed procedure in forecasting normal days of the French national electrical consumption.